Claude for HR: Building AI Workflows for Performance Reviews (Webinar)

https://www.youtube.com/watch?v=cctFWGecvhg

[00:04] Everyone, thanks for joining the Claude for HR webinar.
[00:09] for HR webinar.
[00:10] We're going to give everyone a few minutes um and then we'll get going.
[00:19] And while we wait, would love to hear in the chat where people are calling in from today.
[00:43] All right, we got people from Denver, Florida, Poland, Vegas, Canada, Berlin.
[00:54] Nigeria, we're global.
[01:02] Awesome.
[01:05] We're going to get going in a minute or so here.
[01:10] Uh we're going to send out our first poll.
[01:13] We want this webinar to be as interactive as possible.
[01:17] So you'll see a poll pop up on the screen in a second here.
[01:21] Uh if anyone's new to StreamYard, the way to respond to the poll is to type one, two, or three in the chat.
[01:28] Uh the first question we have is what level of AI fluency uh do you have?
[01:34] So people just want to put one, two, or three in the chat, we'll see those results populate.
[01:46] Nice.
[01:48] All right, we have a few masters.
[01:51] Most people are claiming to be at the pro level.
[02:05] Any more masters?
[02:20] We have Heather as a master.
[02:24] We have Heather as a master.
[02:24] Anyone else?
[02:27] Anyone else?
[02:27] Mia.
[02:33] All right.
[02:36] We'll get going in one more minute.
[03:03] All right, let's get going.
[03:03] Uh so this webinar will be Claude for HR.
[03:05] We're going to be chatting for about an hour.
[03:10] going to be chatting for about an hour.
[03:10] Uh a lot of that is going to be Q&A.
[03:12] So if you have questions while I'm going, just put those right into the chat on the right-hand side.
[03:16] I will answer a bunch of questions as we go, as we wait for AI to think cuz some of our demos take a little bit of time.
[03:22] Um and then we will have a dedicated Q&A question section at the end.
[03:29] Today, what are we going to cover?
[03:31] Um overall, I want this to be as demo um Hm, few people saying they can't hear.
[03:37] Let's see.
[03:43] Okay. Um hopefully people can hear.
[03:46] Um so we're going to cover three use cases.
[03:49] One is building a chatbot for policy Q&A, one is setting up a performance review cycle, and then one is doing data analysis uh on a bunch of survey results.
[03:58] Uh I want this to be as demo heavy as possible.
[03:59] We have very few slides and most of the time I'm going to be demoing live um right within Claude.
[04:05] Um so for each of these different demos,
[04:10] so for each of these different demos, we're going to go through a novice and we're going to go through a novice and then a pro and then a master level.
[04:14] then a pro and then a master level.
[04:15] Uh we'll start with simple stuff and then we're going to get pretty complicated pretty quickly.
[04:17] we're going to get pretty complicated pretty quickly.
[04:19] pretty quickly.
[04:21] Uh for people who are lagging, um I'm not sure the best fix there.
[04:23] It seems like a lot of people have it working well.
[04:25] well.
[04:26] Um but hopefully uh people can hear it.
[04:27] but hopefully uh people can hear it.
[04:29] If you can't hear it, we will send out this recording afterwards.
[04:30] you can't hear it, we will send out this recording afterwards.
[04:32] recording afterwards.
[04:37] Um apologies for any audio issues.
[04:40] By way of introduction, uh my name is Max Shaw.
[04:42] I'm the CEO and co-founder of Windmill and I'm joined by Nicole.
[04:44] Nicole, if you want to introduce yourself.
[04:45] yourself.
[04:47] Hi everyone, I'm Nicole.
[04:48] So great to see so many people here.
[04:51] I know that a lot of people probably found this webinar from the Claude for HR cheat sheet we posted earlier this week.
[04:52] of people probably found this webinar from the Claude for HR cheat sheet we posted earlier this week.
[04:55] Hope that you found a lot of value in that and hope that everyone finds value in what we're presenting today.
[04:57] posted earlier this week.
[04:58] Hope that you found a lot of value in that and hope that everyone finds value in what we're presenting today.
[05:00] that everyone finds value in what we're presenting today.
[05:03] Awesome.
[05:04] Awesome.
[05:06] Um so I'm the founder of Windmill.
[05:07] I'm the founder of Windmill.
[05:10] Uh previously, I ran product at a New York City startup um and I found all of the
[05:13] City startup um and I found all of the HR tools uh specifically around HR tools uh specifically around performance really lacking.
[05:17] performance really lacking um and I saw the opportunity with AI and um and I saw the opportunity with AI and that's why we started Windmill uh to build the first AI native solution for performance management.
[05:23] build the first AI native solution for performance management.
[05:25] Um our first product was uh performance reviews.
[05:28] We launched this uh just a few months ago and the response has been incredible.
[05:31] Uh it allows you to generate performance reviews uh using AI in much faster time um and the results have been phenomenal.
[05:37] reviews uh using AI in much faster time um and the results have been phenomenal.
[05:39] Um and so if anyone's looking for AI performance reviews or one-on-ones or a bunch of different tools, Windmill is your spot.
[05:43] performance reviews or one-on-ones or a bunch of different tools, Windmill is your spot.
[05:45] Uh that is the end of my Windmill commercial.
[05:47] Uh for now, we are going to be talking Claude and how can you most effectively use Claude to uh accelerate your daily workflows.
[05:50] going to be talking Claude and how can you most effectively use Claude to uh accelerate your daily workflows.
[05:54] accelerate your daily workflows.
[05:56] Again, we're going to start simple but then we're going to ramp up in complexity pretty quickly.
[05:58] then we're going to ramp up in complexity pretty quickly.
[05:59] complexity pretty quickly.
[06:01] So let's jump into some live demos.
[06:09] All right, I'm going to share my screen.
[06:15] Just.
[06:19] All right, uh everyone should be seeing a drawing of this uh right now.
[06:24] Um and to start, I wanted to ask what are people using daily for AI?
[06:26] I've we five different options uh ChatGPT, Codex, Claude Chat, Claude Code, or Claude Co-work.
[06:37] Um there's a lot of different options here.
[06:38] If you use something else, just comment it in there.
[06:40] But I'm curious what people are using today.
[06:41] We're primarily going to focus on Claude today.
[06:45] Uh it seems like the majority of people are using ChatGPT.
[06:47] Um we have a few people on Claude Code and then there's some people on chat and co-work.
[06:55] Awesome.
[06:57] Well, I'm going to talk through we're going to all the demos are going to be in Claude today.
[06:57] I'm going to talk through the differences between chat and co-work uh cuz those are really important.
[07:04] Now, if you look if you think about these different tools, you can kind of plot them on a graph of I would say complexity going to the right and ability uh on top of the scale.
[07:14] So ChatGPT and Claude Chat are pretty much the
[07:18] and Claude Chat are pretty much the same.
[07:20] They're both very smart um and they're very simple to use um but they don't have that much ability compared to some of the other options that are out there.
[07:26] Going up the scale, you have Claude Co-work which is a great option and that's where I'm going to be doing the majority of my demos today.
[07:34] And then a level above is Claude Code.
[07:37] So if you're using ChatGPT or Claude Chat today, I would definitely recommend trying out Claude Co-work.
[07:44] It's a really good option that gives you a lot more power.
[07:46] Now, one of the most important things between these different options is whether they work on the web or they are local.
[07:53] So ChatGPT and Claude Chat are on the web and what that means is that they don't have access to your local computer or your local file system.
[08:02] Claude Co-work, Claude Code, and also Codex is over here.
[08:06] It runs locally on your computer and so it can access the files you have on your computer.
[08:13] And so for a bunch of the demos, you'll see this in action how it can actually access my local file system.
[08:19] access my local file system.
[08:21] Now, there's a million different terms out there.
[08:23] Um AI is moving at a pretty insane rate uh and you'll hear things like skills
[08:27] uh and you'll hear things like skills and commands and MCPs and integrations
[08:29] and commands and MCPs and integrations and apps
[08:31] and apps um and there's all these different terms out there.
[08:34] And to simplify it down, I think of it as two categories.
[08:36] On the left here, you have different ways of injecting information into the prompt.
[08:44] If you've used AI a lot, you know that the information in the prompt is absolutely critical.
[08:48] This is often called the context.
[08:50] You can type all this stuff manually or you can use these different ways of injecting context at the right time.
[08:55] And I'll show you a bunch of examples of how that works.
[08:58] On the right-hand side, you have different ways of connecting to other integrations.
[09:04] In order to get the most out of AI, you want to be able to connect your apps, right?
[09:07] This allows you not just to turn this from a chatbot, but this actually turns it into an agent that can get work done for you.
[09:13] They come in different terms, connectors, MCPs, integrations, apps.
[09:16] They're all pretty much the same thing.
[09:19] Um
[09:20] Um and I will show you exactly how they and I will show you exactly how they work.
[09:23] work. Gemini, a bunch of people are asking Gemini, a bunch of people are asking about Gemini.
[09:26] Gemini is based on the web. Um they do have a local CLI as well, but no one really uses it.
[09:31] Um and I'll just talk more about some of these different pieces.
[09:34] So with that, let's jump into a demo.
[09:37] So as I mentioned, we're going to show three use cases of different ways you can use Claude.
[09:42] These are anchored around HR, but most of these use cases and tips are applicable to really anyone uh for any kind of workflow.
[09:50] So the first thing we're going to do is a policy Q&A.
[09:51] Uh I'm sure most people in HR, one of the most annoying things you get is the same questions over and over again.
[09:58] And so I'm going to build a little chatbot here that can show you how to answer questions.
[10:03] So let's start with something super simple. What's the vacation policy?
[10:08] Everyone's always asking about vacation.
[10:10] And so I'm in Claude chat right now. You see at the top there's chat, co-work and code.
[10:14] I'm in Claude chat and I asked what's the vacation policy and it can't give me an answer.
[10:20] and it can't give me an answer.
[10:22] It says Windmill performance review tool.
[10:23] cool but it doesn't know what my vacation policy is and that's because I haven't added anything to the prompt that shows you what the vacation policy is.
[10:31] So unsurprisingly can't answer.
[10:33] Now I have a PDF right here with company policies and I will put that in the chat and I'll say here are the policies.
[10:47] And now it's able to read that PDF and it's able to understand the PTO allotment is 20 days, right?
[10:54] Super simple but you can see how important it is to have the right information in the prompt.
[10:58] Now let's jump over to co-work which is tab two and one of the cool things with co-work is I can tell it to work in a project.
[11:07] So I'm going to have it work in the Claude HR webinar which is a folder and I'm going to ask the same question.
[11:17] and we ask that same question. This is the folder I gave it access to.
[11:21] the folder I gave it access to.
[11:23] You can see the PDF is in there and so in co-work it can look through that folder.
[11:27] It's going to see if it finds any PDFs.
[11:29] It found the Windmill company's policies PDF.
[11:32] I didn't even have to tell it about it and it answered the same question.
[11:35] Now this is the same answer but conceptually it's pretty different.
[11:39] Co-work is working on my computer.
[11:41] It automatically found this PDF and was able to use it to answer a question.
[11:47] So when you start using co-work it's really important what is the folder you're giving it access to.
[11:52] Now this is nice, pretty cool and this could work in PDF or any type of format but the issue here is probably pretty obvious.
[11:59] What happens when you change that policy?
[12:01] Immediately that PDF is out of date and having to resend out that PDF to every single person every time there's a change and expecting every employee to do that just isn't realistic.
[12:11] So let's take this a step further.
[12:14] What if we use Notion as our database?
[12:18] So you could use anything.
[12:19] I'm going to use Notion here.
[12:20] I'm going to go to connectors and if you click manage
[12:22] connectors and if you click manage connectors right here or if you go to connectors right here or if you go to customize over here.
[12:26] customize over here and you go to connectors here are the and you go to connectors here are the connectors that I've hooked up and I can go down to Notion.
[12:31] go down to Notion and then we can connect our Notion.
[12:34] and then we can connect our Notion.
[12:36] So I'm just going to do this and we are going to this is just a fake Notion account.
[12:40] We're going to connect it live right here.
[12:44] Okay, so now Notion is integrated.
[12:48] Now on my computer I have the PDF and I also have Notion connected.
[12:53] So how does co-work know what to use?
[12:56] This is where skills come in and so I'm going to upload a skill here.
[12:58] So I have this upload a skill here.
[13:00] So I have this skill.
[13:01] I'll show you what it looks like.
[13:03] I'm going to go plus.
[13:05] I'm going to upload a skill and I'm just going to grab this zip file.
[13:10] and I'm just going to grab this zip file and now I have this policy chatbot Notion.
[13:13] I'll show you in a second what that skill actually does.
[13:15] So now we go back and let's say the same thing.
[13:17] and let's say the same thing.
[13:20] What is the vacation
[13:26] and we're going to run that.
[13:28] and we're going to run that and on the right hand side you'll see the context start to fill out.
[13:35] Let's give it a second here.
[13:41] And while we're waiting on that I'm seeing some questions come in about security.
[13:44] We will talk about security, sharing skills, a few of those things that were asked throughout the rest of the presentation.
[13:52] So here you see it invoked the skill this policy chatbot Notion.
[13:55] If I click here I can see what's in the skill.
[13:59] It's super simple actually.
[14:01] It just basically says if someone has a question about HR you should use the Notion MCP to find the answer.
[14:08] So that's all this skill is doing.
[14:10] You can use Claude to create your own skills and I think about skills as reusable workflows.
[14:17] If you want to go I can show you how you can create them.
[14:18] So if I go to skills I can create one manually where you just go write skill instructions or I could go in and actually upload a
[14:26] or I could go in and actually upload a skill.
[14:28] So here at Windmill the way we skill.
[14:30] So here at Windmill the way we generally share skills is just by generally share skills is just by sending around those zip files which I
[14:32] sending around those zip files which I showed before and then people can add those.
[14:33] those. There are more advanced ways to do it as well that I'll talk about in a
[14:36] do it as well that I'll talk about in a little bit but that's the simplest way to get started.
[14:37] little bit but that's the simplest way to get started.
[14:39] to get started.
[14:40] So if we go back to this you can see it got the same answer of 20 PTO days per year.
[14:42] got the same answer of 20 PTO days per year.
[14:45] Now the cool thing is that because this is in Notion let's go over to Notion.
[14:47] is that because this is in Notion let's go over to Notion.
[14:50] let's go over to Notion.
[14:53] This is where it's pulling from and if I update the annual PTO to be 25 days and then we go back to Claude
[14:55] update the annual PTO to be 25 days and then we go back to Claude and let's start a new task.
[14:59] and then we go back to Claude and let's start a new task.
[15:01] and let's start a new task.
[15:04] Task is the same as chat. They just call it a new task
[15:06] They just call it a new task in co-work and I ask again what's vacation policy at Windmill.
[15:08] it a new task in co-work and I ask again what's vacation policy at Windmill.
[15:10] what's vacation policy at Windmill.
[15:16] You'll see it's going to rerun the Notion search.
[15:17] You'll see it's going to rerun the Notion search.
[15:19] Notion search. If you ever want to see what co-work is doing you can drill into these things.
[15:20] what co-work is doing you can drill into these things.
[15:23] these things. This can be helpful to understanding what the agent's doing behind the scenes.
[15:24] understanding what the agent's doing behind the scenes.
[15:25] behind the scenes. So in this case it's running a query for vacation policy. You
[15:28] running a query for vacation policy.
[15:28] You can see them all right here.
[15:29] It fetched the answer and now it says 25 days per year.
[15:31] So you can distribute this skill once and then you just keep the Notion up to date and anytime someone asks a question it will get the right answer.
[15:33] Notion is just a data store here.
[15:36] You could use anything else as well but it's a pretty simple way to show off this concept.
[15:40] And Max we're getting some questions about skills.
[15:42] Do you mind sharing what the difference is between a skill and a prompt, when people should create skills?
[15:45] Yes skills are I would think about skills as anytime you want to reuse a prompt over and over again.
[15:47] So if you find yourself and this has happened to me where you like copy and paste snippets or you're constantly rewriting the same text every time skills are a really good way of reusing that.
[15:49] I can type slash and I can see the skills.
[15:51] Here are the skills I have right now.
[15:52] So it's just a shortcut.
[15:54] All it is is a shortcut.
[15:55] It's literally just taking the text of the skill and pushing it right into the prompt and so it just saves you from having to reprompt it all the stuff that you already know about.
[15:57] So anytime
[16:28] that you already know about.
[16:30] So anytime you're doing something reusable you want you're doing something reusable you want to do this a lot, save it as a skill and
[16:32] to do this a lot, save it as a skill and you can ask Claude just to create a
[16:33] you can ask Claude just to create a skill
[16:35] skill and it can do it from there.
[16:37] and it can do it from there.
[16:38] So that's the first use case, really simple example of a vacation policy.
[16:41] simple example of a vacation policy.
[16:43] So the next use case I want to talk about is performance reviews.
[16:46] So say I'm in HR and and I'm in charge of setting up the performance review cycle.
[16:48] in HR and and I'm in charge of setting up the performance review cycle.
[16:49] I'm going to use WhisperFlow here to dictate my prompt because it's a lot
[16:52] dictate my prompt because it's a lot faster
[16:53] faster and I'm going to say I'm in HR and I'm
[16:56] and I'm going to say I'm in HR and I'm in charge of setting up our performance review cycle.
[16:58] in charge of setting up our performance review cycle.
[17:00] Write me three questions for a self review and three questions
[17:02] for a self review and three questions for a manager review.
[17:04] You can see I'm using Sonnet 4.6 here.
[17:08] You can see I'm using Sonnet 4.6 here.
[17:11] There's a lot of good models here.
[17:12] I would recommend using Sonnet.
[17:14] There's a reason it's the default.
[17:16] It's really efficient.
[17:18] Opus would be a more powerful one and then Haiku is faster
[17:20] powerful one and then Haiku is faster but Sonnet is generally a really good
[17:21] but Sonnet is generally a really good default.
[17:22] default.
[17:24] And yes I said WhisperFlow.
[17:26] That's how I'm doing the voice to text that I just did there really quickly.
[17:28] did there really quickly.
[17:29] And so here we get an answer.
[17:32] So six questions, three self review and three manager review.
[17:34] Now these are good questions but they're super generic.
[17:37] They're not anything about my business.
[17:40] They're actually not that good of questions from what I'm trying to solve for.
[17:43] So one of the most helpful prompting tips I have is that when you do a question like this.
[17:46] I'm going to say ask me four clarifying questions before.
[17:54] So what you do is you actually have Claude interview you before it writes the answer.
[18:01] You could have just included all this information in the prompt but I find asking Claude to interview you is a lot more effective way to making sure you're giving all the most helpful information about what this is and what it's about, right?
[18:14] So in this case what type of company you are?
[18:15] Let's say we're tech and software.
[18:19] I want it to be growth focused.
[18:21] It's going to be all levels.
[18:26] How are responses be used?
[18:26] Development
[18:29] How are responses be used?
[18:29] Development planning.
[18:30] planning.
[18:30] And you can go through this.
[18:32] And you can go through this.
[18:32] I can also ask additional information.
[18:35] I can also ask additional information.
[18:35] Can you add some questions about AI
[18:39] Can you add some questions about AI excellence?
[18:40] excellence?
[18:40] I want everyone to get really good
[18:44] I want everyone to get really good at AI, right?
[18:46] So now I'm got back going back and forth.
[18:47] back and forth.
[18:47] Don't worry about typos when you're talking to AI.
[18:51] It will figure it out.
[18:51] Don't waste time with that.
[18:52] And now you can see how it did bunch more questions, right?
[18:57] Now I have five of each.
[18:59] Okay.
[18:59] And so now I have my questions set up, right?
[19:03] So I have my self review.
[19:03] I have my manager review.
[19:05] You could keep going back and forth and now the next thing is I want to set up this cycle.
[19:10] Now let's say I'm using Windmill for performance reviews.
[19:10] A big mistake I see people make is that they start copying and pasting all this stuff into the tool.
[19:20] This is where Claude and really using it as an agent can be extremely powerful.
[19:22] So instead of that I'm going to say okay I want to create this cycle for real.
[19:29] real.
[19:31] Can you use the Windmill connector to create the cycle in Windmill?
[19:36] And we'll try this.
[19:38] So this is using the Windmill MCP which we're going to be releasing over the next few weeks.
[19:43] And if I go to connectors you can see Windmill's hooked up and let's see as this goes.
[19:49] What this is supposed to do is look into Windmill and then take all those questions and turn it into a cycle.
[19:54] So here you can see it working.
[19:57] While we're waiting, are there any questions, Nicole?
[20:00] Yes, we've gotten a couple of good ones here.
[20:04] Uh a lot about security and privacy.
[20:06] I don't know if that's something you want to touch on now or wait until later.
[20:10] Yeah.
[20:12] Um so security and privacy obviously critical and there's a few different ways to look at it.
[20:15] Um the first is you want to use a tool that has been vetted by your IT department and your legal team so that you feel comfortable sharing whatever information you want.
[20:23] If you're really limited about the information you can share, it the AI becomes a lot less effective.
[20:28] Now when
[20:30] becomes a lot less effective. Now when you're choosing tools, there are things
[20:32] you're choosing tools, there are things like zero-day retention, um which means
[20:34] like zero-day retention, um which means none of the logs will be saved by the
[20:36] none of the logs will be saved by the model provider, uh none of them will be
[20:38] model provider, uh none of them will be used for training. Um and so of course
[20:40] used for training. Um and so of course you want to be using approved tools, but
[20:43] you want to be using approved tools, but I would encourage you to work with your
[20:45] I would encourage you to work with your IT teams to make sure that you use a
[20:46] IT teams to make sure that you use a tool that you feel comfortable sharing
[20:49] tool that you feel comfortable sharing anything you want so that you can really
[20:50] anything you want so that you can really get the full power of AI. And all of
[20:52] get the full power of AI. And all of these models have the ability to have
[20:54] these models have the ability to have zero-day retention, which means their
[20:55] zero-day retention, which means their logs are never saved, they're never used
[20:57] logs are never saved, they're never used for training or anything like that.
[20:59] for training or anything like that. All the top companies that are using AI
[21:01] All the top companies that are using AI super well are doing it in a pretty
[21:03] super well are doing it in a pretty free-flowing way where you can feel
[21:04] free-flowing way where you can feel comfortable sharing anything you want.
[21:06] comfortable sharing anything you want. So this cycle has now been created in
[21:08] So this cycle has now been created in Windmill. So I actually can just open up
[21:10] Windmill. So I actually can just open up this link.
[21:11] this link. And if we go into admin, so I'm inside
[21:13] And if we go into admin, so I'm inside of Windmill right now, and we go to
[21:14] of Windmill right now, and we go to configuration,
[21:16] configuration, you can see I have these questions
[21:18] you can see I have these questions already filled out, right? So I didn't
[21:20] already filled out, right? So I didn't have to go through and manually copy and
[21:23] have to go through and manually copy and paste stuff. It was just all there
[21:25] paste stuff. It was just all there waiting for me.
[21:26] waiting for me. Um and this saved you so much time
[21:28] Um and this saved you so much time rather than going through the tedious
[21:30] rather than going through the tedious process of setting up a performance
[21:31] process of setting up a performance review cycle.
[21:33] review cycle. Now we can take this a step further,
[21:34] Now we can take this a step further, right? Now let's say tomorrow I have to
[21:36] right? Now let's say tomorrow I have to present uh this entire plan to the
[21:39] present uh this entire plan to the entire company, let's create a
[21:41] entire company, let's create a presentation.
[21:43] presentation. Tomorrow I want to present this
[21:45] Tomorrow I want to present this performance review cycle to the entire
[21:47] performance review cycle to the entire company. Can you create a presentation
[21:49] company. Can you create a presentation in Gamma that explains the process,
[21:52] in Gamma that explains the process, explains the questions, and then also
[21:54] explains the questions, and then also explains a little bit about Windmill and
[21:56] explains a little bit about Windmill and how Windmill works?
[21:58] how Windmill works? We're going to do that. I'm actually
[22:00] We're going to do that. I'm actually going to copy a link here, right? So it
[22:03] going to copy a link here, right? So it knows a little bit about Windmill, but
[22:04] knows a little bit about Windmill, but if I just send it this document, it can
[22:07] if I just send it this document, it can know a lot more.
[22:10] know a lot more. And so this is a really good way of
[22:10] And so this is a really good way of feeding in context is just by sending
[22:12] feeding in context is just by sending links to relevant information. So this
[22:14] links to relevant information. So this is information about how Windmill works.
[22:16] is information about how Windmill works. Um
[22:17] Um actually
[22:19] actually I got a bad transcription.
[22:21] I got a bad transcription. Gamma, not Gemini.
[22:23] Gamma, not Gemini. So Gamma is a uh spreadsheet tool.
[22:26] So Gamma is a uh spreadsheet tool. Um and it's similar to Google Slides,
[22:29] Um and it's similar to Google Slides, but it's really AI native, so I find it
[22:30] but it's really AI native, so I find it work really well for these kind of use
[22:32] work really well for these kind of use cases. Uh you can see I have it
[22:34] cases. Uh you can see I have it connected. So my Gamma account is hooked
[22:37] connected. So my Gamma account is hooked up. And one thing about uh security here
[22:40] up. And one thing about uh security here is when I'm connecting Gamma or Notion
[22:41] is when I'm connecting Gamma or Notion or Slack, it is connecting using my
[22:44] or Slack, it is connecting using my personal authentication, right? So I'm
[22:46] personal authentication, right? So I'm confident that my agent can only do
[22:49] confident that my agent can only do things in these tools that I could also
[22:51] things in these tools that I could also do myself. Um and so that's a really
[22:54] do myself. Um and so that's a really helpful way to think about the security
[22:55] helpful way to think about the security side. Uh you are connecting tools um
[22:58] side. Uh you are connecting tools um with the same permissions you have in
[23:00] with the same permissions you have in the app natively. And it's really
[23:02] the app natively. And it's really important and similar similar about the
[23:04] important and similar similar about the actual being able to send sensitive
[23:05] actual being able to send sensitive data.
[23:06] data. Talk to your IT team. It's really
[23:07] Talk to your IT team. It's really important that you're able to connect
[23:08] important that you're able to connect these tools in a few order you if you
[23:10] these tools in a few order you if you want to get the maximum value out of
[23:11] want to get the maximum value out of them. So what it's actually doing right
[23:13] them. So what it's actually doing right now is actually creating a full
[23:14] now is actually creating a full presentation.
[23:15] presentation. Um
[23:16] Um I will let it we'll see.
[23:18] I will let it we'll see. I'll let it work for a little bit second
[23:20] I'll let it work for a little bit second here.
[23:21] here. Uh Dominick is saying Notion can also do
[23:23] Uh Dominick is saying Notion can also do presentations. Um yep, that's definitely
[23:25] presentations. Um yep, that's definitely true. Uh there's a lot of different
[23:26] true. Uh there's a lot of different tools you can do that.
[23:28] tools you can do that. Um and so right there it actually
[23:30] Um and so right there it actually created a full slide deck, right? And so
[23:32] created a full slide deck, right? And so if I open it in Gamma,
[23:35] if I open it in Gamma, you'll see this.
[23:37] you'll see this. It's generating. I'll actually jump to
[23:39] It's generating. I'll actually jump to one that was just I did right before
[23:41] one that was just I did right before this, which is the same thing. And here
[23:43] this, which is the same thing. And here is the presentation it created.
[23:45] is the presentation it created. And you can see it has an overview, it
[23:49] And you can see it has an overview, it has why we're doing this, it has the
[23:50] has why we're doing this, it has the cycle dates. You can see all this
[23:52] cycle dates. You can see all this information and the dates are from me,
[23:55] information and the dates are from me, the fact that self-reviews happen in
[23:56] the fact that self-reviews happen in Slack, that was from that Windy Windmill
[23:58] Slack, that was from that Windy Windmill help article.
[23:59] help article. Uh self-review, self-review questions,
[24:01] Uh self-review, self-review questions, manager review, all this information
[24:03] manager review, all this information right in here. Now this presentation
[24:05] right in here. Now this presentation maybe isn't perfectly on brand, uh maybe
[24:07] maybe isn't perfectly on brand, uh maybe you don't love the styling. There's a
[24:09] you don't love the styling. There's a lot of different ways of improving this
[24:10] lot of different ways of improving this with themes, right? So I could change
[24:12] with themes, right? So I could change the theme, um all sorts of different
[24:14] the theme, um all sorts of different things you can do, but I I always find
[24:17] things you can do, but I I always find that doing this is just a great first
[24:19] that doing this is just a great first step, right? And maybe it's good enough,
[24:20] step, right? And maybe it's good enough, maybe it's not, but it's so much faster
[24:22] maybe it's not, but it's so much faster to get all the content I need into this
[24:24] to get all the content I need into this presentation rather than doing it all
[24:25] presentation rather than doing it all completely manually. And you can see how
[24:27] completely manually. And you can see how quick that was. I just did that in 5
[24:28] quick that was. I just did that in 5 minutes and then I have a full
[24:29] minutes and then I have a full presentation. It's not perfect, but it's
[24:31] presentation. It's not perfect, but it's pretty damn good.
[24:35] Any questions, Nicole, that are
[24:37] Any questions, Nicole, that are worthwhile answering here?
[24:40] worthwhile answering here? Yeah, we've had a couple of them come
[24:41] Yeah, we've had a couple of them come in, one around some of the nuances for
[24:47] in, one around some of the nuances for performance reviews. So beyond creating
[24:49] performance reviews. So beyond creating the review questions and templates,
[24:50] the review questions and templates, digging into nuance around navigating
[24:52] digging into nuance around navigating things like LOAs, eligibility, etc.
[24:55] things like LOAs, eligibility, etc. Yeah, so you can certainly do all all
[24:57] Yeah, so you can certainly do all all that kind of stuff working with Claude
[24:59] that kind of stuff working with Claude and I could have showed a more advanced
[25:00] and I could have showed a more advanced use case where you only choose people
[25:02] use case where you only choose people who have been at the company for over
[25:04] who have been at the company for over than 6 months or anything like that. Um
[25:07] than 6 months or anything like that. Um most of that will depend that more
[25:08] most of that will depend that more complex logic will depend on the tool
[25:10] complex logic will depend on the tool you're using, right? So here we're using
[25:12] you're using, right? So here we're using Windmill and we can support all that
[25:13] Windmill and we can support all that type of functionality. Um when you're
[25:15] type of functionality. Um when you're prompting the model,
[25:17] prompting the model, if I go back to the presentation, I kind
[25:20] if I go back to the presentation, I kind of skipped over this. Um but it
[25:22] of skipped over this. Um but it basically said, "Hey, here's a preview."
[25:24] basically said, "Hey, here's a preview." Like the it shows a 7-day window for
[25:26] Like the it shows a 7-day window for self-reviews, a 7-day windows for
[25:27] self-reviews, a 7-day windows for manager reviews. Um I actually think
[25:30] manager reviews. Um I actually think these are way too long. We like to do
[25:31] these are way too long. We like to do things a lot faster. Uh but you can go
[25:33] things a lot faster. Uh but you can go back and forth kind of chatting about
[25:34] back and forth kind of chatting about what it is. And then often for that
[25:36] what it is. And then often for that final adjustments, I'll do it directly
[25:39] final adjustments, I'll do it directly in the tool itself, right? That's where
[25:40] in the tool itself, right? That's where it sometimes gets a little bit easier,
[25:41] it sometimes gets a little bit easier, whether it's a presentation or I want to
[25:43] whether it's a presentation or I want to make any other final tweaks, I can go
[25:45] make any other final tweaks, I can go into here and make any adjustments I
[25:46] into here and make any adjustments I want. Um so different options there, but
[25:49] want. Um so different options there, but I would recommend trying to get to like
[25:51] I would recommend trying to get to like 80 to 90% good in the AI and then if you
[25:54] 80 to 90% good in the AI and then if you want to jump into the tool, you can.
[25:57] want to jump into the tool, you can. So that's a quick overview of
[25:58] So that's a quick overview of performance reviews and creating cycle.
[26:01] performance reviews and creating cycle. The next thing I want to do is to talk
[26:04] The next thing I want to do is to talk about data analysis and we can actually
[26:06] about data analysis and we can actually do this in Claude Code or we could do
[26:09] do this in Claude Code or we could do this in Claude Co-work. Um we're going
[26:12] this in Claude Co-work. Um we're going to be uh putting up a quick poll and to
[26:15] to be uh putting up a quick poll and to see what people want me to do, whether
[26:17] see what people want me to do, whether it's Claude Code or it is Claude
[26:19] it's Claude Code or it is Claude Co-work.
[26:20] Co-work. Um and while we do that, uh I'm seeing a
[26:23] Um and while we do that, uh I'm seeing a bunch of questions around can it create
[26:25] bunch of questions around can it create presentations in Google Drive? Can
[26:27] presentations in Google Drive? Can Claude Code connect to Workday? Um
[26:31] Claude Code connect to Workday? Um the answer to all of that is it depends
[26:32] the answer to all of that is it depends and it's changing at a pretty rapid
[26:34] and it's changing at a pretty rapid pace. Um and so you're going to want
[26:37] pace. Um and so you're going to want when you're buying or using external
[26:39] when you're buying or using external tools, one of the things that you should
[26:41] tools, one of the things that you should always check out for is does this tool
[26:44] always check out for is does this tool support MCP? Does it have APIs? Uh not
[26:48] support MCP? Does it have APIs? Uh not every tool does it today. Over time more
[26:50] every tool does it today. Over time more and more tools are going to be
[26:51] and more tools are going to be supporting MCPs and APIs. So make sure
[26:54] supporting MCPs and APIs. So make sure you're using tools um that are helpful.
[26:57] you're using tools um that are helpful. All right, we're seeing a bunch of for
[26:58] All right, we're seeing a bunch of for Claude Code and Claude Co-work. Maybe
[26:59] Claude Code and Claude Co-work. Maybe we'll do a little bit of both. Um so
[27:02] we'll do a little bit of both. Um so let's get going.
[27:03] let's get going. So I'm going to open up this project.
[27:06] So I'm going to open up this project. And I have in here
[27:09] And I have in here this CSV
[27:11] this CSV and I'm going to ask Claude, can
[27:15] and I'm going to ask Claude, can you find the CSV
[27:17] you find the CSV and tell me what's in it?
[27:20] and tell me what's in it? So when you're doing data analysis, one
[27:22] So when you're doing data analysis, one of the first things I like to do is to
[27:25] of the first things I like to do is to have Claude understand the data set. Um
[27:28] have Claude understand the data set. Um and so here it found the CSV, this
[27:30] and so here it found the CSV, this onboarding pulse survey,
[27:32] onboarding pulse survey, and you can see it has 102 survey
[27:34] and you can see it has 102 survey responses, 13 columns. Some of this is
[27:38] responses, 13 columns. Some of this is numerical data, some of it's open-ended
[27:40] numerical data, some of it's open-ended feedback. Here are the five rating
[27:42] feedback. Here are the five rating questions. Uh here are the employees who
[27:44] questions. Uh here are the employees who covered themes at a glance. You can see
[27:46] covered themes at a glance. You can see it's already starting to do some
[27:46] it's already starting to do some analysis, right? Uh some basic stuff.
[27:52] Awesome. Now
[27:53] Awesome. Now when you're doing data analysis, one
[27:54] when you're doing data analysis, one thing to keep in mind is that the bigger
[27:57] thing to keep in mind is that the bigger the context window or the the more the
[27:59] the context window or the the more the number of rows, the more you need to be
[28:02] number of rows, the more you need to be careful
[28:03] careful um about what AI can reason over. So
[28:05] um about what AI can reason over. So every AI has a context window, uh some
[28:08] every AI has a context window, uh some somewhere around 100,000 to a million
[28:11] somewhere around 100,000 to a million tokens. And if you were doing this
[28:12] tokens. And if you were doing this analysis with say 100,000 survey
[28:15] analysis with say 100,000 survey responses, you're going to need to be a
[28:17] responses, you're going to need to be a little bit more careful about how it
[28:18] little bit more careful about how it reasons over it. With 100 survey
[28:20] reasons over it. With 100 survey responses, it's totally fine just to use
[28:22] responses, it's totally fine just to use the Claude context window.
[28:23] the Claude context window. So let's start asking some questions.
[28:26] So let's start asking some questions. I'm going to say, "What trend do you
[28:28] I'm going to say, "What trend do you identify
[28:30] identify in the data? Do some basic analysis."
[28:34] in the data? Do some basic analysis." And while that's running, I'll actually
[28:35] And while that's running, I'll actually just show you some people want a Claude
[28:37] just show you some people want a Claude Code, so I'll show you that. So here I'm
[28:39] Code, so I'll show you that. So here I'm going to open what's called Ghosty,
[28:41] going to open what's called Ghosty, which is a terminal. I have Claude Code
[28:43] which is a terminal. I have Claude Code open. You can see I'm in the folder and
[28:46] open. You can see I'm in the folder and I can do the exact same prompt.
[28:48] I can do the exact same prompt. And this is Claude Code. The all the
[28:49] And this is Claude Code. The all the techniques I'm showing you work really
[28:51] techniques I'm showing you work really well in Claude Code as well as uh Claude
[28:53] well in Claude Code as well as uh Claude Co-work. Uh you can try out both tools,
[28:56] Co-work. Uh you can try out both tools, they both work really well. Overall, the
[28:58] they both work really well. Overall, the level of capability is pretty much the
[29:00] level of capability is pretty much the same between Co-work and Code. So it's
[29:01] same between Co-work and Code. So it's more of a personal preference than
[29:03] more of a personal preference than anything else. And you can see we get
[29:05] anything else. And you can see we get the same type of information.
[29:08] We'll jump back to Claude Code.
[29:10] We'll jump back to Claude Code. >> [snorts]
[29:10] >> [snorts] >> Now one of the cool things, um
[29:13] >> Now one of the cool things, um so it says everywhere is working well
[29:15] so it says everywhere is working well except uh or engineer scores dropped
[29:17] except uh or engineer scores dropped sharply after the first month.
[29:19] sharply after the first month. Everywhere except engineering. Sales is
[29:21] Everywhere except engineering. Sales is in a crisis, it's saying. Uh right, so a
[29:23] in a crisis, it's saying. Uh right, so a bunch of information here.
[29:25] bunch of information here. Um and then I could say,
[29:28] Um and then I could say, "Can you give me some
[29:36] Can you give me some concrete
[29:37] Can you give me some concrete instructions to improve onboarding?"
[29:40] instructions to improve onboarding?" And so here I've turned it from just
[29:41] And so here I've turned it from just doing data analysis into actually giving
[29:43] doing data analysis into actually giving me recommendations. Now Now I find this,
[29:46] me recommendations. Now Now I find this, when I do stuff like this,
[29:48] when I do stuff like this, AI is not perfect. The expectation
[29:50] AI is not perfect. The expectation should not be that every single response
[29:52] should not be that every single response here is amazing and I'm going to
[29:53] here is amazing and I'm going to implement it blindly. But generally I
[29:55] implement it blindly. But generally I find it gives me some good ideas. So I
[29:57] find it gives me some good ideas. So I would look through these different
[29:58] would look through these different ideas. Some of these probably actually
[30:00] ideas. Some of these probably actually make a lot of sense. And I would then go
[30:02] make a lot of sense. And I would then go to implement them. So this is where
[30:04] to implement them. So this is where you're starting to use AI not just for
[30:05] you're starting to use AI not just for data analysis, but actually to give
[30:06] data analysis, but actually to give recommendations. And this is a super
[30:08] recommendations. And this is a super helpful way of using AI, but you can't
[30:11] helpful way of using AI, but you can't just take them all blindly, right? It's
[30:12] just take them all blindly, right? It's not going to be perfect, but generally
[30:13] not going to be perfect, but generally gives you really good ideas. And so
[30:15] gives you really good ideas. And so here's a bunch of ideas I could have a
[30:16] here's a bunch of ideas I could have a bunch more chat there.
[30:17] bunch more chat there. Now one of the cool things about using
[30:19] Now one of the cool things about using Claude for data analysis is historically
[30:21] Claude for data analysis is historically if you asked questions that were
[30:23] if you asked questions that were open-ended, it was extremely tedious
[30:26] open-ended, it was extremely tedious to understand what was in all the
[30:28] to understand what was in all the open-ended comments. And so what you did
[30:29] open-ended comments. And so what you did end up doing is create a lot of
[30:30] end up doing is create a lot of numerical questions just to make it
[30:32] numerical questions just to make it easier to analyze the data. With AI,
[30:36] easier to analyze the data. With AI, we've actually found that open-ended
[30:37] we've actually found that open-ended questions are by far the most helpful
[30:39] questions are by far the most helpful and it's super easy to reason about
[30:40] and it's super easy to reason about them. So, what are the common trends
[30:44] them. So, what are the common trends using the open-ended?
[30:50] And so now it's going to specifically
[30:51] And so now it's going to specifically look in these open-ended feedback,
[30:52] look in these open-ended feedback, right? Which you see here.
[30:57] All this stuff, right? Reading through
[30:58] All this stuff, right? Reading through all this would be insanely hard and
[31:00] all this would be insanely hard and time-consuming. Now it's extremely
[31:01] time-consuming. Now it's extremely trivial, right? Where Claude can just
[31:03] trivial, right? Where Claude can just reason over all those different things.
[31:05] reason over all those different things. I can show you also in Claude code.
[31:10] And it will do the same thing. Just a
[31:11] And it will do the same thing. Just a different UI, it's all pretty much the
[31:13] different UI, it's all pretty much the exact same.
[31:15] exact same. So here it is. Uh 101s make a better
[31:17] So here it is. Uh 101s make a better experience. Weekly 101s from day one is
[31:18] experience. Weekly 101s from day one is the most frequently praised thing.
[31:20] the most frequently praised thing. No 101s, right? So in this case it's
[31:21] No 101s, right? So in this case it's seeing that hey, people who have 101s
[31:23] seeing that hey, people who have 101s they like them and people who don't have
[31:25] they like them and people who don't have 101s they don't uh are having not nearly
[31:27] 101s they don't uh are having not nearly as good of an experience.
[31:29] as good of an experience. And I think with AI it's never been more
[31:31] And I think with AI it's never been more important to collect information in the
[31:34] important to collect information in the moment, right? And so for something like
[31:36] moment, right? And so for something like this onboarding, uh if you wait weeks
[31:39] this onboarding, uh if you wait weeks and weeks and weeks and undo like a
[31:40] and weeks and weeks and undo like a twice a year survey, everyone's going to
[31:43] twice a year survey, everyone's going to forget all these details.
[31:45] forget all these details. It's now much better to try to collect
[31:46] It's now much better to try to collect that information in the moment and then
[31:48] that information in the moment and then you you can use AI to understand it at
[31:50] you you can use AI to understand it at scale whenever you want. And so in
[31:52] scale whenever you want. And so in Windmill we have these lightweight
[31:53] Windmill we have these lightweight pulses where you can send Slack messages
[31:55] pulses where you can send Slack messages um at timed cadences to get this
[31:58] um at timed cadences to get this information. It doesn't really matter
[32:00] information. It doesn't really matter how you do it. The important thing is
[32:01] how you do it. The important thing is that you collect this information so
[32:03] that you collect this information so that you can then analyze it later.
[32:05] that you can then analyze it later. So we can take this a step further.
[32:07] So we can take this a step further. Let's say
[32:08] Let's say um who is a nutrition risk?
[32:13] And it will look through the data to try
[32:14] And it will look through the data to try to understand who is nutrition risk.
[32:18] And all this is fake data by the way. Um
[32:21] And all this is fake data by the way. Um just so we're all clear.
[32:23] just so we're all clear. Uh and if you wanted to omit employee
[32:24] Uh and if you wanted to omit employee names or anything before that, you can
[32:26] names or anything before that, you can certainly do that.
[32:27] certainly do that. So now it's going to understand who is
[32:29] So now it's going to understand who is try to understand who is nutrition risk
[32:31] try to understand who is nutrition risk across the company.
[32:33] across the company. While that's waiting I'll actually do a
[32:35] While that's waiting I'll actually do a follow-up question and take it one step
[32:38] follow-up question and take it one step further.
[32:42] Create a new column called attrition
[32:45] Create a new column called attrition risk that uses a weighted average of the
[32:47] risk that uses a weighted average of the other scores. Focus on the columns you
[32:49] other scores. Focus on the columns you think would be the highest indicator of
[32:50] think would be the highest indicator of attrition and weight those more heavily.
[32:52] attrition and weight those more heavily. Create a new CSV with this additional
[32:54] Create a new CSV with this additional column.
[32:57] So it did a basic uh
[32:59] So it did a basic uh attrition risk analysis where we see
[33:00] attrition risk analysis where we see some people um in the drop.
[33:03] some people um in the drop. All right? Um
[33:04] All right? Um but let's say I want to put a number on
[33:06] but let's say I want to put a number on every single person.
[33:08] every single person. This is where you can start to have
[33:10] This is where you can start to have Claude do some data analysis. So it's
[33:12] Claude do some data analysis. So it's actually going to create a new version
[33:14] actually going to create a new version of the CSV with an additional column,
[33:16] of the CSV with an additional column, right? So there's no attrition risk
[33:18] right? So there's no attrition risk column here. And that's super zoomed in
[33:21] column here. And that's super zoomed in zoomed out. Um but now I'm going to have
[33:23] zoomed out. Um but now I'm going to have Claude write a formula that's going to
[33:25] Claude write a formula that's going to weight the different columns and turn
[33:27] weight the different columns and turn this into a numerical score called
[33:29] this into a numerical score called attrition risk.
[33:31] attrition risk. Now the risk is going to run from one to
[33:33] Now the risk is going to run from one to five. Five is the highest attrition
[33:35] five. Five is the highest attrition risk.
[33:36] risk. One is the lowest.
[33:38] One is the lowest. You can see what's actually happening
[33:39] You can see what's actually happening under the hood. Um it's actually going
[33:41] under the hood. Um it's actually going to write code. It's going to write
[33:43] to write code. It's going to write Python. So this is what a kind of a if
[33:46] Python. So this is what a kind of a if you hired a data analyst, this is what
[33:47] you hired a data analyst, this is what they would do. They would write a script
[33:49] they would do. They would write a script like this. They would weight the
[33:51] like this. They would weight the different columns differently.
[33:53] different columns differently. And then it would create the score.
[33:55] And then it would create the score. I'm going to let that run.
[33:57] I'm going to let that run. It's validating hey, the scores make
[33:59] It's validating hey, the scores make sense.
[34:00] sense. And now it created a new CSV and because
[34:02] And now it created a new CSV and because this is Claude co-work, it did it right
[34:04] this is Claude co-work, it did it right on my file system.
[34:06] on my file system. And now there's a new column called
[34:07] And now there's a new column called attrition risk right at the end.
[34:10] attrition risk right at the end. Now is this is this number perfect?
[34:12] Now is this is this number perfect? Probably not. Um it's just doing data
[34:14] Probably not. Um it's just doing data analysis, but this is the exact same
[34:16] analysis, but this is the exact same workflow that like a people data analyst
[34:19] workflow that like a people data analyst or someone like that would do to try to
[34:20] or someone like that would do to try to formulate this kind of data.
[34:25] Now this is all helpful, but it's kind
[34:27] Now this is all helpful, but it's kind of back and forth chat. And all this
[34:28] of back and forth chat. And all this chat is private to me, right? No one
[34:31] chat is private to me, right? No one else can see this chat. Um no one else
[34:33] else can see this chat. Um no one else can see what's going on here. Let's say
[34:35] can see what's going on here. Let's say I wanted to turn this into more of an
[34:36] I wanted to turn this into more of an interactive dashboard so that anyone is
[34:39] interactive dashboard so that anyone is able to understand kind of some of these
[34:41] able to understand kind of some of these trends and look at it more visually.
[34:44] trends and look at it more visually. So this is where we can take it a step
[34:45] So this is where we can take it a step further.
[34:46] further. Um for this I'm going to use the web
[34:49] Um for this I'm going to use the web artifacts builder
[34:51] artifacts builder which is a way of it's a skill built
[34:53] which is a way of it's a skill built into Claude.
[34:55] into Claude. And it's and you might have to turn it
[34:57] And it's and you might have to turn it on.
[34:57] on. Um but it's a way of generating
[34:59] Um but it's a way of generating dashboards uh from data. And I'm going
[35:01] dashboards uh from data. And I'm going to give it a prompt.
[35:03] to give it a prompt. Can you create an interactive dashboard
[35:06] Can you create an interactive dashboard where on the homepage we see stuff by
[35:08] where on the homepage we see stuff by department showing a spider chart like a
[35:10] department showing a spider chart like a radar radar spider thing and then um on
[35:13] radar radar spider thing and then um on each department I want to have a drill
[35:14] each department I want to have a drill down um across 30 60 90. And then if I
[35:17] down um across 30 60 90. And then if I drill into a department I want to be
[35:18] drill into a department I want to be able to filter down um and everything
[35:21] able to filter down um and everything should change once you run that filter.
[35:23] should change once you run that filter. Um I want this to be very interactive.
[35:24] Um I want this to be very interactive. It should start as high level, but allow
[35:26] It should start as high level, but allow me to drill deep into the data.
[35:32] And that's going to start to work. Now
[35:33] And that's going to start to work. Now one thing with prompting is I was kind
[35:35] one thing with prompting is I was kind of rambling here.
[35:37] of rambling here. It's much better to ramble and give a
[35:38] It's much better to ramble and give a lot of thoughts than to try to like
[35:41] lot of thoughts than to try to like perfect something
[35:43] perfect something um perfectly, right? The more context
[35:45] um perfectly, right? The more context you give it the better. And so that's a
[35:47] you give it the better. And so that's a a lot why I like voice to text is it
[35:48] a lot why I like voice to text is it just causes you you generally speak a
[35:50] just causes you you generally speak a lot more than you type. And so doing
[35:52] lot more than you type. And so doing something like this um will often give a
[35:54] something like this um will often give a lot more context to to the system. Now
[35:57] lot more context to to the system. Now here it's actually building a full web
[35:59] here it's actually building a full web app. Uh it's installing PNPM. This is
[36:02] app. Uh it's installing PNPM. This is the stuff that a software engineer would
[36:03] the stuff that a software engineer would do. You no longer need to rely on some a
[36:05] do. You no longer need to rely on some a software engineer to do this kind of
[36:07] software engineer to do this kind of stuff.
[36:07] stuff. And so it's actually building this
[36:08] And so it's actually building this entire dashboard.
[36:11] entire dashboard. I'm not going to wait for it to finish
[36:12] I'm not going to wait for it to finish cuz I did this before. This will take a
[36:13] cuz I did this before. This will take a few minutes.
[36:14] few minutes. Let me just show you the result I got.
[36:16] Let me just show you the result I got. This was from one prompt.
[36:18] This was from one prompt. Um and I got it. Where is it? Right
[36:21] Um and I got it. Where is it? Right here.
[36:22] here. This is what Claude produced in one
[36:23] This is what Claude produced in one prompt. This is the same data set, one
[36:25] prompt. This is the same data set, one prompt asking it to build a web app. Um
[36:28] prompt asking it to build a web app. Um and here we see onboarding pulse
[36:29] and here we see onboarding pulse dashboard.
[36:30] dashboard. Total responses, average score.
[36:32] Total responses, average score. We can see by department. You can see
[36:35] We can see by department. You can see this is where you see sales is starting
[36:36] this is where you see sales is starting high, but then really dropping down. And
[36:38] high, but then really dropping down. And engineering is starting up there and
[36:40] engineering is starting up there and going there.
[36:42] going there. Here we see by department with these the
[36:43] Here we see by department with these the spider charts I asked for.
[36:45] spider charts I asked for. We see this information.
[36:47] We see this information. And this isn't just static. I can drill
[36:49] And this isn't just static. I can drill into this, right? So if I want to drill
[36:50] into this, right? So if I want to drill into engineering
[36:51] into engineering I can see it right here.
[36:53] I can see it right here. I can choose filters, right? You can
[36:56] I can choose filters, right? You can filter like that. Um and then I can see
[36:57] filter like that. Um and then I can see the actual employees sorted by the
[37:00] the actual employees sorted by the attrition risk column that I just
[37:02] attrition risk column that I just created. And then I could go in here and
[37:05] created. And then I could go in here and I could see the actual survey responses
[37:07] I could see the actual survey responses right from here.
[37:10] right from here. So this was one prompt. It took a few
[37:11] So this was one prompt. It took a few minutes to do this. But you can start to
[37:13] minutes to do this. But you can start to see how powerful this is for data
[37:16] see how powerful this is for data analysis. You can have that back and
[37:18] analysis. You can have that back and forth conversation. But to take it a
[37:20] forth conversation. But to take it a step further you can create full HTML
[37:23] step further you can create full HTML dashboards and then you can share those
[37:25] dashboards and then you can share those out. You could send the HTML to other
[37:27] out. You could send the HTML to other people or you could host it on a
[37:29] people or you could host it on a platform like Vercel and set up
[37:30] platform like Vercel and set up authentication and build a little mini
[37:32] authentication and build a little mini web app to share across the company or
[37:34] web app to share across the company or to the executive team or whatever the
[37:36] to the executive team or whatever the right audience is.
[37:38] right audience is. Um if you want to build this dashboard
[37:41] Um if you want to build this dashboard and keep live data into it uh that is
[37:44] and keep live data into it uh that is totally possible. You're going to need a
[37:45] totally possible. You're going to need a full API integration there. Um but you
[37:47] full API integration there. Um but you can totally do it. Uh for data analysis
[37:49] can totally do it. Uh for data analysis I like to start with just like an
[37:50] I like to start with just like an offline copy just cuz it's a lot easier.
[37:52] offline copy just cuz it's a lot easier. But if you wanted to power this with
[37:54] But if you wanted to power this with real data so it's kept up to date in
[37:55] real data so it's kept up to date in real time, that is also totally
[37:57] real time, that is also totally possible.
[38:01] So that is as you can see how this is
[38:03] So that is as you can see how this is still working. Um this will take a few
[38:04] still working. Um this will take a few minutes to run. When you're doing these
[38:05] minutes to run. When you're doing these more advanced agentic workflows, it is
[38:07] more advanced agentic workflows, it is going to take time to run. Um and that's
[38:10] going to take time to run. Um and that's where you want to often you'll see a lot
[38:12] where you want to often you'll see a lot of people talking about having multiple
[38:13] of people talking about having multiple tasks going at once or you just let this
[38:15] tasks going at once or you just let this go in the background and wait for it to
[38:16] go in the background and wait for it to finish.
[38:17] finish. So you can start to see with data
[38:19] So you can start to see with data analysis, my general workflow is to
[38:21] analysis, my general workflow is to start simple understand the scope of the
[38:23] start simple understand the scope of the data. You can then start to analyze a
[38:25] data. You can then start to analyze a little bit more.
[38:27] little bit more. You can add columns to turn unstructured
[38:30] You can add columns to turn unstructured data into structured data. And if you
[38:32] data into structured data. And if you want full visualizations, just have
[38:34] want full visualizations, just have Claude create a little web app. I find
[38:36] Claude create a little web app. I find it much easier than a lot of the
[38:37] it much easier than a lot of the visualization like BI tools. And this
[38:39] visualization like BI tools. And this was spun up on the fly and if I want to
[38:41] was spun up on the fly and if I want to make any adjustments, I could just go
[38:42] make any adjustments, I could just go back and forth with Claude.
[38:44] back and forth with Claude. And Max, we're getting some questions
[38:46] And Max, we're getting some questions about how can you save this context and
[38:48] about how can you save this context and reuse it each time?
[38:50] reuse it each time? Yeah, so that's where you'd want to
[38:51] Yeah, so that's where you'd want to create a skill, right? So if I really
[38:53] create a skill, right? So if I really like what this did I would want to turn
[38:55] like what this did I would want to turn it into a skill and I could have Claude
[38:58] it into a skill and I could have Claude do it. I could just say it's working
[38:59] do it. I could just say it's working right now.
[39:00] right now. Um but can you turn this process
[39:04] Um but can you turn this process into a reusable
[39:06] into a reusable skill?
[39:07] skill? I'm going to use the skill creator skill
[39:08] I'm going to use the skill creator skill which is right here.
[39:10] which is right here. Um and I will queue that up which means
[39:12] Um and I will queue that up which means it's going to run after this is started
[39:15] it's going to run after this is started to finish.
[39:16] to finish. Um and so we'll see what that happens.
[39:18] Um and so we'll see what that happens. Now
[39:19] Now there's a few questions around where can
[39:20] there's a few questions around where can I find that skill? So if I go to skills,
[39:23] I find that skill? So if I go to skills, I'm in Claude co-work. They have these
[39:25] I'm in Claude co-work. They have these example skills
[39:27] example skills which you're just seeing right here and
[39:29] which you're just seeing right here and I turned on the web artifacts builder.
[39:32] I turned on the web artifacts builder. Now there's nothing like mad if I zoom
[39:34] Now there's nothing like mad if I zoom out.
[39:35] out. There's nothing magical about
[39:36] There's nothing magical about [clears throat]
[39:37] [clears throat] the skill. Skills are really simple.
[39:39] the skill. Skills are really simple. They're just text documents. And so this
[39:41] They're just text documents. And so this is the web artifacts builder.
[39:43] is the web artifacts builder. It's just telling it kind of hey, use
[39:45] It's just telling it kind of hey, use react, typescript, V, parcel, tailwind,
[39:47] react, typescript, V, parcel, tailwind, right? It's just engineering stuff. It's
[39:49] right? It's just engineering stuff. It's telling it to try to avoid the AI slop,
[39:51] telling it to try to avoid the AI slop, so it doesn't have any purple,
[39:53] so it doesn't have any purple, um which you constantly see in in AI
[39:55] um which you constantly see in in AI slop. Um and yeah, skills are the same
[39:57] slop. Um and yeah, skills are the same thing as commands. They come in
[39:58] thing as commands. They come in different names. It's all the same
[39:59] different names. It's all the same stuff. It's all it is is like if I was
[40:01] stuff. It's all it is is like if I was just to copy and paste this in,
[40:04] just to copy and paste this in, literally take this,
[40:06] literally take this, and I go back and I go new task and I
[40:09] and I go back and I go new task and I say like create an HR dashboard,
[40:13] and I just copy and paste it in. That's
[40:14] and I just copy and paste it in. That's the exact same thing as the skill is
[40:16] the exact same thing as the skill is doing. It's just like a shortcut or like
[40:18] doing. It's just like a shortcut or like a little snippet. It's just a lot easier
[40:20] a little snippet. It's just a lot easier than having to remember to do that every
[40:22] than having to remember to do that every time. And some of them have been written
[40:24] time. And some of them have been written by people who um
[40:26] by people who um like understand have done this a bunch
[40:28] like understand have done this a bunch of times. And so, that's where you're
[40:29] of times. And so, that's where you're able to sh- like they they've tested the
[40:31] able to sh- like they they've tested the skill out. They made some tweaks.
[40:32] skill out. They made some tweaks. They've run this workflow a bunch of
[40:34] They've run this workflow a bunch of times. And so, they know it's a really
[40:36] times. And so, they know it's a really good way to do things.
[40:38] good way to do things. Um skills are available to download, um
[40:41] Um skills are available to download, um as well as you can create your own.
[40:42] as well as you can create your own. Right? So, if I go to skills, here's a
[40:43] Right? So, if I go to skills, here's a bunch of example skills. Um
[40:46] bunch of example skills. Um I could create my own from here. Um you
[40:48] I could create my own from here. Um you can also like the plugin. So, like the
[40:50] can also like the plugin. So, like the the HR plugin was a thing a lot of
[40:52] the HR plugin was a thing a lot of people talked about. Um
[40:54] people talked about. Um and this is the HR plugin right here.
[40:56] and this is the HR plugin right here. Here are the skills in the HR plugin.
[40:58] Here are the skills in the HR plugin. And so, I could
[41:00] And so, I could I don't know why it's not loading.
[41:01] I don't know why it's not loading. Claude co-work can be a little bit
[41:03] Claude co-work can be a little bit janky. Um here's a skill around comp
[41:05] janky. Um here's a skill around comp analysis. So, if anyone saw
[41:08] analysis. So, if anyone saw uh Claude release this HR plugin, this
[41:10] uh Claude release this HR plugin, this is the plugin. Um I can show you how to
[41:12] is the plugin. Um I can show you how to add a plugin. I can browse plugins.
[41:14] add a plugin. I can browse plugins. Here are all the different plugins,
[41:16] Here are all the different plugins, right? This is the human resources one
[41:17] right? This is the human resources one which I added. There's design,
[41:19] which I added. There's design, engineering. Under the hood, all of
[41:21] engineering. Under the hood, all of these are just bundles of skills. Um
[41:24] these are just bundles of skills. Um And so, this comp analysis skill you see
[41:26] And so, this comp analysis skill you see right here, it's actually pretty
[41:27] right here, it's actually pretty straightforward. It's not that
[41:28] straightforward. It's not that complicated. It's just basically, hey,
[41:30] complicated. It's just basically, hey, this is what a market benchmark should
[41:32] this is what a market benchmark should look like. So, skill these built-in
[41:34] look like. So, skill these built-in skills I find are a good place to kind
[41:35] skills I find are a good place to kind of get started, but generally you're
[41:37] of get started, but generally you're going to want to customize them a lot
[41:39] going to want to customize them a lot more to fit exactly your use case. And
[41:41] more to fit exactly your use case. And so, I could go in here and I could edit
[41:43] so, I could go in here and I could edit um I could open it in like cursor or I
[41:45] um I could open it in like cursor or I could customize it with Claude.
[41:47] could customize it with Claude. Skills are the same as gems in Gemini.
[41:50] Skills are the same as gems in Gemini. For people who use Gemini, skills are
[41:51] For people who use Gemini, skills are the exact same thing.
[41:53] the exact same thing. Um if you want to install the plugins,
[41:55] Um if you want to install the plugins, go right into Claude co-work,
[41:57] go right into Claude co-work, browse plugins,
[41:59] browse plugins, and here are the different plugins. So,
[42:00] and here are the different plugins. So, this is how you install it. I don't have
[42:02] this is how you install it. I don't have it enabled right now. I actually find
[42:03] it enabled right now. I actually find the skill sometimes is annoying, but if
[42:05] the skill sometimes is annoying, but if I wanted to enable it, I could enable it
[42:06] I wanted to enable it, I could enable it right there. Uh all it is is a list of
[42:09] right there. Uh all it is is a list of skills. And a skill is just instructions
[42:12] skills. And a skill is just instructions to the model of what to do.
[42:17] All right. Any other questions, Nicole,
[42:19] All right. Any other questions, Nicole, about this demo?
[42:22] about this demo? One that we've gotten across the demos,
[42:24] One that we've gotten across the demos, a couple of questions around how do you
[42:26] a couple of questions around how do you share and collaborate with what you're
[42:27] share and collaborate with what you're doing in Claude with the rest of your
[42:28] doing in Claude with the rest of your team, whether that be skills you want to
[42:30] team, whether that be skills you want to share or other work that you're working
[42:32] share or other work that you're working on.
[42:33] on. Yeah, so there's not actually sharing
[42:35] Yeah, so there's not actually sharing skills is not as easy as it should be
[42:37] skills is not as easy as it should be right now.
[42:38] right now. Um I'll show you a few examples of how
[42:40] Um I'll show you a few examples of how you can share stuff. So, one thing you
[42:42] you can share stuff. So, one thing you can do is just share a chat. Um very
[42:44] can do is just share a chat. Um very helpful thing to do. Uh if I go Where is
[42:47] helpful thing to do. Uh if I go Where is the share chat?
[42:49] the share chat? They've moved the button on me.
[42:53] Hm.
[42:54] Hm. >> [clears throat]
[42:54] >> [clears throat] >> Um maybe you can't you not share
[42:57] >> Um maybe you can't you not share Maybe you can't share it in um co-work.
[43:00] Maybe you can't share it in um co-work. But in chat, if I have a chat, um so
[43:02] But in chat, if I have a chat, um so here, I don't want to share it. I go
[43:04] here, I don't want to share it. I go right here and I can share the chat. So,
[43:06] right here and I can share the chat. So, this is a helpful way to share like
[43:07] this is a helpful way to share like here's how I prompt, here are common
[43:08] here's how I prompt, here are common workflows. If you have a reusable thing,
[43:11] workflows. If you have a reusable thing, that's where you're going to want to
[43:12] that's where you're going to want to share a skill. And so, if I click here,
[43:14] share a skill. And so, if I click here, I can also turn into skill or ask
[43:16] I can also turn into skill or ask Claude. Um if we go turn to skill,
[43:19] Claude. Um if we go turn to skill, all it does is just put this prompt in
[43:20] all it does is just put this prompt in here. Um
[43:21] here. Um and it would create that skill as we do
[43:22] and it would create that skill as we do this.
[43:24] this. It will create the skill for me. Now,
[43:26] It will create the skill for me. Now, how do you share the skill with other
[43:27] how do you share the skill with other people at the company?
[43:29] people at the company? The simplest way is actually the way I
[43:31] The simplest way is actually the way I showed before, where you just take this
[43:33] showed before, where you just take this zip file, right?
[43:35] zip file, right? Maybe zoom in a little bit. People can
[43:37] Maybe zoom in a little bit. People can see that.
[43:38] see that. Um you just send this We just send the
[43:41] Um you just send this We just send the zip file in Slack. Um it's not the
[43:43] zip file in Slack. Um it's not the fanciest thing, but it's a really easy
[43:44] fanciest thing, but it's a really easy way of sharing skills right now. And a
[43:47] way of sharing skills right now. And a skill, as I said before, is just a
[43:49] skill, as I said before, is just a markdown file, and markdown is just
[43:51] markdown file, and markdown is just text. It sounds a lot fancier than it
[43:54] text. It sounds a lot fancier than it is. All it is is sharing text snippets.
[43:57] is. All it is is sharing text snippets. Um so, sharing zip files is a really
[43:59] Um so, sharing zip files is a really easy way to do it. If you want to go one
[44:01] easy way to do it. If you want to go one step further, you can set up a GitHub
[44:05] step further, you can set up a GitHub plugin, um which it seems to be what
[44:07] plugin, um which it seems to be what Claude is moving towards. And so, if I
[44:09] Claude is moving towards. And so, if I go to custom and I go to
[44:11] go to custom and I go to add a plugin,
[44:13] add a plugin, and I see these your organization. I
[44:16] and I see these your organization. I don't We don't have any set up for us,
[44:17] don't We don't have any set up for us, but if you go into GitHub and you create
[44:19] but if you go into GitHub and you create a GitHub repository and put the skills
[44:22] a GitHub repository and put the skills in there, you can make it so anyone can
[44:24] in there, you can make it so anyone can install them with one click right from
[44:26] install them with one click right from here. You're going to want IT
[44:28] here. You're going to want IT um to obviously help with this. Uh but
[44:30] um to obviously help with this. Uh but if you really want to take it to the
[44:31] if you really want to take it to the next level, you can set it up. It is
[44:33] next level, you can set it up. It is still pretty clunky, and I think you
[44:34] still pretty clunky, and I think you need to give everyone GitHub access,
[44:36] need to give everyone GitHub access, which is why we haven't done it. So,
[44:38] which is why we haven't done it. So, what we do is share skills just in
[44:40] what we do is share skills just in Slack. Um if you're trying to drive AI
[44:42] Slack. Um if you're trying to drive AI adoption, the more learning and
[44:44] adoption, the more learning and development you can do, the better. Uh
[44:45] development you can do, the better. Uh we find lunch and learns to be extremely
[44:47] we find lunch and learns to be extremely helpful, where people just share cool
[44:49] helpful, where people just share cool thing they've done.
[44:50] thing they've done. Uh we also have a um shout-outs channel,
[44:52] Uh we also have a um shout-outs channel, where people do something cool
[44:54] where people do something cool uh with uh AI. They just show exactly
[44:57] uh with uh AI. They just show exactly what they did. They show a demo or a
[44:58] what they did. They show a demo or a loom or a recording. The more you can
[45:00] loom or a recording. The more you can encourage information sharing across the
[45:02] encourage information sharing across the company, the better.
[45:05] company, the better. So, I wanted to stop the demos right
[45:07] So, I wanted to stop the demos right there and go through a few slides. Um
[45:10] there and go through a few slides. Um so, let's
[45:11] so, let's let's just do that. Um some other
[45:14] let's just do that. Um some other questions we've had I just want to kind
[45:15] questions we've had I just want to kind of highlight.
[45:17] of highlight. Let's see.
[45:20] All right.
[45:21] All right. So, if we bring
[45:27] Uh I think some of these will answer a
[45:29] Uh I think some of these will answer a few of the common questions here.
[45:31] few of the common questions here. Um so, the first thing that I was
[45:33] Um so, the first thing that I was talking about before is driving AI
[45:34] talking about before is driving AI adoption. Uh we have a three-step kind
[45:37] adoption. Uh we have a three-step kind of program here of how we do about this
[45:38] of program here of how we do about this internally. Um the first is measuring
[45:42] internally. Um the first is measuring what matters. Uh if you can't measure
[45:44] what matters. Uh if you can't measure it, you're never going to be able to
[45:45] it, you're never going to be able to drive AI adoption as effectively as you
[45:47] drive AI adoption as effectively as you want. And the way we do this is in two
[45:49] want. And the way we do this is in two different ways. One is by actually
[45:51] different ways. One is by actually looking into Claude. Claude has
[45:52] looking into Claude. Claude has analytics. Um so, if we go into here,
[45:55] analytics. Um so, if we go into here, this is like Claude analytics. There's a
[45:57] this is like Claude analytics. There's a bunch of different ways of getting
[45:58] bunch of different ways of getting analytics for your team.
[46:00] analytics for your team. Um oh, sorry. You can't see my screen.
[46:01] Um oh, sorry. You can't see my screen. But in Claude analytics, um
[46:04] But in Claude analytics, um uh you can start to see that. And so,
[46:06] uh you can start to see that. And so, you want to use this telemetry to
[46:07] you want to use this telemetry to understand how are people using Claude
[46:09] understand how are people using Claude today and is it increasing.
[46:11] today and is it increasing. The other thing that's really helpful is
[46:12] The other thing that's really helpful is pulse surveys. So, once a week, we ping
[46:14] pulse surveys. So, once a week, we ping every single employee, ask them, "Hey,
[46:16] every single employee, ask them, "Hey, how how what tools are you using for AI?
[46:18] how how what tools are you using for AI? How is it going? How could it go
[46:20] How is it going? How could it go better?" And you get these point-in-time
[46:22] better?" And you get these point-in-time snapshots that you can track over time.
[46:26] The second thing that's really important
[46:28] The second thing that's really important is to use your existing rituals to drive
[46:30] is to use your existing rituals to drive AI adoption. If AI is really important,
[46:32] AI adoption. If AI is really important, you better be using it to prepare for
[46:34] you better be using it to prepare for one-on-ones. And then within each
[46:35] one-on-ones. And then within each one-on-one, you better be talking about,
[46:37] one-on-one, you better be talking about, "Hey, how is your AI adoption going?
[46:39] "Hey, how is your AI adoption going? What are you struggling? What tools are
[46:41] What are you struggling? What tools are you using? How could we unblock you?"
[46:42] you using? How could we unblock you?" Right? If IT is not letting you install
[46:45] Right? If IT is not letting you install the app that you want to use, you really
[46:47] the app that you want to use, you really want to know that.
[46:48] want to know that. Performance reviews. If AI adoption is
[46:50] Performance reviews. If AI adoption is really critical, it better be a question
[46:52] really critical, it better be a question in your performance review. Managers
[46:54] in your performance review. Managers should be rating their employees about,
[46:55] should be rating their employees about, "Hey, how good are you adopting AI?" It
[46:57] "Hey, how good are you adopting AI?" It should be part of the self-review.
[46:59] should be part of the self-review. Uh and if you want to take it a step
[47:00] Uh and if you want to take it a step further, we've actually heard of some
[47:01] further, we've actually heard of some companies that are tying it directly to
[47:02] companies that are tying it directly to compensation, where the more you use AI,
[47:05] compensation, where the more you use AI, that is going to be an input into
[47:07] that is going to be an input into compensation decisions.
[47:09] compensation decisions. Um and then obviously recognize, right?
[47:10] Um and then obviously recognize, right? It's It's about recognizing the people,
[47:12] It's It's about recognizing the people, but also recognizing the workflows. And
[47:14] but also recognizing the workflows. And this is where lunch and learns um and
[47:16] this is where lunch and learns um and other types of information sharing can
[47:18] other types of information sharing can be extremely helpful.
[47:20] be extremely helpful. The one point of caution is AI slop is
[47:23] The one point of caution is AI slop is what the term everyone uses now, right?
[47:24] what the term everyone uses now, right? It's very easy. You saw how quickly I
[47:26] It's very easy. You saw how quickly I could could create this massive
[47:28] could could create this massive presentation. Um
[47:30] presentation. Um That presentation was pretty good, uh
[47:32] That presentation was pretty good, uh but it certainly wasn't perfect. And so,
[47:34] but it certainly wasn't perfect. And so, you need to be careful about com-
[47:36] you need to be careful about com- basically saying output equals
[47:38] basically saying output equals performance, right? They're very
[47:39] performance, right? They're very different things. It sounds very easy to
[47:40] different things. It sounds very easy to write a 10-page document. It's very easy
[47:43] write a 10-page document. It's very easy to write a 100-page document. It doesn't
[47:44] to write a 100-page document. It doesn't mean it's good. It doesn't mean anyone's
[47:45] mean it's good. It doesn't mean anyone's actually reviewed it. Uh we have a
[47:47] actually reviewed it. Uh we have a policy here where if you're sending out
[47:49] policy here where if you're sending out AI stuff, you you have to review it
[47:52] AI stuff, you you have to review it first. And if anyone ever sends me just
[47:54] first. And if anyone ever sends me just like a totally generated thing that they
[47:55] like a totally generated thing that they never even looked at, I'm going to send
[47:57] never even looked at, I'm going to send it right back to them. Um and this is
[47:58] it right back to them. Um and this is something you need to be careful about.
[47:59] something you need to be careful about. But this isn't a reason not to adopt AI.
[48:01] But this isn't a reason not to adopt AI. Uh slop is a thing, but uh the reason it
[48:04] Uh slop is a thing, but uh the reason it comes is you're not doing a good job
[48:05] comes is you're not doing a good job with your prompting. You're not really
[48:06] with your prompting. You're not really understanding what's going on over the
[48:08] understanding what's going on over the hood.
[48:09] hood. So, any questions about AI adoption? Uh
[48:13] So, any questions about AI adoption? Uh so, um I see a question around when we
[48:15] so, um I see a question around when we ping people for the survey, do you use
[48:16] ping people for the survey, do you use do it with Claude? Um we actually uh use
[48:19] do it with Claude? Um we actually uh use Windmill to do that. Uh Windmill has a
[48:21] Windmill to do that. Uh Windmill has a pulse surveys feature. It sends out a
[48:22] pulse surveys feature. It sends out a Slack.
[48:23] Slack. Uh our agent has a conversation with
[48:25] Uh our agent has a conversation with every single individual. And then I'll
[48:27] every single individual. And then I'll use Claude to do analysis of the uh
[48:29] use Claude to do analysis of the uh information.
[48:33] Um
[48:34] Um Any other questions here, Nicole, that
[48:36] Any other questions here, Nicole, that are worthwhile answering about driving
[48:38] are worthwhile answering about driving adoption?
[48:40] Nothing on adoption right now.
[48:43] Nothing on adoption right now. Um Okay.
[48:43] Um Okay. A few that we can cover, I think, after
[48:45] A few that we can cover, I think, after we handle data privacy security.
[48:48] we handle data privacy security. Okay. Um
[48:49] Okay. Um the next thing I want to quickly mention
[48:50] the next thing I want to quickly mention is around buying tools. I mentioned this
[48:52] is around buying tools. I mentioned this before, but it's in pretty insane to buy
[48:56] before, but it's in pretty insane to buy a tool these days that did not does not
[48:57] a tool these days that did not does not offer kind of MCPs or APIs or at least
[49:00] offer kind of MCPs or APIs or at least it's not on the short-term roadmap. Um
[49:02] it's not on the short-term roadmap. Um and there's been a few companies I
[49:03] and there's been a few companies I actually I I will call out um Workday
[49:06] actually I I will call out um Workday made some
[49:07] made some some comments recently about how they
[49:08] some comments recently about how they might charge for API adoption or
[49:10] might charge for API adoption or something like that.
[49:13] If you want to take advantage of these
[49:15] If you want to take advantage of these workflows, you want to take advantage of
[49:16] workflows, you want to take advantage of AI, you need to work with open
[49:17] AI, you need to work with open platforms. This means at a minimum APIs
[49:20] platforms. This means at a minimum APIs and MCP support, so that your agent can
[49:23] and MCP support, so that your agent can access all the same data that you could
[49:25] access all the same data that you could do otherwise. And so, connectivity, I
[49:27] do otherwise. And so, connectivity, I would say it's a must-have as you think
[49:28] would say it's a must-have as you think about buying or using external tools. Um
[49:31] about buying or using external tools. Um definitely we're in the process of
[49:32] definitely we're in the process of opening up all of Windmill to to be able
[49:34] opening up all of Windmill to to be able to do all this kind of stuff. Uh if you
[49:36] to do all this kind of stuff. Uh if you want to do the workflows that I showed,
[49:37] want to do the workflows that I showed, like that Gamma creation, the reason I
[49:39] like that Gamma creation, the reason I use Gamma not Google Slides is cuz they
[49:41] use Gamma not Google Slides is cuz they had a better API to do that.
[49:43] had a better API to do that. Um
[49:43] Um the next thing is around asking the
[49:45] the next thing is around asking the right questions
[49:47] right questions uh when it comes to security. You're
[49:48] uh when it comes to security. You're only going to be able to use AI in a
[49:50] only going to be able to use AI in a really effective way if you feel
[49:51] really effective way if you feel comfortable pretty much sharing any kind
[49:53] comfortable pretty much sharing any kind of information. Now, there's certainly
[49:55] of information. Now, there's certainly exceptions, maybe social security
[49:57] exceptions, maybe social security numbers and credit card numbers and
[49:58] numbers and credit card numbers and stuff like that, but in general, you
[50:00] stuff like that, but in general, you want to feel pretty comfortable sharing
[50:02] want to feel pretty comfortable sharing information in order to get the value
[50:04] information in order to get the value out of it, right? The same way you would
[50:05] out of it, right? The same way you would share information with another employee.
[50:07] share information with another employee. And so when you're looking at tools, a
[50:08] And so when you're looking at tools, a few things to look at. One is zero day
[50:09] few things to look at. One is zero day retention. Um pretty much everyone
[50:11] retention. Um pretty much everyone offers this. Um this means that the
[50:14] offers this. Um this means that the model provider will not be saving the
[50:16] model provider will not be saving the data, will not be saving the logs, and
[50:18] data, will not be saving the logs, and will not train off of your data. This is
[50:21] will not train off of your data. This is generally something you can opt into or
[50:22] generally something you can opt into or at least at the enterprise levels, it's
[50:24] at least at the enterprise levels, it's always a good thing to look at. Um
[50:27] always a good thing to look at. Um so that's that's a key thing. The other
[50:28] so that's that's a key thing. The other thing is are they going to be using data
[50:29] thing is are they going to be using data for training and inference? Sometimes
[50:31] for training and inference? Sometimes this is opt in or opt out. Um if you're
[50:33] this is opt in or opt out. Um if you're using sensitive data, you can always opt
[50:35] using sensitive data, you can always opt out of this. Pretty much everyone
[50:36] out of this. Pretty much everyone supports opting out of training um at
[50:38] supports opting out of training um at least if you pay enough money. Um and so
[50:40] least if you pay enough money. Um and so you if you're working with sensitive
[50:42] you if you're working with sensitive data, I would recommend opting out of
[50:43] data, I would recommend opting out of training
[50:44] training um and making sure that they can't use
[50:46] um and making sure that they can't use your data to train on it. And once
[50:47] your data to train on it. And once you've done that, um and most companies
[50:49] you've done that, um and most companies I talked to, you start to feel pretty
[50:51] I talked to, you start to feel pretty comfortable sending the data and the
[50:52] comfortable sending the data and the benefits greatly out lie outweigh the
[50:55] benefits greatly out lie outweigh the risks. Uh I don't think you're able to
[50:57] risks. Uh I don't think you're able to get the full value of AI if you're super
[50:59] get the full value of AI if you're super sensitive about what data is being sent
[51:00] sensitive about what data is being sent in and what's not.
[51:03] in and what's not. Um and then this third thing is around,
[51:04] Um and then this third thing is around, hey, when you find a workflow that
[51:05] hey, when you find a workflow that works, share it widely, share it across
[51:07] works, share it widely, share it across your organization, let everyone know
[51:08] your organization, let everyone know about it. Everyone's trying to up level
[51:10] about it. Everyone's trying to up level their game. A lot of this stuff is
[51:12] their game. A lot of this stuff is clunky. Cowork is like super buggy. Uh
[51:14] clunky. Cowork is like super buggy. Uh there's a lot of issues right now and
[51:15] there's a lot of issues right now and stuff's changing on like a daily basis.
[51:17] stuff's changing on like a daily basis. It's really hard to stay up to date on
[51:18] It's really hard to stay up to date on everything. Uh so definitely creating a
[51:21] everything. Uh so definitely creating a culture of information sharing is an
[51:23] culture of information sharing is an absolute must.
[51:26] The final thing I wanted to talk about
[51:28] The final thing I wanted to talk about um is where AI, especially with like uh
[51:31] um is where AI, especially with like uh difficult
[51:32] difficult uh management decisions. Where you
[51:34] uh management decisions. Where you should AI use AI and where you
[51:35] should AI use AI and where you shouldn't. Uh and this is really
[51:37] shouldn't. Uh and this is really critical and this is a lot about how we
[51:38] critical and this is a lot about how we think about the design of Windmill.
[51:41] think about the design of Windmill. Um
[51:42] Um you want to use AI where AI is really
[51:43] you want to use AI where AI is really good. AI is really good at gathering
[51:45] good. AI is really good at gathering information, um
[51:47] information, um pulling it all together, looking at tons
[51:49] pulling it all together, looking at tons of data sources, drafting a narrative,
[51:50] of data sources, drafting a narrative, finding patterns, right? AI is really
[51:52] finding patterns, right? AI is really good at that.
[51:53] good at that. Where AI is not that good and where I
[51:55] Where AI is not that good and where I still 100% think AI should uh humans
[51:57] still 100% think AI should uh humans should be in full control is around the
[51:59] should be in full control is around the key decisions. Especially key decisions
[52:02] key decisions. Especially key decisions around hiring or promotions or
[52:04] around hiring or promotions or terminations or performance actions or
[52:07] terminations or performance actions or review ratings or calibration decisions.
[52:09] review ratings or calibration decisions. All of this critical stuff, you should
[52:11] All of this critical stuff, you should never delegate to AI, especially right
[52:13] never delegate to AI, especially right now. Not only is I don't think the
[52:14] now. Not only is I don't think the results you get are going to be very
[52:15] results you get are going to be very good, um
[52:17] good, um but you just lose accountability across
[52:18] but you just lose accountability across the organization. So you want to use AI
[52:20] the organization. So you want to use AI where it's helpful, but you still want
[52:22] where it's helpful, but you still want humans and managers and whoever it is to
[52:24] humans and managers and whoever it is to be the ultimate judge, right? And they
[52:25] be the ultimate judge, right? And they are using AI to make better decisions,
[52:27] are using AI to make better decisions, not to kind of skip over this really
[52:29] not to kind of skip over this really critical thing. Um and just as an
[52:31] critical thing. Um and just as an example,
[52:32] example, if you blindly use AI to like evaluate
[52:34] if you blindly use AI to like evaluate someone's performance and you're just
[52:36] someone's performance and you're just like, hey, how's this person performing?
[52:37] like, hey, how's this person performing? Pretty much all the time AI is going to
[52:39] Pretty much all the time AI is going to say something pretty positive. It's
[52:40] say something pretty positive. It's going to say, hey, this person's doing
[52:41] going to say, hey, this person's doing well. Uh AI has been fine-tuned to be
[52:43] well. Uh AI has been fine-tuned to be pretty nice uh if you use any of these
[52:45] pretty nice uh if you use any of these chatbots. And you can't at this point
[52:48] chatbots. And you can't at this point really trust it to make these critical
[52:49] really trust it to make these critical judgment calls, nor do you really want
[52:51] judgment calls, nor do you really want to. Uh this is still the exact place
[52:53] to. Uh this is still the exact place where you want people to make those
[52:55] where you want people to make those calls, but you don't want them to have
[52:56] calls, but you don't want them to have to rely on their memory. You want them
[52:58] to rely on their memory. You want them to be informed by data and AI is really
[53:00] to be informed by data and AI is really good at reasoning over large sets of
[53:01] good at reasoning over large sets of data.
[53:03] data. Uh another thing to call out here is the
[53:04] Uh another thing to call out here is the legal landscape. Uh we're in New York
[53:06] legal landscape. Uh we're in New York City, so uh local law 144 is uh
[53:09] City, so uh local law 144 is uh something we talk a lot about, but every
[53:10] something we talk a lot about, but every state is starting to enact uh different
[53:12] state is starting to enact uh different regulation. In Europe, there's a bunch
[53:13] regulation. In Europe, there's a bunch of different rules in Germany. Um you've
[53:15] of different rules in Germany. Um you've got to be very careful about how you're
[53:17] got to be very careful about how you're using AI. In general, if you follow this
[53:20] using AI. In general, if you follow this framework at the top, you're generally
[53:22] framework at the top, you're generally going to fit pretty cleanly within the
[53:23] going to fit pretty cleanly within the laws. The laws are all primarily about
[53:25] laws. The laws are all primarily about things like um hey, if you're
[53:27] things like um hey, if you're automatically rejecting candidates using
[53:29] automatically rejecting candidates using only AI, um and there was actually a
[53:31] only AI, um and there was actually a recent case about that, or you're
[53:33] recent case about that, or you're deciding promotions entirely using AI,
[53:35] deciding promotions entirely using AI, uh we are still not ready for that type
[53:37] uh we are still not ready for that type of automation.
[53:38] of automation. Um but we'll see how things evolve over
[53:40] Um but we'll see how things evolve over time. This is a pretty safe spot to be
[53:42] time. This is a pretty safe spot to be right now, but definitely be aware of
[53:44] right now, but definitely be aware of all legal laws that affect your company
[53:46] all legal laws that affect your company or or or your employees.
[53:50] All right. Um
[53:51] All right. Um I know I was talking fast. I had a lot
[53:53] I know I was talking fast. I had a lot to cover.
[53:54] to cover. Um if there's we now have time for a few
[53:58] Um if there's we now have time for a few more questions, but I wanted to turn
[53:59] more questions, but I wanted to turn over to Nicole for some next steps.
[54:02] over to Nicole for some next steps. Yeah, thank you so much to everyone who
[54:03] Yeah, thank you so much to everyone who joined us today. A couple of quick
[54:05] joined us today. A couple of quick things. So I did see some requests for a
[54:06] things. So I did see some requests for a demo, which is fantastic. Uh Ben
[54:09] demo, which is fantastic. Uh Ben actually shared a direct link to book a
[54:11] actually shared a direct link to book a a Windmill demo if you want to do so.
[54:13] a Windmill demo if you want to do so. That uh comment is pinned right now.
[54:15] That uh comment is pinned right now. Also, after this, I saw a lot of
[54:17] Also, after this, I saw a lot of questions about what we'll be sharing.
[54:18] questions about what we'll be sharing. We'll be sharing the slides, the video,
[54:20] We'll be sharing the slides, the video, and a couple of other resources
[54:22] and a couple of other resources including an AI adoption survey. We'd
[54:24] including an AI adoption survey. We'd love to get a really good sense of how
[54:26] love to get a really good sense of how people are using AI internally at their
[54:28] people are using AI internally at their companies right now. What are barriers
[54:30] companies right now. What are barriers entry? What are the issues that people
[54:32] entry? What are the issues that people are currently facing in making that
[54:34] are currently facing in making that adoption more widespread? That's
[54:36] adoption more widespread? That's something we'd love to package up and
[54:37] something we'd love to package up and share as a resource to people as well.
[54:39] share as a resource to people as well. And lastly, if you learned anything from
[54:41] And lastly, if you learned anything from this webinar and you want to tag us on
[54:43] this webinar and you want to tag us on LinkedIn, share a post, uh we'd love to
[54:45] LinkedIn, share a post, uh we'd love to hear about it. We're going to share some
[54:46] hear about it. We're going to share some written notes as well using Granola,
[54:48] written notes as well using Granola, another fantastic uh recording app. If
[54:51] another fantastic uh recording app. If you have if you haven't used it before,
[54:52] you have if you haven't used it before, highly recommend doing so. That'll share
[54:54] highly recommend doing so. That'll share a summary, key points, everything that
[54:56] a summary, key points, everything that Max talked about in a very easy to
[54:57] Max talked about in a very easy to digest written format. Um if there's
[54:59] digest written format. Um if there's anything else you're curious in seeing
[55:01] anything else you're curious in seeing as a resource, just let me know and I'll
[55:03] as a resource, just let me know and I'll add it up to the follow-up email.
[55:05] add it up to the follow-up email. All right. And I will stay on as long as
[55:07] All right. And I will stay on as long as there's questions.
[55:09] there's questions. Um
[55:10] Um or maybe at least for another 30 minutes
[55:12] or maybe at least for another 30 minutes if we want. Um okay. So we have some
[55:14] if we want. Um okay. So we have some questions uh that I'll kind of go
[55:16] questions uh that I'll kind of go through. Um so let's see what we got.
[55:20] through. Um so let's see what we got. Yeah, so there was a question around do
[55:22] Yeah, so there was a question around do HR practitioners feeling confident
[55:25] HR practitioners feeling confident Actually, let's see if I can show this.
[55:26] Actually, let's see if I can show this. This is my first StreamYard
[55:28] This is my first StreamYard webinar, but hopefully it's working
[55:29] webinar, but hopefully it's working well. So here's the question. Um are HR
[55:31] well. So here's the question. Um are HR practitioners feeling confident in these
[55:33] practitioners feeling confident in these weighted scores? I feel like I'm having
[55:34] weighted scores? I feel like I'm having trouble reconciling that against AI
[55:36] trouble reconciling that against AI hallucinations. Um and so this was a
[55:38] hallucinations. Um and so this was a question around the attrition risk
[55:40] question around the attrition risk demo that I showed where generated those
[55:42] demo that I showed where generated those scores. Now, the interesting thing like
[55:46] scores. Now, the interesting thing like the way to think about this is not like
[55:48] the way to think about this is not like I don't think AI hallucinations is the
[55:49] I don't think AI hallucinations is the right way to think about this. You saw
[55:51] right way to think about this. You saw the code that it wrote. Um and you
[55:53] the code that it wrote. Um and you should audit what it's doing, but it's
[55:55] should audit what it's doing, but it's basically doing the same thing that a
[55:57] basically doing the same thing that a data analyst would do, right? And this
[55:58] data analyst would do, right? And this is where I'm using Cowork, so I know it
[56:00] is where I'm using Cowork, so I know it has access to my computer, I know it's
[56:02] has access to my computer, I know it's writing code, so I see a little bit even
[56:03] writing code, so I see a little bit even if I don't fully understand it, I see a
[56:06] if I don't fully understand it, I see a little bit about what's going on under
[56:07] little bit about what's going on under the hood. And for that reason, I'm
[56:09] the hood. And for that reason, I'm confident that this isn't an AI
[56:11] confident that this isn't an AI hallucination problem. Now, it's still
[56:13] hallucination problem. Now, it's still not might be a good score, right? If you
[56:14] not might be a good score, right? If you ask the person
[56:16] ask the person try to predict attrition, they might not
[56:17] try to predict attrition, they might not get it right either, right? But it's not
[56:18] get it right either, right? But it's not an issue around AI hallucination, it's
[56:21] an issue around AI hallucination, it's more about, hey, how did it weight the
[56:22] more about, hey, how did it weight the scores? Could it have done something
[56:23] scores? Could it have done something differently? And I'm actually going to
[56:25] differently? And I'm actually going to share my screen again.
[56:28] share my screen again. See if we can get it back up.
[56:31] See if we can get it back up. Um I'm going to jump back into Claude.
[56:39] Did I do that right, Nicole?
[56:43] Yep, we can see your screen. Okay. Um so
[56:45] Yep, we can see your screen. Okay. Um so if I go back to what I showed there,
[56:49] if I go back to what I showed there, so if I scroll back in the chat,
[56:53] so if I scroll back in the chat, so it created this. When it created this
[56:56] so it created this. When it created this column, you can see how it weighted the
[56:57] column, you can see how it weighted the columns, right? So it's not doing
[56:59] columns, right? So it's not doing anything magical here. It is simply
[57:01] anything magical here. It is simply weighting these different things with
[57:02] weighting these different things with different It said 30% for would
[57:04] different It said 30% for would recommend company. Um and these seem
[57:06] recommend company. Um and these seem reasonable. If I want to change any of
[57:07] reasonable. If I want to change any of them, I can change it and have it rerun
[57:09] them, I can change it and have it rerun the analysis. Um I can also look,
[57:13] the analysis. Um I can also look, and this is where Claude Cowork is super
[57:14] and this is where Claude Cowork is super nice, um is if I go into here,
[57:18] nice, um is if I go into here, you can actually see this is the script
[57:19] you can actually see this is the script it wrote, right? Now, you might not
[57:21] it wrote, right? Now, you might not understand all of this, but this gives
[57:23] understand all of this, but this gives you confidence that you understand at
[57:25] you confidence that you understand at least a little bit about what's going on
[57:26] least a little bit about what's going on and you can always ask questions back to
[57:28] and you can always ask questions back to the AI. So I think with something like
[57:30] the AI. So I think with something like this, it's not a a risk of AI
[57:32] this, it's not a a risk of AI hallucination, it's a risk of you don't
[57:34] hallucination, it's a risk of you don't have the right data to make this
[57:35] have the right data to make this prediction, which is the same risk you'd
[57:37] prediction, which is the same risk you'd have otherwise.
[57:39] have otherwise. All right. Um I'm going to throw another
[57:41] All right. Um I'm going to throw another question.
[57:43] question. This one is about security.
[57:45] This one is about security. Um let me just pull this back up. Can we
[57:48] Um let me just pull this back up. Can we go back to security? How do you know all
[57:49] go back to security? How do you know all the data isn't just public information?
[57:51] the data isn't just public information? Keeping things confiden- confidential is
[57:53] Keeping things confiden- confidential is what we have to do in protect employees
[57:54] what we have to do in protect employees in the company.
[57:55] in the company. Um
[57:57] Um so I think there's a lot of with like
[57:59] so I think there's a lot of with like fear, uncertainty, and doubt about AI
[58:01] fear, uncertainty, and doubt about AI and where the AI is being used. Um
[58:04] and where the AI is being used. Um I think in practice, if you're using
[58:07] I think in practice, if you're using like Claude or ChatGPT or most of these
[58:10] like Claude or ChatGPT or most of these vendors and you understand things like
[58:11] vendors and you understand things like zero day retention, um you're
[58:13] zero day retention, um you're authenticated, none of these companies
[58:15] authenticated, none of these companies are in would It doesn't make any sense
[58:17] are in would It doesn't make any sense for them to try to like see your logs
[58:19] for them to try to like see your logs and understand what's going on. They
[58:21] and understand what's going on. They have enormous PR issues uh and like
[58:24] have enormous PR issues uh and like lawsuits and everything like that. I At
[58:26] lawsuits and everything like that. I At least the way I run my company is I'm
[58:27] least the way I run my company is I'm pretty comfortable
[58:29] pretty comfortable having our
[58:31] having our uh employees use the approved tools to
[58:33] uh employees use the approved tools to do their job, right? The same way I
[58:35] do their job, right? The same way I actually think there's a bigger risk of
[58:36] actually think there's a bigger risk of an employee leaking information than any
[58:39] an employee leaking information than any of these AI providers. Now, you need to
[58:40] of these AI providers. Now, you need to make your own decisions for your own
[58:41] make your own decisions for your own company, but I will say if you try to
[58:44] company, but I will say if you try to really crack down
[58:46] really crack down on the legal side or try to really
[58:48] on the legal side or try to really restrict what people can say and what
[58:49] restrict what people can say and what they can't say, you're just going to be
[58:51] they can't say, you're just going to be making AI pretty ineffective. Um and if
[58:54] making AI pretty ineffective. Um and if you want to get the power out of it,
[58:55] you want to get the power out of it, then my the process I would have is it's
[58:57] then my the process I would have is it's a clear set of approved tools. Don't let
[58:59] a clear set of approved tools. Don't let them use their personal ChatGPT account.
[59:01] them use their personal ChatGPT account. Have a clear set of approved tools and
[59:03] Have a clear set of approved tools and with those tools, work through the uh do
[59:06] with those tools, work through the uh do the work up front to make sure you feel
[59:07] the work up front to make sure you feel comfortable, and then let people use
[59:09] comfortable, and then let people use those tools for real. Um otherwise,
[59:12] those tools for real. Um otherwise, you're just not going to be able to get
[59:13] you're just not going to be able to get the same effectiveness uh you'd
[59:14] the same effectiveness uh you'd otherwise get.
[59:18] All right. Uh we had a question about
[59:20] All right. Uh we had a question about what's an API versus an MCP. Um the
[59:24] what's an API versus an MCP. Um the short answer I have is you probably
[59:25] short answer I have is you probably don't really need to worry about this.
[59:28] don't really need to worry about this. Um
[59:29] Um this is like a pretty in the weeds, um
[59:31] this is like a pretty in the weeds, um but basically MCPs talk to APIs
[59:34] but basically MCPs talk to APIs and then APIs are the thing that talk to
[59:36] and then APIs are the thing that talk to your
[59:38] your your system. Ultimately, they're pretty
[59:40] your system. Ultimately, they're pretty much the same thing and connectors are
[59:41] much the same thing and connectors are pretty much the same thing unless you're
[59:43] pretty much the same thing unless you're building one yourself. This
[59:44] building one yourself. This determination doesn't matter that much
[59:46] determination doesn't matter that much and most products have MCPs that you can
[59:49] and most products have MCPs that you can now interact with and now things like
[59:51] now interact with and now things like Claude are making it pretty easy to
[59:52] Claude are making it pretty easy to interact with those different things.
[59:56] All right.
[59:58] All right. Let's see what else we got.
[01:00:03] Here's a question. Um, I'm curious about
[01:00:05] Here's a question. Um, I'm curious about maintaining anonymity for employees when
[01:00:09] maintaining anonymity for employees when Vibe coding survey or engagement tools.
[01:00:12] Vibe coding survey or engagement tools. Uh, this is the kind of thing that I
[01:00:13] Uh, this is the kind of thing that I actually would not trust Vibe coding to.
[01:00:16] actually would not trust Vibe coding to. I don't use we don't we don't have our
[01:00:18] I don't use we don't we don't have our own Vibe coding tools here. This is the
[01:00:19] own Vibe coding tools here. This is the kind of thing that's pretty hard to get
[01:00:22] kind of thing that's pretty hard to get right, especially if you're the one
[01:00:24] right, especially if you're the one building the tool. It's very hard to be
[01:00:25] building the tool. It's very hard to be like fully anonymous.
[01:00:27] like fully anonymous. We use Windmill like obviously
[01:00:28] We use Windmill like obviously internally to do pulse surveys and that
[01:00:30] internally to do pulse surveys and that has
[01:00:31] has anonymous built into it. This is the
[01:00:33] anonymous built into it. This is the kind of thing that I would probably be a
[01:00:34] kind of thing that I would probably be a little bit more careful about trying to
[01:00:36] little bit more careful about trying to Vibe code some of that stuff. So like
[01:00:38] Vibe code some of that stuff. So like that data analysis thing, especially
[01:00:39] that data analysis thing, especially when it's working entirely on my
[01:00:40] when it's working entirely on my computer, I feel very confident cuz I'm
[01:00:42] computer, I feel very confident cuz I'm just building a tool for myself. If
[01:00:44] just building a tool for myself. If you're building a tool for your entire
[01:00:46] you're building a tool for your entire company with authentication and
[01:00:48] company with authentication and permissions and roles and let's say you
[01:00:49] permissions and roles and let's say you want it to be anonymous to certain
[01:00:51] want it to be anonymous to certain people and not to others,
[01:00:52] people and not to others, that's where I probably recommend using
[01:00:54] that's where I probably recommend using a more off-the-shelf tool and then
[01:00:56] a more off-the-shelf tool and then hooking up your agent into that tool,
[01:00:58] hooking up your agent into that tool, right? So it's not all or nothing like
[01:01:00] right? So it's not all or nothing like external tool versus internal. What we
[01:01:02] external tool versus internal. What we do mostly here internally like we use
[01:01:04] do mostly here internally like we use Notion a ton, we use Adios as CRM, it's
[01:01:06] Notion a ton, we use Adios as CRM, it's great. We obviously use Windmill, our
[01:01:08] great. We obviously use Windmill, our own product.
[01:01:09] own product. And I love all those tools cuz they
[01:01:11] And I love all those tools cuz they allow me to hook up my own agents to
[01:01:12] allow me to hook up my own agents to talk to them. Um, and so it's not all or
[01:01:15] talk to them. Um, and so it's not all or nothing.
[01:01:16] nothing. I think finding tools that are open that
[01:01:17] I think finding tools that are open that you can connect to is a really good
[01:01:19] you can connect to is a really good workflow, but allowing using kind of the
[01:01:21] workflow, but allowing using kind of the right software for the right job.
[01:01:25] All right, let's see what else we got.
[01:01:29] Okay.
[01:01:30] Okay. So here is a question about the
[01:01:31] So here is a question about the dashboard.
[01:01:33] dashboard. Um,
[01:01:34] Um, for the dashboard, how can you export
[01:01:36] for the dashboard, how can you export that so you can share it out and someone
[01:01:38] that so you can share it out and someone else can drill down to each department?
[01:01:39] else can drill down to each department? So
[01:01:40] So what I actually created there was just
[01:01:42] what I actually created there was just an HTML file. I could send that and we
[01:01:45] an HTML file. I could send that and we can we'll send it out after this. Um, we
[01:01:48] can we'll send it out after this. Um, we can send you the HTML file. Anyone can
[01:01:50] can send you the HTML file. Anyone can open that and look at the data. All of
[01:01:52] open that and look at the data. All of the data is embedded inside of that HTML
[01:01:54] the data is embedded inside of that HTML file. Now, there's no authentication
[01:01:56] file. Now, there's no authentication around that. So if that HTML file got
[01:01:58] around that. So if that HTML file got sent to someone random, right? Think of
[01:02:00] sent to someone random, right? Think of it similar to like a Word document. Um,
[01:02:02] it similar to like a Word document. Um, if you want to take it a step further,
[01:02:04] if you want to take it a step further, you can host that little mini web app on
[01:02:06] you can host that little mini web app on a product like Vercel, which is what we
[01:02:08] a product like Vercel, which is what we do. And inside of Vercel, you can have
[01:02:11] do. And inside of Vercel, you can have authentication so that only your
[01:02:13] authentication so that only your approved employees can view that and
[01:02:14] approved employees can view that and they don't need to open the HTML, they
[01:02:15] they don't need to open the HTML, they just go to a URL. Um,
[01:02:18] just go to a URL. Um, but
[01:02:19] but uh that's a little bit of how to do
[01:02:20] uh that's a little bit of how to do that, but sharing HTML files is like a
[01:02:22] that, but sharing HTML files is like a pretty low friction way that I'm
[01:02:23] pretty low friction way that I'm starting to see a lot of people do.
[01:02:27] All right, another question.
[01:02:30] All right, another question. Yeah, so the first demo showed how
[01:02:32] Yeah, so the first demo showed how adding the vacation policy helps
[01:02:33] adding the vacation policy helps answering avoiding answering the same
[01:02:35] answering avoiding answering the same questions repeatedly. How can employees
[01:02:37] questions repeatedly. How can employees access this themselves so they can ask
[01:02:39] access this themselves so they can ask those questions and get automated
[01:02:40] those questions and get automated answers on their own? So the way to do
[01:02:42] answers on their own? So the way to do that, the last demo I showed you is
[01:02:44] that, the last demo I showed you is probably the best way to do that is you
[01:02:46] probably the best way to do that is you have everyone at the company, you put
[01:02:48] have everyone at the company, you put everything in Notion and then you have
[01:02:49] everything in Notion and then you have everyone at the company connect the
[01:02:50] everyone at the company connect the Notion integration to your Claude
[01:02:52] Notion integration to your Claude account. Um, and then I also had a skill
[01:02:54] account. Um, and then I also had a skill set that I distributed, which you could
[01:02:56] set that I distributed, which you could just send out that skill
[01:02:58] just send out that skill and then people could install that skill
[01:02:59] and then people could install that skill to improve kind of the quality. That
[01:03:01] to improve kind of the quality. That would be the way to do that. If you want
[01:03:02] would be the way to do that. If you want to take it a step further, I think I
[01:03:04] to take it a step further, I think I would build an actual Slack bot here.
[01:03:07] would build an actual Slack bot here. And you could have the Slack bot talk to
[01:03:09] And you could have the Slack bot talk to Notion. We've actually done this. We
[01:03:10] Notion. We've actually done this. We have an internal
[01:03:12] have an internal uh agent we call PIM, PIM, which does
[01:03:15] uh agent we call PIM, PIM, which does like revenue operations for us. We
[01:03:16] like revenue operations for us. We actually put out a public blog post
[01:03:18] actually put out a public blog post about it. Uh, it's a little bit more
[01:03:19] about it. Uh, it's a little bit more technical so I didn't go through it
[01:03:20] technical so I didn't go through it today, but that's been extremely
[01:03:23] today, but that's been extremely effective because
[01:03:24] effective because then anyone can see
[01:03:26] then anyone can see the questions that are being asked right
[01:03:28] the questions that are being asked right in Slack and then our agent PIM
[01:03:30] in Slack and then our agent PIM responds. So you could build a little HR
[01:03:31] responds. So you could build a little HR buddy that is hooked up to the knowledge
[01:03:33] buddy that is hooked up to the knowledge base. It's actually pretty easy. There's
[01:03:35] base. It's actually pretty easy. There's a lot of tools to do it. You could just
[01:03:36] a lot of tools to do it. You could just build it yourself.
[01:03:37] build it yourself. And that's a good way of making it so
[01:03:39] And that's a good way of making it so that no one even needs to install
[01:03:40] that no one even needs to install anything. Now, in that case, it's not in
[01:03:42] anything. Now, in that case, it's not in Claude, it's more like in Slack. Um, and
[01:03:45] Claude, it's more like in Slack. Um, and I in ideal world, you probably have
[01:03:46] I in ideal world, you probably have both, right? People should be able to
[01:03:47] both, right? People should be able to ask these questions where wherever they
[01:03:49] ask these questions where wherever they want.
[01:03:54] Uh, Kyle asked a question. What is the
[01:03:56] Uh, Kyle asked a question. What is the material difference between creating a
[01:03:58] material difference between creating a skill
[01:04:00] skill versus an agent who is scheduled to do
[01:04:01] versus an agent who is scheduled to do repetitive tasks?
[01:04:03] repetitive tasks? Um, when you create a skill, the person
[01:04:06] Um, when you create a skill, the person is still in charge of invoking it at the
[01:04:08] is still in charge of invoking it at the right time. So if I create a skill, I
[01:04:10] right time. So if I create a skill, I still have to go out and like choose
[01:04:12] still have to go out and like choose when I want to run that skill.
[01:04:14] when I want to run that skill. Um, if you have an agent that's
[01:04:16] Um, if you have an agent that's scheduled to do repetitive task,
[01:04:18] scheduled to do repetitive task, especially if you've set that up on an
[01:04:19] especially if you've set that up on an automation so it runs like every morning
[01:04:22] automation so it runs like every morning at 8:00 a.m., it's going to run
[01:04:23] at 8:00 a.m., it's going to run automatically without anyone actually
[01:04:25] automatically without anyone actually having to do anything. So that's one
[01:04:27] having to do anything. So that's one difference, but in general like skills
[01:04:30] difference, but in general like skills versus commands versus agents with
[01:04:31] versus commands versus agents with custom instructions, it's all kind of
[01:04:33] custom instructions, it's all kind of the same stuff. You're just putting text
[01:04:35] the same stuff. You're just putting text into the prompt. That's all they're
[01:04:37] into the prompt. That's all they're doing. It's not more complicated than
[01:04:39] doing. It's not more complicated than that.
[01:04:40] that. Um, and the I guess the key difference
[01:04:41] Um, and the I guess the key difference between like is it something that a user
[01:04:43] between like is it something that a user invokes or is it automated? If this
[01:04:45] invokes or is it automated? If this agent was something a user invoked, it's
[01:04:47] agent was something a user invoked, it's it's the same thing. Like a skill is
[01:04:48] it's the same thing. Like a skill is it's literally just a block of text.
[01:04:52] Um,
[01:04:54] Um, we have a question from Rachel. What are
[01:04:56] we have a question from Rachel. What are your recommendations for building a
[01:04:57] your recommendations for building a knowledge base? I find our problem is
[01:04:59] knowledge base? I find our problem is that we don't have reliable knowledge
[01:05:00] that we don't have reliable knowledge bases for Claude chat GPT to pull from.
[01:05:02] bases for Claude chat GPT to pull from. Um,
[01:05:03] Um, ultimately,
[01:05:04] ultimately, someone has to do this. AI can't fully
[01:05:07] someone has to do this. AI can't fully solve this problem. Um, but a few things
[01:05:10] solve this problem. Um, but a few things to think about here. One is to
[01:05:11] to think about here. One is to consolidate vendors when it comes to
[01:05:13] consolidate vendors when it comes to like knowledge bases,
[01:05:15] like knowledge bases, it's a little bit better to have less
[01:05:16] it's a little bit better to have less knowledge base vendors. I've talked to
[01:05:17] knowledge base vendors. I've talked to some companies that have like 10
[01:05:19] some companies that have like 10 different knowledge bases.
[01:05:20] different knowledge bases. If you can cut that down to one, two, or
[01:05:22] If you can cut that down to one, two, or three, that would be nice.
[01:05:23] three, that would be nice. Notion is a great option. I like them
[01:05:26] Notion is a great option. I like them personally for really lightweight
[01:05:28] personally for really lightweight knowledge base management. Um, and when
[01:05:31] knowledge base management. Um, and when you're trying to create the knowledge
[01:05:33] you're trying to create the knowledge base, one of the most effective things
[01:05:34] base, one of the most effective things you can do
[01:05:36] you can do is have an agent and like if you're
[01:05:38] is have an agent and like if you're using Slack Slack heavily and there's a
[01:05:40] using Slack Slack heavily and there's a question where people
[01:05:41] question where people a thread or a channel where everyone's
[01:05:43] a thread or a channel where everyone's asking questions, have Claude
[01:05:46] asking questions, have Claude um
[01:05:47] um go through that Slack channel, read all
[01:05:50] go through that Slack channel, read all of the logs, and then generate the
[01:05:51] of the logs, and then generate the knowledge base from that or find gaps in
[01:05:53] knowledge base from that or find gaps in the knowledge base. So it's really
[01:05:55] the knowledge base. So it's really helpful when you're trying to build this
[01:05:56] helpful when you're trying to build this knowledge base from from scratch or have
[01:05:58] knowledge base from from scratch or have migrated or make like a bunch of edits,
[01:06:00] migrated or make like a bunch of edits, use the agent to build the knowledge
[01:06:02] use the agent to build the knowledge base. And then you got to keep it up to
[01:06:03] base. And then you got to keep it up to date and ideally you have some sort of
[01:06:05] date and ideally you have some sort of workflow where someone asks a question
[01:06:07] workflow where someone asks a question that can't be answered, the knowledge
[01:06:08] that can't be answered, the knowledge base kind of gets automatically updated.
[01:06:11] base kind of gets automatically updated. All right. We have a question around
[01:06:14] All right. We have a question around co-worker is still in research preview.
[01:06:16] co-worker is still in research preview. There's no logs to audit the activities.
[01:06:19] There's no logs to audit the activities. How do you handle security at the
[01:06:20] How do you handle security at the enterprise level? I've kind of answered
[01:06:21] enterprise level? I've kind of answered this a few times, but ultimately that's
[01:06:23] this a few times, but ultimately that's a question for your IT team
[01:06:25] a question for your IT team or or your executive team. I do think a
[01:06:28] or or your executive team. I do think a lot of people are way too obsessed with
[01:06:31] lot of people are way too obsessed with thinking that someone like Anthropic is
[01:06:33] thinking that someone like Anthropic is going to leak all your data
[01:06:36] going to leak all your data or use all of your data in
[01:06:39] or use all of your data in a public way. Now, I totally understand
[01:06:40] a public way. Now, I totally understand it with different levels of um
[01:06:44] it with different levels of um uh enterprise levels, you might not be
[01:06:45] uh enterprise levels, you might not be comfortable with the current set. I
[01:06:47] comfortable with the current set. I definitely recommend getting on an
[01:06:49] definitely recommend getting on an enterprise plan for all these accounts
[01:06:51] enterprise plan for all these accounts and then you can talk to someone and
[01:06:53] and then you can talk to someone and understand what the details are.
[01:06:55] understand what the details are. Um,
[01:06:55] Um, and you security is obviously really
[01:06:57] and you security is obviously really critical, but I don't think the place to
[01:06:59] critical, but I don't think the place to focus on is like worrying about
[01:07:01] focus on is like worrying about Anthropic or OpenAI leaking your data. I
[01:07:04] Anthropic or OpenAI leaking your data. I think the place to be worried about a
[01:07:06] think the place to be worried about a little bit more actually is people doing
[01:07:08] little bit more actually is people doing rogue stuff in like personal accounts or
[01:07:10] rogue stuff in like personal accounts or hooking up random stuff that isn't
[01:07:11] hooking up random stuff that isn't approved. So I would make sure and and
[01:07:14] approved. So I would make sure and and if you want people to use the approved
[01:07:15] if you want people to use the approved tools, you need them let them use the
[01:07:16] tools, you need them let them use the tools.
[01:07:17] tools. And so that's where there is a balance.
[01:07:19] And so that's where there is a balance. I think you're in a best much better
[01:07:20] I think you're in a best much better spot to really focus on getting a really
[01:07:24] spot to really focus on getting a really good list of approved tools and letting
[01:07:26] good list of approved tools and letting people use them freely rather than
[01:07:28] people use them freely rather than trying to like make it impossible to use
[01:07:29] trying to like make it impossible to use anything and then everyone just jumps
[01:07:31] anything and then everyone just jumps to like their own personal accounts,
[01:07:32] to like their own personal accounts, which is the wild west, which is the
[01:07:34] which is the wild west, which is the worst worst case scenario you can be in.
[01:07:36] worst worst case scenario you can be in. So few ideas there. I don't know if
[01:07:38] So few ideas there. I don't know if that's a perfect answer, but that's how
[01:07:39] that's a perfect answer, but that's how I think about it.
[01:07:43] Uh, okay, we got that question answered.
[01:07:48] Here's another question.
[01:07:49] Here's another question. Um,
[01:07:50] Um, can other people including admins like
[01:07:52] can other people including admins like IT see your skills?
[01:07:54] IT see your skills? Uh, no. Skills are local. Other people
[01:07:57] Uh, no. Skills are local. Other people cannot see them. If you created a plugin
[01:08:00] cannot see them. If you created a plugin which is shared across the organization,
[01:08:02] which is shared across the organization, then yes, other people could see them,
[01:08:04] then yes, other people could see them, but by default, skills are local. And so
[01:08:06] but by default, skills are local. And so if you want to create a skill about
[01:08:08] if you want to create a skill about sensitive information, you do not need
[01:08:10] sensitive information, you do not need to be worried about at least in Claude,
[01:08:12] to be worried about at least in Claude, the way I use it, this is not something
[01:08:14] the way I use it, this is not something you need to be worried about. Now, there
[01:08:15] you need to be worried about. Now, there are obviously places how you can share
[01:08:17] are obviously places how you can share skills, but skills are private by
[01:08:20] skills, but skills are private by default.
[01:08:29] Yes, so how do you know which systems
[01:08:30] Yes, so how do you know which systems are
[01:08:31] are uh which system is better API to use
[01:08:33] uh which system is better API to use with Claude? I feel like that's a big
[01:08:35] with Claude? I feel like that's a big barrier for all the twos out there,
[01:08:36] barrier for all the twos out there, calling out the twos. Um, and how to
[01:08:38] calling out the twos. Um, and how to choose the right system to connect.
[01:08:41] choose the right system to connect. Um, this is honestly pretty difficult. I
[01:08:44] Um, this is honestly pretty difficult. I don't have a simple answer here other
[01:08:46] don't have a simple answer here other than you just try them and see which
[01:08:48] than you just try them and see which tools work better. That's where I
[01:08:50] tools work better. That's where I mentioned like Gamma versus Google
[01:08:51] mentioned like Gamma versus Google Slides. I think Google Slides has an
[01:08:53] Slides. I think Google Slides has an MCP. I just know it doesn't work well
[01:08:56] MCP. I just know it doesn't work well cuz I tried it a while ago. That said,
[01:08:58] cuz I tried it a while ago. That said, this stuff changes extremely quickly.
[01:09:01] this stuff changes extremely quickly. And so you need to be re-evaluating some
[01:09:03] And so you need to be re-evaluating some of like your choices around this kind of
[01:09:04] of like your choices around this kind of stuff
[01:09:05] stuff almost on like a weekly basis.
[01:09:07] almost on like a weekly basis. It's not realistic for
[01:09:10] It's not realistic for every employee to be redoing their tools
[01:09:11] every employee to be redoing their tools every single week. And this is where
[01:09:12] every single week. And this is where you're probably going to have in your
[01:09:13] you're probably going to have in your organization a few people who love AI
[01:09:15] organization a few people who love AI and love trying all this kind of stuff.
[01:09:17] and love trying all this kind of stuff. And so if you can set it up where those
[01:09:19] And so if you can set it up where those people who are obsessed about it and
[01:09:20] people who are obsessed about it and using it every day cuz they love it and
[01:09:22] using it every day cuz they love it and having them share best practices is
[01:09:25] having them share best practices is probably the best way to figure it out
[01:09:26] probably the best way to figure it out rather than trying to have every single
[01:09:28] rather than trying to have every single person audit different APIs.
[01:09:31] All right.
[01:09:39] Um, can we use the tool effectively if
[01:09:42] Um, can we use the tool effectively if we get the team subscription for HR or
[01:09:44] we get the team subscription for HR or will it be limited if the whole company
[01:09:46] will it be limited if the whole company is not on Claude?
[01:09:47] is not on Claude? Um,
[01:09:49] Um, you can definitely get a lot of value by
[01:09:51] you can definitely get a lot of value by just getting it for
[01:09:53] just getting it for your team.
[01:09:54] your team. Um, you'll lose The main thing you'll
[01:09:56] Um, you'll lose The main thing you'll lose is a lot of like the information
[01:09:58] lose is a lot of like the information sharing, right? If everyone's on
[01:10:00] sharing, right? If everyone's on diff- like Claude is complicated enough.
[01:10:01] diff- like Claude is complicated enough. You have like chat, co-work, and code,
[01:10:03] You have like chat, co-work, and code, and different things of different
[01:10:04] and different things of different capabilities. If everyone's on different
[01:10:06] capabilities. If everyone's on different fragmented tools across the
[01:10:08] fragmented tools across the organization,
[01:10:10] organization, um,
[01:10:11] um, information sharing becomes a lot
[01:10:12] information sharing becomes a lot harder. Um, and one nice thing at least
[01:10:14] harder. Um, and one nice thing at least at this moment in time about Claude is
[01:10:16] at this moment in time about Claude is that engineers like Claude code,
[01:10:19] that engineers like Claude code, um, and co-work is really good, and chat
[01:10:20] um, and co-work is really good, and chat is pretty easy. And so, seems like they
[01:10:22] is pretty easy. And so, seems like they have a pretty good offering for each of
[01:10:23] have a pretty good offering for each of those different types of of of
[01:10:25] those different types of of of employees. Um, but if you just get the
[01:10:28] employees. Um, but if you just get the team subscription HR, you'll get a ton
[01:10:29] team subscription HR, you'll get a ton of value out of it. You don't need like
[01:10:31] of value out of it. You don't need like I you can download that data set of all
[01:10:33] I you can download that data set of all those pulse survey responses
[01:10:35] those pulse survey responses uh from across the company, and you can
[01:10:36] uh from across the company, and you can load it with any data you want.
[01:10:48] Uh, question. What are your thoughts on
[01:10:49] Uh, question. What are your thoughts on HRIS vendor that has zero-day
[01:10:51] HRIS vendor that has zero-day uh policy or opt out of training models
[01:10:53] uh policy or opt out of training models using client data, but has the ability
[01:10:55] using client data, but has the ability to change the policy at any time, and a
[01:10:57] to change the policy at any time, and a company's only recourse is to turn off
[01:10:59] company's only recourse is to turn off AI features?
[01:11:00] AI features? Um, again, this is up to you what you
[01:11:02] Um, again, this is up to you what you want to decide on. Um, certainly you
[01:11:04] want to decide on. Um, certainly you could talk to the vendor and see if you
[01:11:06] could talk to the vendor and see if you can remove that ability to to change
[01:11:08] can remove that ability to to change these key policies if that's really
[01:11:09] these key policies if that's really important. Um, so that would probably be
[01:11:11] important. Um, so that would probably be my response there is if you feel
[01:11:14] my response there is if you feel strongly about this, then just talk to
[01:11:15] strongly about this, then just talk to the vendor and see if you can get that
[01:11:16] the vendor and see if you can get that part taken out.
[01:11:19] Um,
[01:11:20] Um, for Windmill, what's available for
[01:11:22] for Windmill, what's available for career development and growth? Uh, I
[01:11:24] career development and growth? Uh, I assume this is a question about our
[01:11:25] assume this is a question about our product. Um,
[01:11:28] product. Um, so I'll give a quick overview. In our
[01:11:29] so I'll give a quick overview. In our product, we offer pulse surveys and
[01:11:32] product, we offer pulse surveys and one-on-ones and performance reviews,
[01:11:35] one-on-ones and performance reviews, um, and career development and growth
[01:11:37] um, and career development and growth are kind of baked into all those
[01:11:38] are kind of baked into all those different pieces. So, in our
[01:11:39] different pieces. So, in our one-on-ones, uh, we allow you to uh
[01:11:42] one-on-ones, uh, we allow you to uh upload what we're going to be calling
[01:11:43] upload what we're going to be calling company knowledge, where you can upload
[01:11:45] company knowledge, where you can upload leveling docs and personal goals and
[01:11:46] leveling docs and personal goals and stuff like that. Um, so it can inform
[01:11:49] stuff like that. Um, so it can inform those one-on-ones, and then those
[01:11:51] those one-on-ones, and then those one-on-ones and all those documents
[01:11:52] one-on-ones and all those documents inform the performance review cycle
[01:11:53] inform the performance review cycle itself. So, if you want more information
[01:11:55] itself. So, if you want more information here, definitely just sign up for a
[01:11:57] here, definitely just sign up for a demo. We're happy to show you it even if
[01:11:58] demo. We're happy to show you it even if you just want to learn kind of how we
[01:12:00] you just want to learn kind of how we built what we built. Um, definitely
[01:12:01] built what we built. Um, definitely check out a demo and we can go into more
[01:12:03] check out a demo and we can go into more details there.
[01:12:05] details there. Let's see. Any other questions?
[01:12:10] Let's see. Any other questions? And I know we've been answering a lot of
[01:12:11] And I know we've been answering a lot of questions that came in earlier. There
[01:12:13] questions that came in earlier. There are some people still on the chat. Uh,
[01:12:16] are some people still on the chat. Uh, if anyone who's still here has
[01:12:17] if anyone who's still here has questions, feel free to drop them in as
[01:12:18] questions, feel free to drop them in as well.
[01:12:20] well. All right. Um, there's a question around
[01:12:23] All right. Um, there's a question around do you know any HRIS systems that can
[01:12:24] do you know any HRIS systems that can integrate with Claude? I'm
[01:12:27] integrate with Claude? I'm not sure at the moment. Um,
[01:12:29] not sure at the moment. Um, I know historically a lot of HRIS
[01:12:31] I know historically a lot of HRIS vendors are pretty restrictive with even
[01:12:34] vendors are pretty restrictive with even like giving out like employee data,
[01:12:35] like giving out like employee data, which to me is a little bit crazy coming
[01:12:38] which to me is a little bit crazy coming from I don't have um my previous role
[01:12:40] from I don't have um my previous role was outside of HR.
[01:12:42] was outside of HR. Um, HR definitely historically had a
[01:12:44] Um, HR definitely historically had a different philosophy when it comes to
[01:12:45] different philosophy when it comes to APIs. I do think this will change pretty
[01:12:48] APIs. I do think this will change pretty dramatically over the next few years.
[01:12:51] dramatically over the next few years. Uh, Workday will probably be the last
[01:12:52] Uh, Workday will probably be the last holdout. Um,
[01:12:54] holdout. Um, but uh I do think it's really critical
[01:12:56] but uh I do think it's really critical that you at least have an HRIS system
[01:12:57] that you at least have an HRIS system that has plans to open up the APIs. Um,
[01:13:00] that has plans to open up the APIs. Um, that it just I don't know how you're
[01:13:02] that it just I don't know how you're going to be able to use AI effectively
[01:13:03] going to be able to use AI effectively without that. So, I think that's coming,
[01:13:05] without that. So, I think that's coming, and if not, I would switch to a vendor
[01:13:06] and if not, I would switch to a vendor that does support that.
[01:13:10] All right.
[01:13:11] All right. >> [snorts]
[01:13:14] >> Curious if there's a prompt library to
[01:13:16] >> Curious if there's a prompt library to help optimize or build specific for HR
[01:13:18] help optimize or build specific for HR operations. Uh, there are tons of prompt
[01:13:21] operations. Uh, there are tons of prompt libraries. Everyone is
[01:13:23] libraries. Everyone is sharing skills everywhere. There is the
[01:13:25] sharing skills everywhere. There is the official
[01:13:26] official um Claude one. I'll jump back and show
[01:13:29] um Claude one. I'll jump back and show it again so people can see it.
[01:13:37] All right.
[01:13:40] All right. Okay. So, if I go here and I go to
[01:13:42] Okay. So, if I go here and I go to customize, everyone see my screen?
[01:13:45] customize, everyone see my screen? Um, this is this plugin, this HR plugin,
[01:13:48] Um, this is this plugin, this HR plugin, right? And so, here are the skills that
[01:13:50] right? And so, here are the skills that they have here.
[01:13:51] they have here. Um, I would caution you these are not
[01:13:54] Um, I would caution you these are not like incredible. I think
[01:13:56] like incredible. I think Claude made this by just having AI write
[01:13:58] Claude made this by just having AI write a bunch of skills,
[01:14:00] a bunch of skills, um, and you can just read them, right?
[01:14:01] um, and you can just read them, right? So, here's an example of one. Org
[01:14:03] So, here's an example of one. Org planning.
[01:14:05] planning. Help plan organizational structure.
[01:14:08] Help plan organizational structure. Planning dimensions, headcount,
[01:14:10] Planning dimensions, headcount, sequencing budget.
[01:14:12] sequencing budget. Span of control, they're saying five day
[01:14:13] Span of control, they're saying five day direct reports, management layers, IC to
[01:14:15] direct reports, management layers, IC to manager ratio, right?
[01:14:17] manager ratio, right? So, that's that's literally the entire
[01:14:19] So, that's that's literally the entire skill. It's like these this paragraph.
[01:14:20] skill. It's like these this paragraph. So, maybe it's a decent starting point,
[01:14:22] So, maybe it's a decent starting point, but these are not like some incredible
[01:14:25] but these are not like some incredible thing. These are decent starting points
[01:14:26] thing. These are decent starting points that you're going to want to customize
[01:14:29] that you're going to want to customize for your business. Um, and it's actually
[01:14:31] for your business. Um, and it's actually really easy to create these. I just tell
[01:14:33] really easy to create these. I just tell Claude you want to create a skill about
[01:14:34] Claude you want to create a skill about XYZ, have it interview you, it asks
[01:14:36] XYZ, have it interview you, it asks questions. Yeah, I can just show it.
[01:14:39] questions. Yeah, I can just show it. People are curious.
[01:14:40] People are curious. Yeah.
[01:14:43] Yeah. I want to create a skill that
[01:14:44] I want to create a skill that automatically generates,
[01:14:46] automatically generates, um, offer letters for new employees. Can
[01:14:48] um, offer letters for new employees. Can you ask me five clarifying questions and
[01:14:50] you ask me five clarifying questions and then generate the skill?
[01:14:54] Cool. Do that.
[01:14:55] Cool. Do that. That's literally what I would do. And
[01:14:56] That's literally what I would do. And now it's going to interview me, ask me
[01:14:58] now it's going to interview me, ask me what I care about, and then it will
[01:14:59] what I care about, and then it will generate this skill file.
[01:15:02] generate this skill file. Uh, I think that something like this is
[01:15:03] Uh, I think that something like this is going to be much more effective than
[01:15:05] going to be much more effective than using a built-in off-the-shelf thing
[01:15:07] using a built-in off-the-shelf thing just cuz every company's different.
[01:15:13] What else?
[01:15:17] All right. We can do this real quickly.
[01:15:19] All right. We can do this real quickly. Uh,
[01:15:20] Uh, let's just do like we just say do you
[01:15:22] let's just do like we just say do you want to pull from Notion or Windmill?
[01:15:23] want to pull from Notion or Windmill? Let's do manual input only. I want to be
[01:15:26] Let's do manual input only. I want to be a
[01:15:27] a Let's do a PDF.
[01:15:29] Let's do a PDF. Should the skill use an existing offer
[01:15:31] Should the skill use an existing offer or generate from scratch?
[01:15:33] or generate from scratch? Base salary.
[01:15:35] Base salary. And equity incentive bonus. Cool.
[01:15:41] It's asking me one more question.
[01:15:44] Warm and professional, sure.
[01:15:48] Warm and professional, sure. All right. So, do it.
[01:15:49] All right. So, do it. Like the best tech companies.
[01:15:54] Now it has the information it needs.
[01:15:57] Now it has the information it needs. Um, and then it's going to actually
[01:15:58] Um, and then it's going to actually create the skill.
[01:16:00] create the skill. It's going to make it like Stripe or
[01:16:01] It's going to make it like Stripe or Google, sure.
[01:16:03] Google, sure. Thinking. You can see now it's using the
[01:16:05] Thinking. You can see now it's using the skill creator skill.
[01:16:07] skill creator skill. Um, in order for that to work,
[01:16:09] Um, in order for that to work, you just need to enable it before,
[01:16:10] you just need to enable it before, right? So, these example skills,
[01:16:13] right? So, these example skills, um, that's where I've
[01:16:15] um, that's where I've um
[01:16:17] um added them. The web artifacts builder is
[01:16:19] added them. The web artifacts builder is what I showed before and the skill
[01:16:20] what I showed before and the skill creator. If I wanted to, here's an
[01:16:22] creator. If I wanted to, here's an internal comms one.
[01:16:24] internal comms one. All right. So, it's saying for internal
[01:16:25] All right. So, it's saying for internal comms, you can use the three P's,
[01:16:27] comms, you can use the three P's, progress, plans, problems, right? If you
[01:16:29] progress, plans, problems, right? If you like that, you can just turn it on, but
[01:16:31] like that, you can just turn it on, but then you can of course customize it,
[01:16:32] then you can of course customize it, which you're probably going to want to
[01:16:33] which you're probably going to want to do.
[01:16:34] do. So, we go back here,
[01:16:36] So, we go back here, and it's building the skill. It'll run
[01:16:38] and it's building the skill. It'll run for a second, but that will run.
[01:16:41] All right. Let's see what else we got.
[01:16:48] All right.
[01:16:50] All right. Um,
[01:16:55] Here's a question here. When using
[01:16:57] Here's a question here. When using Claude over longer back and forth, I've
[01:16:59] Claude over longer back and forth, I've noticed it can start going off on
[01:17:01] noticed it can start going off on tangents and produce a lot non-useful
[01:17:03] tangents and produce a lot non-useful text.
[01:17:05] text. So, the issue here that you're running
[01:17:06] So, the issue here that you're running into is,
[01:17:08] into is, um,
[01:17:10] um, an issue with the prompt window and
[01:17:11] an issue with the prompt window and context lengths, right? Um, so I'll
[01:17:14] context lengths, right? Um, so I'll actually dive in. So, when you have a
[01:17:16] actually dive in. So, when you have a long chat, that means the context window
[01:17:18] long chat, that means the context window is really full, and AI will start to
[01:17:20] is really full, and AI will start to degrade.
[01:17:21] degrade. And so, you got to be careful about
[01:17:23] And so, you got to be careful about this. If you're having a really
[01:17:24] this. If you're having a really long-running chat, um, this is where
[01:17:26] long-running chat, um, this is where definitely stuff will get
[01:17:28] definitely stuff will get unfocused and off-task. I would
[01:17:30] unfocused and off-task. I would recommend starting a new chat if you're
[01:17:32] recommend starting a new chat if you're over like 10 or 20 messages. Uh, and
[01:17:35] over like 10 or 20 messages. Uh, and this is all because
[01:17:40] windows. Um, and so if I do like Claude
[01:17:44] windows. Um, and so if I do like Claude context windows.
[01:17:46] context windows. Let's see. I can quickly
[01:17:51] This is a little technical, but I know
[01:17:52] This is a little technical, but I know people are curious.
[01:17:54] people are curious. Let me show this.
[01:17:57] Let me show this. Uh, here's Claude's docs about context
[01:17:59] Uh, here's Claude's docs about context windows.
[01:18:00] windows. Um, and you'll actually see, where does
[01:18:02] Um, and you'll actually see, where does it say it?
[01:18:05] Um,
[01:18:08] Um, model comparison.
[01:18:10] model comparison. So, this is the context window. Where is
[01:18:11] So, this is the context window. Where is it? Context window. So, 1 million
[01:18:14] it? Context window. So, 1 million tokens, 1 million tokens, 200,000
[01:18:15] tokens, 1 million tokens, 200,000 tokens.
[01:18:17] tokens. So, what this means is if you use Sonnet
[01:18:19] So, what this means is if you use Sonnet and and Opus, it's going to degrade a
[01:18:21] and and Opus, it's going to degrade a lot less slowly than if you use Haiku.
[01:18:23] lot less slowly than if you use Haiku. So, this is why I was recommending
[01:18:25] So, this is why I was recommending Sonnet. 1 million tokens, this is up
[01:18:26] Sonnet. 1 million tokens, this is up from like it used to be like 40,000
[01:18:28] from like it used to be like 40,000 tokens. So, if you had problems in the
[01:18:30] tokens. So, if you had problems in the past, definitely things are a lot better
[01:18:32] past, definitely things are a lot better now. But even with this, at some point
[01:18:34] now. But even with this, at some point you'll run out of it you you'll run out
[01:18:35] you'll run out of it you you'll run out of space, and you're going to want to
[01:18:36] of space, and you're going to want to start a new chat. So, I would definitely
[01:18:37] start a new chat. So, I would definitely just recommend starting a new chat
[01:18:39] just recommend starting a new chat anytime you want to do something new.
[01:18:41] anytime you want to do something new. And if you find yourself reusing old
[01:18:42] And if you find yourself reusing old chats cuz you want to reuse some prompt,
[01:18:45] chats cuz you want to reuse some prompt, that's where a skill could generally be
[01:18:46] that's where a skill could generally be very helpful.
[01:18:50] All right. See what else we got.
[01:18:56] Did I answer this question? Um,
[01:18:59] Did I answer this question? Um, co-work is still in research preview,
[01:19:00] co-work is still in research preview, and there's no logs to Yeah, I think I
[01:19:02] and there's no logs to Yeah, I think I answered this question. Um, this is
[01:19:03] answered this question. Um, this is ultimately up to you for your own call,
[01:19:05] ultimately up to you for your own call, but I would definitely recommend
[01:19:07] but I would definitely recommend getting approved list of tools and
[01:19:09] getting approved list of tools and letting people use them freely rather
[01:19:10] letting people use them freely rather than trying or else people are just
[01:19:12] than trying or else people are just going to use their own like personal
[01:19:13] going to use their own like personal tools.
[01:19:14] tools. Um,
[01:19:18] All right. Let's see.
[01:19:20] All right. Let's see. Let's see.
[01:19:27] There's a question. If you can't
[01:19:28] There's a question. If you can't integrate directly, you could generate a
[01:19:30] integrate directly, you could generate a token to work through the API. Yeah, you
[01:19:32] token to work through the API. Yeah, you can start to do some of this stuff if
[01:19:33] can start to do some of this stuff if you know what you're doing,
[01:19:35] you know what you're doing, um, where you can start to use APIs,
[01:19:36] um, where you can start to use APIs, but, um,
[01:19:38] but, um, this will get a lot better. And
[01:19:40] this will get a lot better. And definitely if you're using some tool
[01:19:41] definitely if you're using some tool that doesn't support it, just ask them.
[01:19:45] Um,
[01:19:46] Um, here we go.
[01:19:47] here we go. If you, let's see this question.
[01:19:54] For those of you who already have Claude
[01:19:57] For those of you who already have Claude adopted across the HR team, how did you
[01:19:58] adopted across the HR team, how did you get started? Did you run any workshops,
[01:20:00] get started? Did you run any workshops, pair with some technical, or learn how
[01:20:02] pair with some technical, or learn how I'm doing? I don't know if anyone who's
[01:20:03] I'm doing? I don't know if anyone who's still here wants to answer Jade, but my
[01:20:05] still here wants to answer Jade, but my answer is all the above. Um, there's not
[01:20:07] answer is all the above. Um, there's not going to be one-size-fits-all. Uh, we
[01:20:09] going to be one-size-fits-all. Uh, we did a one lunch and learn where we had
[01:20:11] did a one lunch and learn where we had one of our engineers literally sit with
[01:20:13] one of our engineers literally sit with everyone side by side at their computer
[01:20:15] everyone side by side at their computer and like literally download the app and
[01:20:17] and like literally download the app and like install it for them to get them
[01:20:18] like install it for them to get them going. I do find that one misconception
[01:20:22] going. I do find that one misconception with AI is it's really easy to use. It's
[01:20:24] with AI is it's really easy to use. It's not, it's complicated. It's like any
[01:20:25] not, it's complicated. It's like any other skill. It takes time to learn. And
[01:20:28] other skill. It takes time to learn. And so you need to spend the time early on
[01:20:30] so you need to spend the time early on teaching people how to do things and
[01:20:32] teaching people how to do things and teaching them the ropes. And then once
[01:20:34] teaching them the ropes. And then once they get some level of expertise, the
[01:20:37] they get some level of expertise, the rest becomes a lot easier cuz then they
[01:20:38] rest becomes a lot easier cuz then they know how to use AI to make themselves
[01:20:40] know how to use AI to make themselves better. Um, but the initial cold start
[01:20:42] better. Um, but the initial cold start is hard. Um,
[01:20:44] is hard. Um, it's not easy to use AI. This Cohere
[01:20:45] it's not easy to use AI. This Cohere stuff's confusing. I'm talking about
[01:20:47] stuff's confusing. I'm talking about context windows and skills and MCPs and
[01:20:49] context windows and skills and MCPs and APIs.
[01:20:50] APIs. It is not a simple thing and you should
[01:20:52] It is not a simple thing and you should not expect people, especially
[01:20:53] not expect people, especially non-technical people, to get it in like
[01:20:57] non-technical people, to get it in like an hour, right? It's going to take days
[01:20:58] an hour, right? It's going to take days and days, but once they get it, then
[01:21:00] and days, but once they get it, then they become really self-sufficient.
[01:21:02] they become really self-sufficient. It's, and so getting people just over
[01:21:04] It's, and so getting people just over the hurdle and maybe means sitting down
[01:21:06] the hurdle and maybe means sitting down with them and setting up their laptop
[01:21:08] with them and setting up their laptop and showing them a few example
[01:21:09] and showing them a few example workflows. Um, we did that a few months
[01:21:11] workflows. Um, we did that a few months ago and now everyone at Windmill is like
[01:21:13] ago and now everyone at Windmill is like doing incredible things on a daily
[01:21:15] doing incredible things on a daily basis.
[01:21:16] basis. Um, and so it's just getting them over
[01:21:17] Um, and so it's just getting them over that early, um,
[01:21:19] that early, um, early hurdle.
[01:21:21] early hurdle. Yeah, hackathons are awesome. Uh, I
[01:21:22] Yeah, hackathons are awesome. Uh, I would recommend that. We do we do a lot
[01:21:24] would recommend that. We do we do a lot of hackathons, really fun way. Uh, and
[01:21:26] of hackathons, really fun way. Uh, and historically hackathons really only made
[01:21:27] historically hackathons really only made sense for engineers. Now they're great
[01:21:29] sense for engineers. Now they're great for anyone. Uh, cuz people can build
[01:21:31] for anyone. Uh, cuz people can build cool things. Like that dashboard. Like a
[01:21:32] cool things. Like that dashboard. Like a good hackathon project is give everyone
[01:21:34] good hackathon project is give everyone a bunch of data,
[01:21:36] a bunch of data, um, and maybe you anonymize it, right?
[01:21:37] um, and maybe you anonymize it, right? Or or something that's not sensitive and
[01:21:39] Or or something that's not sensitive and you let everyone try to build understand
[01:21:41] you let everyone try to build understand the data or build a dashboard. That's a
[01:21:42] the data or build a dashboard. That's a really good way.
[01:21:44] really good way. Um, question here.
[01:21:46] Um, question here. Not sure if it was just answered on the
[01:21:47] Not sure if it was just answered on the call. Do you recommend any specific
[01:21:49] call. Do you recommend any specific models for simple chats? Uh, I would
[01:21:51] models for simple chats? Uh, I would just use Sonnet is my short answer.
[01:21:53] just use Sonnet is my short answer. Um, really safe option.
[01:21:55] Um, really safe option. Uh, if you're using Claude, uh, if I use
[01:21:57] Uh, if you're using Claude, uh, if I use Opus when I'm doing software
[01:21:59] Opus when I'm doing software engineering, outside of that Sonnet's
[01:22:00] engineering, outside of that Sonnet's totally good. I would not recommend
[01:22:02] totally good. I would not recommend Haiku. Um, it's another similar thing
[01:22:04] Haiku. Um, it's another similar thing where
[01:22:05] where similar to like security, I think some
[01:22:07] similar to like security, I think some companies that really try to clamp down
[01:22:09] companies that really try to clamp down on AI costs are shooting themselves in
[01:22:11] on AI costs are shooting themselves in the foot where they're trying to be
[01:22:13] the foot where they're trying to be like, "Hey, we can't spend a lot of
[01:22:14] like, "Hey, we can't spend a lot of money on AI, so you have to use the
[01:22:16] money on AI, so you have to use the shitty models." Um, I would not
[01:22:18] shitty models." Um, I would not recommend that. I think at this point in
[01:22:19] recommend that. I think at this point in time, you want to be doing everything
[01:22:21] time, you want to be doing everything you can to let people use AI. I think
[01:22:24] you can to let people use AI. I think the ROI is pretty apparent for most
[01:22:26] the ROI is pretty apparent for most roles and so letting them use whatever
[01:22:28] roles and so letting them use whatever model they want, um, I think makes a lot
[01:22:31] model they want, um, I think makes a lot of sense. But Sonnet, the default model
[01:22:32] of sense. But Sonnet, the default model in Claude is great. It can do,
[01:22:34] in Claude is great. It can do, like unless you're trying to solve like
[01:22:36] like unless you're trying to solve like PhD level math problems or doing like
[01:22:38] PhD level math problems or doing like technical software engineering, it's
[01:22:40] technical software engineering, it's really good and it can do all this kind
[01:22:41] really good and it can do all this kind of stuff.
[01:22:44] of stuff. Um, anything else, Nicole?
[01:22:47] Um, anything else, Nicole? Any other questions?
[01:22:52] No, I think you covered most of them.
[01:22:56] No, I think you covered most of them. Okay. Well, we still got 170 people
[01:22:57] Okay. Well, we still got 170 people here, which is awesome. But I think
[01:22:59] here, which is awesome. But I think we're going to have to end it. Um, if
[01:23:01] we're going to have to end it. Um, if there's no more questions, uh, I wanted
[01:23:02] there's no more questions, uh, I wanted to thank everyone for coming. Hopefully
[01:23:04] to thank everyone for coming. Hopefully this was helpful. Uh, we'll definitely
[01:23:06] this was helpful. Uh, we'll definitely do another one of these. I think
[01:23:07] do another one of these. I think probably a little bit more focused on
[01:23:08] probably a little bit more focused on managers,
[01:23:10] managers, um, and how they can use AI to run their
[01:23:11] um, and how they can use AI to run their teams.
[01:23:13] teams. Um, and if you have any questions,
[01:23:15] Um, and if you have any questions, definitely reach out to us. If you want
[01:23:16] definitely reach out to us. If you want a demo of Windmill, we're happy to show
[01:23:18] a demo of Windmill, we're happy to show you what we're building.
[01:23:20] you what we're building. Um, and yeah, thank you, everyone.

Claude for HR: Building AI Workflows for Performance Reviews (Webinar)

https://www.youtube.com/watch?v=cctFWGecvhg

Summary

TL;DR — This webinar demonstrates how to leverage Claude AI for HR workflows, focusing on practical applications like building a policy Q&A chatbot, setting up performance review cycles, and analyzing survey data. It highlights the differences between Claude's chat, co-work, and code interfaces, emphasizing the power of co-work for local file access and integration with external tools like Notion and Gamma.

Key points

• Claude's co-work and code interfaces offer enhanced capabilities over basic chat, including local file system access and the ability to build complex workflows.

• Skills and connectors (MCPs) are crucial for turning Claude into an agent that can interact with other applications, automating tasks and providing dynamic data access.

• Practical HR use cases demonstrated include creating an interactive policy chatbot using Notion, automating performance review cycle setup with tools like Windmill, and performing data analysis on survey results to generate insights and even create interactive dashboards.

• Security and privacy are paramount; using approved tools with zero-day retention and understanding data usage policies are essential for effective AI adoption.

• Driving AI adoption requires a multi-faceted approach, including measuring usage, integrating AI into existing workflows (like one-on-ones and performance reviews), and fostering a culture of information sharing.

• While AI can draft content and analyze data, critical decisions in HR (hiring, promotions, terminations) should remain under human control, with AI serving as an informed assistant.

Takeaway — By understanding Claude's different interfaces, leveraging skills and integrations, and focusing on practical, data-driven HR use cases, organizations can significantly enhance efficiency and decision-making, while maintaining crucial human oversight for critical functions.