Claude Code Advanced Masterclass in Under 81 Mins

1. 0:00 – 2:14

   Intro

   1. CVCarl Vellotti0:00

      I spend all day in Claude Code because it can do more and more parts of my workflow. It gets to the point where, as a product manager or anyone working with these tools, you never really have to even leave Claude Code.

   2. AGAakash Gupta0:10

      This will really save PMs a lot of time. My first episode with Carl Vellotti got 30,000 views. It's been my most popular episode of the last couple months. So we have Carl back for a masterclass in advanced Claude Code usage.

   3. CVCarl Vellotti0:22

      Opus 4.5 is basically AGI, where you can give it a task, and you can kind of just trust it to, like, walk away from your computer, and it will just keep going. Most of the work, of course, that a product manager does is gonna be in the document world. So, you know, Google Drive, set that up right away, Notion, Confluence. That's the first one I recommend.

   4. AGAakash Gupta0:37

      But you told me your workflows have totally changed. What have you discovered?

   5. CVCarl Vellotti0:41

      We're gonna dive into how you can connect Claude Code to MCP, so you can pull in and push out information from everywhere. Go through the UI and click all these buttons and, like, kind of make sure everything's all organized. Whereas now, with Claude, it can just literally create all those tickets for you.

   6. AGAakash Gupta0:53

      Wow.

   7. CVCarl Vellotti0:54

      But when you're really... as you kind of start to pay attention, just thinking about the whole system that you're building with Claude, rather than just having it do individual tasks. Now, let's see the cool thing. Let's see this in Google Drive. So it gave us the link here. And then one thing, whenever it pushes to Google Drive, it pushes the straight markdown. So kind of a, a pro move here is you can cut it and then paste from markdown, and it will be nice.

   8. AGAakash Gupta1:13

      Whoa.

   9. CVCarl Vellotti1:14

      Because I have workflows for all those things already built, I can do it just all straight in one shot in, like, one morning. So it, it took our PRD, and now it created 19 engineering tickets, and it also gave them priority, which would be based on what was in that PRD.

   10. AGAakash Gupta1:27

       Is there anything else people should be exploring when they're thinking about Claude Code for PM? If you get any value out of this podcast, do me a huge favor and follow on Spotify and Apple Podcasts and subscribe on YouTube. It helps the show tremendously. And if you become an annual subscriber to my newsletter, you get access to nine incredible AI products for an entire year. This is an over $3,000 value across tools like Mobbin, Linear, Descript, Magic Patterns, Reforge Build, Relay, DeepSky, Dovetail, and Arise for AI Evals. Most of these brands have never done a product package like this, so go take advantage at bundle.aakashg.com.

2. 2:14 – 9:35

   Why Claude Code Hit $1B ARR in 6 Months

   1. AGAakash Gupta2:14

      And now, into today's episode. Carl, welcome back to the podcast.

   2. CVCarl Vellotti2:23

      Hey, I'm super excited to be here. Our last episode was really kind of, for me too, like, a good intro into using Claude Code and really figuring out how to use it as a tool in its own right. And, uh, ever since then, it's, it's kind of amazing. Even, I think it's been about two months since we recorded that last episode. I've continued to use Claude Code, continued to explore, like, more of its more advanced features. And now I basically just live in Claude Code all day. I start my day in Claude Code. I can do literally almost anything in Claude Code. If I can't do it in Claude Code, I figure out how to make it possible in Claude Code, and they keep adding new features that just allow more and more powerful workflows. And we're at the point now where every time, you know, you start something from scratch, you have to figure out how to use Claude Code for the first time to do that. But then, when you go from there, you build this workflow that the next time you do that task, it's faster, and then you realize a couple other ways you can improve that workflow, and then the next time it's faster. And, uh, it's at the point where there's just certain, like, many things where it would've taken me, you know, maybe a whole week of work. Because I have workflows for all those things already built, I can do it just all straight in one shot in, like, one morning. So I'm really excited to show kind of these more advanced features today. We're gonna dive into how you can connect Claude Code to MCPs so you can pull in and push out information from everywhere, and then we're gonna look at some of the newest features. Skills are really exciting. Skills are a way for Claude Code to basically automatically know how to do really complicated things without you even having to really prompt it. And then we'll do some, like, really advanced stuff where we'll show how, if you have your PRDs and your, all your documents in a GitHub repo, how you can have Claude take actions in those repos automatically. So you can be, you can be out, you can be at the beach. You can realize, "Oh, I forgot this requirement," and then you can open up your GitHub app. You can make an issue @Claude, and then by the time you get back to your desk, it will already be done. So really excited. There's so much to explore today, and, uh, I think we're gonna really do some more really good work today.

   3. AGAakash Gupta4:11

      There is this crazy news about Claude Code, which I'm not even sure exactly how Anthropic is figuring out what of its revenue is Claude Code versus regular Claude. But they have said that Claude Code has reached a $1 billion ARR milestone. If it hit that on December 2nd, it's been out for six months. That makes it pretty much the fastest ever to 1 billion ARR in less than six months. What's led to the continued success of Claude Code in the face of competition? We saw Google release Antigravity. Obviously, OpenAI has Codex, which we've done an episode on recently. Why has Claude Code continued to do so well?

   4. CVCarl Vellotti4:47

      Yeah, I think what it really comes down to is, you know, you have kind of, like, the main frontier labs. We have OpenAI, we have Google, and we have Anthropic, and they're all taking their own bets on how to approach using their LLMs. So for OpenAI, through ChatGPT, I think you really see kind of, like, general usage. So for kind of, like, regular people who aren't, like, necessarily using it all the time, it's a very good, like, general tool. Like, it's good for search. It's good for images. It can, it can do some amount of coding that's pretty-- It's very good as well. So they're much more general. For Google, you see that they really have taken a lot of big bets on being the absolute king of multimodal, so they can ingest... I think it's still the only LLM that can, like, watch a video and understand what's in it. They have the best image generation now with Nano Banana Pro. But they haven't really gone in depth to, like, the super deep workflows around coding. And they're, they're trying with Antigravity, but they're definitely kind of didn't start there, and that's not where their big bets are. And with Anthropic, they've really kind of focused on, how can this be the best tool for coding and for work? And where I think, and what I think is so powerful about that is that these are the types of things where I kind of was talking about earlier. I spend all day in Claude Code because it can do more and more parts of my workflow. Like, I don't need a general tool, and I don't really need it to be multimodal. I need it to be really good at text and code, and that alone basically lets me keep doing- More and more things in the tool. So I think what they're getting is they're getting maybe fewer users, but they are, like, absolutely just power users. And as soon as you start using it, it almost becomes addictive to where you just keep doing more and more stuff in it. And so I think that's why they're just so successful is because they have a smaller part of the market, but it goes so deep, and it keeps getting deeper. And, um, we're really at the point where the rest of the industry is just kind of trying to catch up. I think if you look at Antigravity or you look at Codex, they're definitely, like, most of their features are features that already exist in Claude Code, and Claude Code's just way ahead and keeps shipping new things. And to see the extent where Anthropic came out with MCPs, they were the first kind of company to come up with a standard for how LLMs can connect to other services. They actually just donated that as part of, like, a, an agentic foundation or something where, like, anybody can use it, so they open-sourced it. Uh, I think just today, Skills, which was something that we're gonna cover today, and Anthropic kind of came up with first as a way for you to really give an LLM, like, cl- clear tasks that it can complete. Now, the rest of the world is adopting that same standard. So they're just way ahead of the, those advanced workflows to the point where competitors kind of just have to copy them to try to keep up.

   5. AGAakash Gupta7:03

      I can say from my personal usage, after our episode, I had to switch to the Claude 5X plan, $100 a month.

   6. CVCarl Vellotti7:10

      Mm-hmm.

   7. AGAakash Gupta7:10

      I was like, "Oh my God." So I went and canceled some other subscriptions that I had to make the budget. And then I started using Claude Code, and I hit the 5X limits quickly, [laughs] and I had to go up to the 20X plan. One of the things about Claude Code that makes it unique is it's basically limitless. Like, with Claude, when you go tell it to do something, you're like, "Hey, I wrote a..." So one of the things I asked it to do, actually, was I said, "Hey, I wrote, like, 90 newsletter articles this year."

   8. CVCarl Vellotti7:33

      Yes.

   9. AGAakash Gupta7:33

      "I want you to, like, pull all those down, and I specifically want you to analyze, like, how my own style has changed over the course of the year."

   10. CVCarl Vellotti7:39

       Mm-hmm.

   11. AGAakash Gupta7:40

       Normal Claude challenged at it. Claude Code, "All right. Let's do it. Downloading all the things." Even hit a rate limit error on [laughs] Substack, but eventually downloaded all the articles. I told it to put it in a place. It was able to put them. It was able to analyze them. And so I think you're right. It's this power user functionality which is insane, and that's what we're gonna teach you guys today.

   12. CVCarl Vellotti7:59

       Yeah. Especially, uh, another thing since we last, uh, had our last podcast, Opus 4.5 came out, which is what I'll use today for our demo. And a lot of people are saying... You know, I think it's always, this thr- this term gets thrown around all the time, but people are really starting to feel the AGI. People are saying Opus 4.5 is basically AGI, where you can give it a task, and you can kind of just trust it to, like, walk away from your computer, and it will just keep going forward in the task. And if it hits a block, it'll find a way forward. Of course, it's not perfect, and we'll talk about that some today. But it's, it's really impressively powerful now.

   13. AGAakash Gupta8:29

       So what's the difference between people who are using Claude Code and not being productive versus people who are actually getting productivity gains out of it? Because sometimes it can just seem like you're watching the AI think. It's taking a long time. Are, am I actually doing something the right way and saving myself time?

   14. CVCarl Vellotti8:44

       Yeah. I think the, the big difference between, like, kind of a Cl- you know, maybe Claude on the browser versus a beginner Claude Code user versus a advanced Claude Code user is that for an advanced Claude Code user, you're really paying attention to what are the workflows that you're doing, you know, multiple times. And then how can you define, you know, files so that you can go step by step and really define those workflows so that the next time you do that, that it can run through all of that in, like, one shot. Versus, I think when you first start using Claude Code, it's easy to say, "Okay, do this. Okay, do this. Okay, do this." But when you're really, as you kind of start to pay attention, you see, oh, this is something I'm doing over and over again. Let me define, like, rules for that so that the next time I can just point it at this file, and it will know exactly how to approach it. So I think that's really what it is, is it's just thinking about the whole system that you're building with Claude rather than just having it do individual tasks.

   15. AGAakash Gupta9:32

       All right. Shall we just get started? I'd love to see this in action.

3. 9:35 – 10:59

   Ad Start

   1. AGAakash Gupta9:35

      Today's episode is brought to you by Amplitude. Replays of mobile user engagement are critical to building better products and experiences, but many session replay tools don't capture the full picture. Some tools take screenshots every second, leading to choppy replays and high storage costs from enormous capture sizes. Others use wireframes, but key moments go missing, creating gaps in your understanding. Neither approach gives you a truly mobile experience. Amplitude does things differently. Their mobile replays capture the full experience, every tap, every scroll, and every gesture, with no lag and no performance hit. It's the most accurate way to understand mobile behavior. See the full story with Amplitude. Today's podcast is brought to you by Pendo, the leading software experience management platform. McKinsey found that 78% of companies are using gen AI, but just as many have reported no bottom-line improvements. So how do you know if your AI agents are actually working? Are they giving users the wrong answers, creating more work instead of less, improving retention, or hurting it? When your software data and AI data are disconnected, you can't answer these questions. But when you bring all your usage data together in one place, you can see what users do before, during, and after they use AI, showing you when agents work, how they help you grow, and when to prioritize on your roadmap. Pendo Agent Analytics is the only solution built to do this for product teams. Start measuring your AI's performance with Agent Analytics at pendo.io/aakash.

4. 10:59 – 11:04

   Ad End: Setting Up MCPs

   1. AGAakash Gupta10:59

      That's P-E-N-D-O.I-O/A-A-K-A-S-H.

5. 11:04 – 14:07

   Setting Up Linear MCP (Live Demo)

   1. AGAakash Gupta11:04

      So should we start with MCP servers?

   2. CVCarl Vellotti11:06

      Yeah. Let's go ahead and start with MCP. So I'm gonna go ahead and share my screen here. And then, so one thing to kind of call out is I definitely recommend, if you haven't seen the first video, this one will still be good, and you'll still see, like, a lot of good demos. And it's actually good 'cause we're not gonna cover too many features that we did in the first video. So w- in the first video, we, we've covered things like if you have a markdown file, how you can point Claude to it, agents that can run stuff in parallel. We're gonna mostly cover new stuff in this video. Everything that we cover stacks with that first video. It's helpful to have that context, but everything here will make sense even if you haven't done it. Uh, the only thing I'll call out is there are lots of different ways you can use Claude Code because it's kind of just, you know, it has the files on your computer, and there's different ways you can show those. Last time we focused, we were mostly in Cursor. I'm gonna be in Cursor this time. We're not using any Cursor specific features. We're just using it kind of, like, as our IDE. So I'll just call that out. So we're in, we're in Cursor here. And let's go ahead and get started. And then the other thing, um, I'm gonna start Claude. I pretty much always use Claude with the dangerously skip permissions turned on 'cause I just, I don't know. I haven't had any issues with it. So that's one other thing to call out. I'm not gonna have to approve any, like, actions or files or things that it wants to do because I have this turned on. So with that, let's go ahead and, and fire this up.

   3. AGAakash Gupta12:15

      Also known as YOLO mode

   4. CVCarl Vellotti12:17

      Yeah, YOLO mode. All right, cool. Okay, so let's go ahead and get started. So we're in Cursor here, and we'll get going. So the first thing that I wanna cover today is one question that people have is, okay, we're in Claude Code, and we're on our computer here, but if I'm, you know, I'm a product manager, I need to write PRDs, I need other people to see those. How do I actually get other people to be able to, like, see my work? How do I get stuff that's in Claude Code out into the world so that someone else can see it? And then if I have, like, stuff in Notion or I have stuff in Google Docs or I have stuff in Confluence, how do I bring that into Claude Code in, in a way that's not just copying and pasting it? So that's what we'll cover first, and the main answer is we're gonna use something called MCPs. So this stands for Model Context Protocol, and it's something that was invented by Anthropic, and it's basically just the standard for how LLMs connect with services. So e-ev- like, many companies now, they have built their own MCP servers that make it so that you can basically just say, "Hey, here's the MCP I want you to connect to." Connect to it, and then it will give you a ton of options kind of with that. So we'll go ahead, and we'll start, and I'll show you just how to set up an MCP 'cause that's a question that a lot of people have. We'll start with setting one up for Linear. Are you familiar with Linear, Aakash?

   5. AGAakash Gupta13:31

      For sure, yes. Linear is one of the presenting sponsors of this podcast, so that was an organic mention there. [chuckles]

   6. CVCarl Vellotti13:37

      Yeah. Cool. So Linear, if, if anyone isn't familiar, it is a, a really sort of very modern, I don't know, I don't know of the exactly approved way to say it, but it's basically a modern version of Jira. Um, it's a way for engineers to, to manage their tickets. It... And it's, it's really nice. The thing that I'm showing here you could do with Jira. I just don't have a personal Jira account, so I can show you with Linear, but all the, the concepts are the same here. So it's pretty easy. So I'm gonna just show you almost, like, in general, like, and this is kind of like a mindset that you should have if you're a Claude Code user, is you should think, "Okay, I want Claude Code to be able to connect

6. 14:07 – 21:25

   Essential MCP Stack for PMs

   1. CVCarl Vellotti14:07

      to this thing." You can... Pretty much you can ask Claude Code, like, "Hey, does Linear have an MCP server?" And then it will search for you. Um, or you can, of course, just do our old school Google search. And, um, Linear does have an MCP server. And then on these, on these pages, it will usually tell you how you can set it up. And so what I'm gonna do here is, this is the command, so it's just... You literally, we're just gonna copy this, and then we're gonna run it, and then this will add this MCP into our Claude Code. So gonna go ahead and copy here, and we'll go back to Claude. That was a, that was, like, not a command that you put into Claude itself. You just put that into your terminal itself. So I'm gonna go ahead and copy that. And then, okay, so what we see here is it said that it added it, so file modified. So basically, what it's saying is it just added it to the Claude JSON so that when we go back into Claude, it will have, now have access to it. So literally, it was just one command. Many companies, they have it set up this easily, so you can do it like this. And then when we come back into Claude, one thing to know is you usually have to restart Claude for it to sort of pick up whatever changes there were to its settings. And now we're gonna start Claude again.

   2. AGAakash Gupta15:12

      And how'd you do that really quickly? What were the keyboard permission or shortcuts there?

   3. CVCarl Vellotti15:15

      Oh, yes. In order to just, like... There's two ways to leave Claude, or there's a couple different ways. You can type exit, which will kill it. You can type... That's the main one. Or the, the way that you can kind of, like, cancel any command w- any time you're in a terminal is you can do Control + C. Um, so here it's saying do it again, and then it will kill it. So that's a way that you can always just, like, totally, if you wanna just stop whatever's happening in the terminal, just Control + C will just immediately kill it. Okay.

   4. AGAakash Gupta15:38

      I've been, like, creating new terminal windows, so gonna save me a little bit.

   5. CVCarl Vellotti15:42

      Yep. Okay, so now what we can do is we go to MCP, and we see... So normally, Linear, I actually already had connected before we did this. What I'm gonna do is I'm gonna go ahead and dis- or I'm gonna re- have to reauthenticate. And so what that does is basically it just pops open a browser where, of course, if you are... You can't just say, like, "Claude, connect to Linear." You're gonna... There's gonna be some sort of authentication process. But here with Linear, it's super easy. So I just approve. So now it's approved. We're in my workspace here, um, and we'll be using this kind of throughout the demo. So authentication successful. So when we go back to, um, Claude Code, now we say that we are, now it's, we see that we're connected to Linear. And so we can do something like create a Linear test issue. And so what's cool is now we have our, our Linear, you know, our actual real Linear board, and you can be connected to your Linear's team. And what's awesome is that it will just now, it'll be able to figure out, like, okay, great, this MCP, I already have it, and it knows what MCPs it has. And then it's searching for it, and it finds the tool, and now it's gonna go ahead and basically create it. What's cool here, and we'll let, we'll let that happen, and we'll come back to it, but I'll start another one while we wait, is when you're looking at your MCPs here, so right now I have Google, I have Linear. Well, there's also Linear Server, which is just 'cause I added it again, but we could delete that one, and then Reddit, which is a demo that we did last time. What's cool is that you can click on any of these, and what's really nice about, uh, an MCP is that it g- they basically come with lots of tools that Claude Code, like, will basically automatically understand you, understand how to use if you've turned it on. So what we can do here is we can go ahead, and we can look at Linear, and this is by connecting Claude Code to the Linear MCP. This is all of the tools that it gets. So there's just a ton here. You can get documents. You can get issues. You can create issues. You can get, like, a list of projects, and then you can get the information from those projects. So just right away, just by connecting to this MCP, we have 25 different tools that Linear has built to allow LLMs, whenever they connect to this MCP, to basically tell it how it can work with Linear.

   6. AGAakash Gupta17:37

      Yep, and this is basically the step one. Like, you need to do this with your task management tool, your product analytics tool. What are the other major things PMs should be hooking up via MCP here?

   7. CVCarl Vellotti17:46

      I mean, really, like I kind of said earlier, a, a good mindset is like, "Okay, I have to do this thing, and I wanna try to live in Claude Code all day. What are there MCPs that already exist for it?" So what we're gonna look at today is I have a Google Workspace MCP set up, so that's gonna be how we pull information out and push it into Google, g- anything with Google Drive, but it can also do Google Calendar, all that stuff. Slack has a good one. All the sort of, like, major document systems like Notion and Confluence, they all have their own MCPs. There's one for, uh, Google Search Analytics, so for some of my websites where I have, like, SEO going, you can connect it to that. So there's- Lots of MCPs, and because it's the, it's sort of the new standard now, most companies, if they want to-- an LLM to be able to interact with them, they're gonna have their own. So it's, it-- there's a lot. Um, those are some of the best ones I just mentioned. But in general, always just try to search to see if there is one that already exists.

   8. AGAakash Gupta18:34

      So I guess if I were just to guess, right, like as a PM, you need to really care about insight and data from inside your company, like Salesforce analytics, BI, product analytics, all those things. User insights, right? Like your Dovetail, your Interpret, whatever your support desk, like your Zendesk or your Intercom. And nowadays, all these products have an MCP, so you should really think about connecting your Claude Code to all of those, I think, if you wanna get the most out of it.

   9. CVCarl Vellotti19:01

      Yep, exactly. And, uh, one kind of... One thing that is worth keeping in mind as you connect your MCP to all these tools is, um, MCPs, they do, like, automatically take up some context window of the LLM as you use it. So it-- you could, you really could, and you probably should have just like, you know, 20 different MCPs set up to all the different tools that you use. It's good to kind of try to be smart around, like, when you have them turned on. So for example, right now I only have Linear and Google Workspace turned on. I'm gonna leave Reddit turned off because it will take up some of that context window. But if I was gonna do something with Reddit, then I would be able to enable it. And one thing that's cool is Claude Code knows what MCPs it has. So, you know, if I said to do something on Reddit, it wouldn't have access to those tools without this turned on, but it would knows that it has Reddit. So it could say, "Hey, would you enable that?" And then it will have access to those tools.

   10. AGAakash Gupta19:46

       So what's a quick win a PM could set up in like five minutes?

   11. CVCarl Vellotti19:49

       Yeah, I would say, you know, whatever the document system that you personally use or is used at your company, set that up first. 'Cause most of the work, of course, that a product manager does is gonna be in the document world. So, you know, Google Drive, set that up right away. Notion, Confluence. That's the first one I recommend 'cause that's just where immediately you're gonna be able to do work on your computer and push it out so that people can see it.

   12. AGAakash Gupta20:09

       Nice.

   13. CVCarl Vellotti20:09

       So one thing that's also worth covering about MCPs, and, and this is kind of like a natural question, and it's, it's really important kind of throughout this demo, is MCPs are, like, kind of the new way for an LLM to be able to interact with all these different tools. But there's kind of the more traditional way, and that is with an API. So for example, one thing that we're gonna do later in this demo is we're gonna use... Like, for example, any time that you, if you wanna build an application that, you know, has an LLM embedded, like you want it to be able to, you know, ask ChatGPT something, then you're gonna get an OpenAI key, and then you're gonna call the OpenAI API, and then that will get a response. So LLMs can also run commands through APIs and get information that way. The way that I would think about, like, when should you use an MCP versus an API. So for an MCP, as we saw, you get all the tools just automatically. An LLM, like right out of the box, it just knows how to use that MCP. Whereas with an API, you just, like, for example, if I said, "Hey, here's the OpenAI API key. Do something with it," it doesn't know what to do. So it needs to have, like, that context around how to actually use that API. And so today we're gonna do some stuff with MCP, and we're gonna do some stuff with APIs. And you'll see that those are kind of like the two ways that you can connect your LLM to the world, and they're good for, like, different use cases. I just kind of want to cover that 'cause that's the other main way you can have your LLM sort of get information from the world.

7. 21:25 – 28:00

   End-to-End PM Workflow Begins

   1. AGAakash Gupta21:25

      Okay. So I think we're gonna start with, like, a email summary demo?

   2. CVCarl Vellotti21:28

      Yes. Okay, so what we're gonna start here is we're gonna start our end-to-end demo. So we're gonna work with a fictional company called Taskflow. You can think of it as a to-do app, kind of like Todoist. And we're gonna work through a feature for a calendar integration. So you can imagine that we know that people want this. There's been, like, some customer support tickets. The CEO thinks it would be a cool feature, but we don't have any user research. So we're gonna start there. We're gonna draft a user research survey. We're gonna get some mock data. We're gonna have Claude analyze it. We're gonna create the PRD. We're gonna create a deck. We'll create some images for that deck. We will write tickets that go straight into Linear. We'll be able to create the PRD in a way that it pulls those tickets directly from Linear through Claude Code, and then it will automatically update that PRD as, you know, engineers start that work. So we'll do an end-to-end flow, just start at the very beginning, and we'll show how from Claude Code you can do all of it in here and just keep the work in here, push it out, and then when you have stuff to pull back in, pull it in as well.

   3. AGAakash Gupta22:22

      Epic.

   4. CVCarl Vellotti22:23

      So to start, we're just gonna go ahead and we'll just start at the very beginning. We'll create a user research survey for a calendar integration based on... Basically, we'll give it information around the feature in our feature context document, and then we'll say that we already have a template. So this is some stuff that we basically covered in the first video. Okay, so what it's doing is it's reading through those two, and now it's gonna go ahead and it'll create that document, and then it'll save it to Google Docs. One thing I know I'm gonna have to do is I, I'll tell it to use my email. Sometimes when it uses this MCP, it, it, like, forgets who it's for. But once I tell it for the first time, it, it won't have that problem again. So, um, MCP. And please also put the markdown file in here. Okay, so what it's gonna do is it's gonna read those two files. It will create a markdown file in here that has the survey, but then it will use the Google Docs MCP to actually push that into Google Drive. So it'll be pretty sweet by the end here 'cause it'll actually have it in Google Docs. And then from there, you can imagine, like, you probably wanna share that with your user research team or someone before you actually created the, like, sort of sent it out. So this is step one. We're just gonna start with some user research based on some documents that we have in here.

   5. AGAakash Gupta23:29

      Makes sense.

   6. CVCarl Vellotti23:29

      Yeah, so we kind of see it running the tool here. And now it's also saving it in here, so we can see this survey in here. And then one way you can always nicely view these, you right click and do preview.

   7. AGAakash Gupta23:41

      Yes. Showed this to another podcast guest who loved it.

   8. CVCarl Vellotti23:44

      Yeah, I saw that actually. There's also a shortcut if you do a Command + Shift + V, and then it'll open it as well. Um, okay, great. So it's basically... This looks pretty good. It followed our own template here. And then, so this is here, but now let's see the cool thing. Let's see this in Google Drive. So it gave us the link here. And then one thing, whenever it pushes to Google Drive, it pushes the straight markdown. So kind of a, a pro move here is you can, like, cut it and then paste from markdown, and it will be nice.

   9. AGAakash Gupta24:09

      Whoa!

   10. CVCarl Vellotti24:11

       Yeah

   11. AGAakash Gupta24:11

       Another staver

   12. CVCarl Vellotti24:11

       I'll, I'll do that again just to, like, show it. So we had a markdown file, or everything's in markdown here. But I think Google Docs knows that many more people are working in markdown files than ever before, so they've added a feature where you can paste from markdown. There's no hotkey for it, but you just right-click and then do paste from markdown, and it will take that and it will make it really nice. So boom, now we have our user research survey. So the first time we've kind of seen it, we've actually pushed something out into Google Docs.

   13. AGAakash Gupta24:38

       Nice.

   14. CVCarl Vellotti24:38

       Okay. So now let's say we send this out and we get responses from it. I have already created a, uh, sort of like a response. Like we, we got some survey responses from this, and then the next step is we're going to have Claude help us analyze that research. So first, I'm just gonna go ahead and ask it to pull in the responses from this survey. Okay, this is gonna use that same MPC again, and I'm asking it not to do any analysis. We just wanna pull the information in. One thing worth calling out, so there's sort of a step in between here where you'd actually want to, you know, draft that survey in a way that users could respond to it. So depending on the thing that your company was using, Qualtrics is like a very common user research platform. So then as a, as a PM who's trying to be an AI native PM, what you should think of is, "Oh, I wonder if Qualtrics has an MPC server." So let's go ahead and search, um, Qualtrics.

   15. AGAakash Gupta25:24

       And they must, right? So then you could basically automatically create the survey in MCP once you've gotten the feedback from your UXR team.

   16. CVCarl Vellotti25:31

       Yep, exactly. So there is in fact a Qualtrics MPC server. So in this demo I don't have Qualtrics, but that is kind of the, the mindset is like, okay, here's the next step. Could I have Claude Code do it? And it turns out for many, many, many things, the answer is almost always yes, which is so awesome. Okay, so it's pulling that in.

   17. AGAakash Gupta25:48

       And if you are a stats tool that hasn't developed an MCP server yet, you might wanna ask yourself why not.

   18. CVCarl Vellotti25:53

       Yeah, it's true. Okay, great. So it pulled that information in, and then this is... It's kind of just showing it here. Um, one thing, the, if you, if you ever have, like, these kind of ugly tables, it's usually just 'cause you need to give it enough room.

   19. AGAakash Gupta26:04

       Yeah. I face this a lot.

   20. CVCarl Vellotti26:05

       Yeah.

   21. AGAakash Gupta26:05

       And then you end up using a lot of Command + minus, right? To go small.

   22. CVCarl Vellotti26:09

       Yeah. [laughs] So-

   23. AGAakash Gupta26:10

       Yeah

   24. CVCarl Vellotti26:10

       ... there it is. Obviously I'll, I'll zoom back in, but that, that's like w- this is like a very long table. Okay, great. So now what we're gonna do, and this won't seem that amazing at first, but something cool is gonna happen here. So w- I'm gonna say synthesize the user research survey responses and put it into new.md. So notice that I'm not giving it any other instructions. This is kind of vague.

   25. AGAakash Gupta26:28

       Is it gonna put it in Google Docs too?

   26. CVCarl Vellotti26:30

       Even, even something even kind of cooler is gonna happen. Okay, so we're, we're waiting. It says it's accomplishing. It's thinking really hard.

   27. AGAakash Gupta26:37

       And if you guys are wondering why his Claude Code is so much faster than yours, he had that bypass permissions, which many of you guys probably don't have on.

   28. CVCarl Vellotti26:44

       It's not... Oh, here we go. Okay. Really thinking hard on this request.

   29. AGAakash Gupta26:48

       [laughs] Yeah. 35 seconds for 300 tokens.

   30. CVCarl Vellotti26:52

       It's kind of a funny element of working with LLM sometimes is, like, there's a lot of just, like, really short amount of downtime. Okay. So, okay, it didn't do what I was hoping it would do. Let me... We'll, we'll start this again. I think it might be because we already had the chat going.
8. 28:00 – 30:01

   Skills Introduction

   1. CVCarl Vellotti28:00

      has this very cool new feature called Skills, and the way that Skills work is I have a skill that is already uploaded into this project that is a research synthesizer skill. And the way that this is supposed to work is you have a name for it, and then you have a description of that skill. And so what this does is that as soon as it uses this skill, then it has like a pre-built sort of template for how it is supposed to analyze that information.

   2. AGAakash Gupta28:26

      Yeah.

   3. CVCarl Vellotti28:26

      That alone is something that we've kind of seen before. You could say, "Hey, Claude, analyze this information and use this file that has a template." The cool thing about Skills and what I'm hoping that we see next... Okay, so-

   4. AGAakash Gupta28:36

      It's supposed to invoke the skill whenever it's relevant, but I found it's not always reliable, so sometimes I'm just prompting, like, use X skill and then do Y. [laughs]

   5. CVCarl Vellotti28:45

      Yeah. L- We'll, that'll be our backup. We'll, we'll, we'll try this one more time and see if it figures it out. But what, what, the way it's supposed to work is we have this like very clear trigger. Triggers on a- on requests to analyze, synthesize, or find patterns in user research. And then the prompt... Oh, it might be... Well, okay. So it's putting it into a new markdown, and then one thing you can do it with Claude is you can, uh, send it a, a message, and then that'll basically queue it. So as soon as it has completed this, then it will do this. It's really nice if you kind of like know what the output is gonna be and you know the next thing you wanna do and then you, like, wanna go do something else, is you can queue up that message. Seems like, um, Claude in general is being a bit slow today. Maybe lots of people trying to get some work done on the end-

   6. AGAakash Gupta29:24

      And this might not be the most, uh, difficult query, so if you're facing that issue and you're saying, "I need to go faster," you can always switch to a Sonnet or even a Haiku.

   7. CVCarl Vellotti29:32

      Yeah, that's true. And, and the, the way you do that is you just type model, and then you could switch it to another one, but not while it's also going. Okay, so it created that. Now we can see all the responses from that file in Google Drive. And then now I asked it to sy- synthesize, so now it says, "Now I'll synthesize the survey responses in a new analysis document." We'll see if it, it finds the skill. If it doesn't find the skill, that's okay. I think there are some other examples from today where we'll see it use it more. In general, I think it's kind of an, an improving feature for

9. 30:01 – 30:53

   Ad Start

   1. CVCarl Vellotti30:01

      Claude.

   2. AGAakash Gupta30:01

      Today's episode is brought to you by Jira Product Discovery. If you're like most product managers, you're probably in Jira tracking tickets and managing the backlog. But what about everything that happens before delivery? Jira Product Discovery helps you move your discovery, prioritization, and even roadmapping work out of spreadsheets and into a purpose-built tool designed for product teams. Capture insights, prioritize what matters, and create roadmaps you can easily tailor for any audience. And because it's built to work with Jira, everything stays connected from idea to delivery. Used by product teams at Canva, Deliveroo, and even The Economist, check out why and try it for free today at atlassian.com/product-discovery. That's A-T-L-A-S-S-I-A-N.com/product-discovery.

10. 30:53 – 38:00

    Ad End: Skills Deep Dive

    1. AGAakash Gupta30:53

       Jira Product Discovery: build the right thing. Today's episode is brought to you by Miro. Let me ask you something. How many tools are you juggling just to get a single project across the finish line? One for brainstorming, another for planning, something else for tracking tickets. That's where Miro comes in. It becomes an all-in-one collaboration workspace. Whether you're consolidating user research from several interviews, developing and synthesizing product briefs or a wireframe, or project managing development, Miro brings everyone into the same space. It's fast, intuitive, and fully loaded with features like project templates, two-way Jira sync, and integration with software like Draw.io and PlantUML. Miro's AI features can be used to synthesize elements in a board to develop a ready-to-review product requirements document in seconds. If you're tired of tab overload and scattered workflows, try Miro. Head to miro.com and see why over 90 million users choose Miro to guide from idea to outcome. Skills themselves, though, are amazing, and I highly recommend people watching build skills. I've built skills for, like, everything I do, um, and when it doesn't invoke the skill, I get very angry because the response is much worse. [chuckles]

    2. CVCarl Vellotti32:04

       Yeah, what are, what are some of the things you use skills for?

    3. AGAakash Gupta32:07

       So personally, I have, uh, just even for this podcast, I have a podcast production guide skill, which is, like, my lifesaver. Like, a- after I create the production guide with my team, I put it into there and it always comes out much better. I also have a skill for editing my newsletters, for editing my Twitter posts, editing my LinkedIn posts. Both of those are pretty good. What else do I have skills for? I have a skill for creating the takeaways from a podcast using a transcript. I have a skill for coming up with new newsletter topics. So yeah, so many skills at this point.

    4. CVCarl Vellotti32:40

       Yeah. That's awesome. Yeah, so that, that's like a-- Those are all really good examples where you've, you've probably done those things a million times and now you're like, "Okay, I should just invest in letting Claude really give, get all the power out of that skill of doing this every time I do it." Okay.

    5. AGAakash Gupta32:52

       And it didn't take as long as you might think. Like, usually what I'll do is, like, let's say I wanna do, like, a editor, right? I'll give, like, two or three examples. Here's version A and version final three times. Notice, like, w- how I edited it, how I improved it. Take that down, and then you do it, and then you can just run vibe checks like two or three times. I'll just have it edit two or three posts. I'll edit the skill two or three times, and it's pretty much good to go after that.

    6. CVCarl Vellotti33:13

       Yeah. That's awesome. And it's cool 'cause this stacks too, like if you had agents, you could give the agents access to different skills, and then when they're working they'll, they'll have those skills. Okay, cool. So all right, so we have this, this synthesis. We'll, we'll go from here. Okay, so in general, the concept of skills is that they're supposed to trigger when you do different things. The question might be, okay, so we already had this skill in our kind of .Claude here. What, where do skills come from? So you can... This one I already had. There's a couple different places that skills can come from. So one is there are a lot of these, like, people share them all the time on Twitter, on LinkedIn, there are, like, uh, these big GitHub repos with tons of skills. So there are a lot of skills that kind of already exist. So let's say, like, this, this is a repo that just has tons of skills. And so what you can do is you can just kind of like look at one of these libraries, and you can choose one, so-

    7. AGAakash Gupta33:59

       I do wanna warn everybody, you guys are very susceptible to prompt injection hacks if you are downloading a skill and running and dangerously skip permissions, so be really, really careful about that.

    8. CVCarl Vellotti34:08

       Good. Yeah, good tip. I should also keep that in mind. I think, yeah, definitely be careful. Uh, maybe it'd be good to even, like, have Claude almost investigate your spill, skill. Maybe you could build your own skill that's a skill investigator whenever you add a new skill. Literally not a terrible idea. Um, but yeah, that's true, that some people will put stuff into these that can be, [chuckles] especially for people like me who are just using them kind of willy-nilly, who knows what the triggers are and what exactly they do. Okay, but there are just tons of different skills that people are building all the time. So here we see an artifact builder, an iOS simulator, an MCP builder. There was one in here that I thought looked good. We'll have, I think it's a document... Oh, internal comms. So that's something that, like, you know, as a product manager, you might want this. So we're gonna go ahead. We're just gonna literally copy and paste this GitHub link, and we'll say to a new session of Claude. Go ahead and clear this.

    9. AGAakash Gupta34:53

       Nice.

    10. CVCarl Vellotti34:54

        We'll say-

    11. AGAakash Gupta34:54

        Yeah, and you guys can also find them from places you trust. Like in my newsletter, I have a PRD writer, a feature results writer, resume customizer. You can find all of those, and there's no prompt injection tech there.

    12. CVCarl Vellotti35:05

        Yes. Good, good point. Um, okay, so here it's gonna go ahead and it'll just, like, fetch this, and then it'll add it to the skills. So we'll come back to that. Another thing that you can do, and this is really cool. So basically, Anthropic has built what are called skills marketplaces, and this is gonna use a new advanced Claude Code feature that we haven't looked at before called plugins. Okay, so this is saying discover plugins. And so what this is, is this is a bunch of sort of like packs of skills that companies have, like, built. So Context Seven, for example, is one where it gets you the most updated information on whatever, like, APIs or whatever you're using in your project. Most LLMs, they have, like, when they're trained, they have past data, so Context Seven is one where it will always pull the most up-to-date information. And so there's, there's a lot of these, like, pre-built plugins. There's one for Supabase, Supabase for your backend, Figma. There's a lot of these. So what we're gonna do is we're gonna... There's a linear one. So these just give you, like, a bunch of different skills and, and MCP configurations and things like that kind of right off the bat. There's one that's really good that is built by Anthropic themselves. And so this one, it's just called-- So we're gonna do /plugin install, and then it's just document skills. So these are just a bunch that Anthropic has built themselves to allow Claude to do things like create PowerPoints and create Word documents. So, oh, it's already installed. So we're gonna go ahead and look at our plugins. We're gonna, up here you see it says "Installed." We're just gonna go click Tab, and then we have document skills already installed. And what you can do is you can kind of like look at all the stuff that comes with this one. So this is a collection of document processing, including Excel, Word, PowerPoint, and PDF. And then it comes with a bunch of sort of components: algorithmic art, brand guidelines, canvas design. The ones that are interesting for us, there's internal comms, there's a skill creator. So if you wanna build your own skill, which we'll look at next, it kind of comes for free in this one. So if nothing else, I definitely recommend at least getting their Anthropic's official skills, 'cause these are things that as Claude works, it will be able to automatically create these, like, pretty nice things for you, especially the skill creator for anything else that you wanna build.

    13. AGAakash Gupta36:58

        I think I need that one 'cause sometimes I create the skill, and then it'll be like YAML front matter error, and it will have added in, like, some extraneous field. So this is a good one.

    14. CVCarl Vellotti37:08

        Yeah, that's a good one. And then so let's go ahead and do it. So, um, I talked earlier about how APIs are the main other way that you can have your LLM kind of interact with the world. And so what I'm gonna go ahead and add to this file is, um- So an, an example of where you might want to use an API where there might not be an MCP, for example, is, um, Google's NanoBanana Pro. So we're gonna get the API for Gemini so that we can generate images inside of Claude Code. So that's something that you might think you have to go to their tool, but because they have an API, we can do it in Claude Code right in our terminal here. So the main things that you need to do that is we're gonna go ahead and create... Oh, I think I already have one. But we have our Gemini API key in here already. And so the thing with using an, uh, an API is that you have to have a script that can run to hit that API.

    15. AGAakash Gupta37:54

        And just so people know really quickly, he just created a .env file. That's where you put in your API

11. 38:00 – 44:12

    Image Generation with Gemini API

    1. AGAakash Gupta38:00

       keys, and that's how you keep them safe. If you're wondering, like, if I'm ever vibe coding or something, you want to create these files.

    2. CVCarl Vellotti38:06

       If you have a .env file, so there's this other thing called a gitignore, which will pretty much come automatically with, like, repos that you create. This will have things that when you are committing to this to GitHub, when you want to save your information, these things will not get sent. And so one of those is your .env file. So that means that, like, if someone else, you know, you want to open source your code, you definitely do not want your keys in there because then someone else could use them. But in, in your gitignore, it'll automatically have that so that you're not committing that. But that, that's one good thing to check. LLMs are also pretty good about knowing not to commit that. But in this case, I have this with our Gemini API. So we're just gonna say use... So just as an, a quick demo here, this script that already exists. So use... What should we make an image of?

    3. AGAakash Gupta38:48

       Yeah, I was gonna ask you, like, why is it better to u- do it in Claude Code? I guess Claude Code could, like, write a better prompt for you or something. So maybe something that requires prompting, like maybe we're trying to create an infographic that doesn't look like the normal NanoBanana infographic style, but instead looks like something more professional and clean and maybe a different size.

    4. CVCarl Vellotti39:08

       Yeah. So let's create a nice infographic professional for, uh, how to use Claude Code skills. Okay. So this is kind of nice because, like, in this case, Claude, of course, knows how Claude Code skills work, so it'll be able to prompt the AI to-- with a lot of that information. The nice thing that-- about doing this in Claude Code is that, for example, if you have... You know, let's say that you really want your images to always have a specific s- type of style, or you want them to have certain colors or things like that. Then what you could do is you could save that as a file, and then you could say, "Claude, please create an image, and then make sure that you use these brand guidelines." And then you could automatically have those brand guidelines here. Otherwise, like, if you're in Gemini, then you have to, like, copy and paste those every time. And you can kind of, like, see what works and then save it. So the nice thing about doing it in Claude Code is that it can, like, make a lot of those, like, sort of workflows or, or things like that. It can write the prompt for you, and it can do it in here. It can also do other cool stuff like if you wanted to, you know, generate like four var-vari-variations of an image at once, then Claude could literally hit the API four times at, in parallel for you, which is something that would be pretty hard to do on your own. So it just, it just kind of unlocks, like, much more advanced capabilities for how you can use this once it's in code versus using, like, the sort of t-typical front ends.

    5. AGAakash Gupta40:21

       Yeah. In, in your experience, like, prompting NanoBanana Pro, some people have been talking about JSON prompting and different things like that. So could it be useful to create, like, a NanoBanana prompting skill as well?

    6. CVCarl Vellotti40:33

       Yeah, definitely. That would, that, that's a great example of, like, help me turn this, like, natural language prompt into a JSON prompt, and then you could have a skill that would automatically kind of understand what you're going for with that. That's a really good example. Okay, so here's what it created. Okay, so it created a Claude Code skills infographic. Honestly, it looks pretty good. Um-

    7. AGAakash Gupta40:50

       Besides the purple gradient. [laughs]

    8. CVCarl Vellotti40:52

       Yeah, which is another thing you could... Yeah, the purple, the classic purple gradient, the LLMs just for some reason love so much.

    9. AGAakash Gupta40:58

       Yeah.

    10. CVCarl Vellotti40:58

        Okay. So now let's go ahead and say create a skill for image generation or product managers for user journeys.

    11. AGAakash Gupta41:08

        Okay. User journeys. Yep. PMs need to create those.

    12. CVCarl Vellotti41:11

        All right. Talk through this diagrams. Okay. So now what we'll see is that for this first one, we kind of had to, like, tell it where that script was. But now what it's gonna do is it will create its own skill, so that next time we ask for it to create an image, it will know that it has that skill. And then this script that we had in there is, uh, part of that skill. So one thing that we've seen is, like, so far, all that we've seen is, like, sort of just text instructions. But the other cool thing about, uh, these instructions is they can, like, point to other parts of your sort of whole system so that it knows, like, where to pull that information from.

    13. AGAakash Gupta41:42

        And what script are we referring to again?

    14. CVCarl Vellotti41:44

        This is just the image generation script. So I'll open that here. So this is basically just, like, how do we actually use that Gemini API?

    15. AGAakash Gupta41:51

        Yeah.

    16. CVCarl Vellotti41:51

        So it tells it, like, okay, this is the model we want to use. Here's, like, the default aspect ratio. And it sort of just defines exactly how you can use that API to actually generate images. Uh-

    17. AGAakash Gupta42:00

        Makes sense.

    18. CVCarl Vellotti42:01

        Yeah. Of course, I had Claude Code write this. So, um, this is sort of a... It's, it's actually a pretty big script. So this is something that... This is an example of where using an API requires you to have, like, these scripts, and the AI has to already kind of know what the API is capable of. Whereas with an MCP, you get, like, all of the tools that it has basically for free.

    19. AGAakash Gupta42:19

        Mm.

    20. CVCarl Vellotti42:20

        Okay. And then now it's gonna test that for us, which is... The nice thing about Opus is it's very smart around it does work, and then it checks its own work, which is something that LLMs used to be so bad at, but they're getting so much better at.

    21. AGAakash Gupta42:31

        So if somebody wanted to recreate that script really quick, how do they do that?

    22. CVCarl Vellotti42:34

        Yeah. So the easiest way would be to basically get the Gemini NanoBanana, like, reference documents.

    23. AGAakash Gupta42:43

        Yeah, the API docs and then-

    24. CVCarl Vellotti42:44

        Yeah

    25. AGAakash Gupta42:45

        ... s- ask Claude Code, "Create a Python file to pull, to make and get requests to the API," or-

    26. CVCarl Vellotti42:53

        Yeah, exactly. So basically, what I would do is I would just say, "Okay, here's the API docs for this." And then I would say, "Create a script for us to create images with the Gemini API. Here's the reference docs." And then it will read those, and then it will create the script.

    27. AGAakash Gupta43:06

        Nice.

    28. CVCarl Vellotti43:06

        Yeah. Okay, cool. So that's, that's kind of, like, a good coverage of skills. So we've seen you can get skills from GitHub. You can get them from... or, like, newsletters, like Akasha's, which are definitely won't have any prompt injection. You can get them through the plug-ins marketplaces, which give you kind of a ton for free by just doing /plugin, or you can build your own. And kind of like a combination of all of those is, is what you're gonna do, like, as you build out your system. You'll include more and more of these skills. And skills are just- They're awesome because they are, they automatically trigger. And what's cool is, like, if you give it a long-running task, when it gets to the points where it could use a skill, it will automatically use that skill, and it will do it sort of in a very specific way, rather than sort of just whatever the natural way that it would do it. Which, if you wanna build these reusable workflows and the, especially for long-running tasks, keeping, uh, the LLM on a narrower sort of don't let it go whichever direction it wants, 'cause as soon as it starts to drift, it'll drift more and more. Skills really help it keep on that same path so that it can keep the output really high, like the high quality all the way through.

    29. AGAakash Gupta44:06

        Hmm.

    30. CVCarl Vellotti44:07

        Okay, cool. So now we have this new skill called PM Diagrams, which we'll go ahead and use a little

12. 44:12 – 51:00

    Creating PRD from Survey Analysis

    1. CVCarl Vellotti44:12

       bit later. Okay, so that was kind of like a big tangential exploration into skills. For now, let's kind of go ahead and continue with our demo of we're building this calendar integration feature. We got those survey results. We had them analyzed. So now let's go ahead and create a PRD based on those survey results.

    2. AGAakash Gupta44:27

       Makes sense. So step one, you send out your survey for feedback to Google Docs. You can even set it up Gmail. You could have it send an email all via Claude Code.

    3. CVCarl Vellotti44:36

       Yep.

    4. AGAakash Gupta44:36

       Step two, you had it create that survey in Qualtrics using Qualtrics MCP, again, in Claude Code. Step three, you have it analyze the results. And now step four, create a PRD.

    5. CVCarl Vellotti44:46

       Yep. And then probably, in this case, I don't have one, but it would make sense to potentially have a PRD writing skill, which would have your template in it. It just makes it, like, kind of nicer to use also, so you're not having to remember, like, "Oh yeah, what was the name of this markdown file that I had with the template?" It will just automatically use that template. Okay, so it's reading the survey results, and then it will go ahead and write our PRD based on kind of all the information we've given it.

    6. AGAakash Gupta45:08

       Yeah. How many times do you just go look for that latest PRD template or that latest example of, "Oh, this is a good PRD I want it to look like"? If you put it into a skill, you can really stop that search.

    7. CVCarl Vellotti45:18

       Yeah. Especially, you know, as you live in Claude Code more and more and you build out more of your system, it's helpful to have more stuff happen automatically so that you don't have to remember exactly how every single thing works. Okay, so I, I asked it to create a markdown file here and then also push that to Google Docs. So once that's done, we'll take a look at it. Cool. Okay, so we now have our calendar integration PRD.

    8. AGAakash Gupta45:39

       It's the moment of truth. Is it any good?

    9. CVCarl Vellotti45:41

       Yes, let's see. We'll, we'll look at it in Google Docs when we, when it pushes it out. Okay, so it found the MCP. Oh, it needs my Google email. One thing I should do, um, that would make this faster, so as a reminder, right now it keeps, every new session it keeps forgetting, like, what my email is, which it needs for the MCP. The way that you can have Claude just always know information is you create a Claude file. So what I really should do is I could sh- should create a claude.md that says, like, "Whenever a user uses Google MCP, use their email," which is my email, Carl Vellotti. Okay, so now it will go ahead and do that. And then I think before that happens, we can also just look at the PRD here. Um, okay, great. So now we see that it's actually creating it and that it was successful, so it'll give us the link in a second. I didn't really give it much of a template, but it is including the user research here, which is helpful. And then it is also writing out the requirements. So, okay, so now we have our PRD for this feature, and then we have the Google Doc, so let's look at that.

    10. AGAakash Gupta46:37

        So that's Opus 4.5 with no instructions. So much better than GPT-4 or something.

    11. CVCarl Vellotti46:42

        Yeah, like-

    12. AGAakash Gupta46:42

        If that's the last time you've done it, guys, like, even just without anything, it's pretty good these days.

    13. CVCarl Vellotti46:47

        Yeah, 100%. And it's, like, it's pretty nice. Like, I mean, I would have to read this more carefully, and I'm sure, you know, whatever company you work at and things. But if you give it a lot of context, then these things can be actually pretty good on their first pass. And so it's really showing that user research in here, which, you know, a lot of times you present to people, it's like, "Well, how do we know this is a problem?" We've already gotten ahead of that with our user research. And we know now, like, what are the most important things for this. Okay, great. So now, like, the next step would be we have our PRD, but, you know, this is pretty text-heavy. So let's say that we wanted to create kind of a user journey flow for this document. Well, luckily, we already created a skill for that, so let's say... Okay, here's the moment of truth. Let's see if it remembers to use the skill. So I'm gonna say-

    14. AGAakash Gupta47:26

        [laughs]

    15. CVCarl Vellotti47:27

        ... add a user journey map showing the calendar integration flow. Okay, so we'll see. If, if this works, and it'll be awesome, it should know that we created... Oh, let me try one thing. Um, we're gonna close this. One thing, it might be that this, we hadn't restarted Claude. I'm not sure if you need to for it to pick up skills. So what we're gonna do is we're going to, we're just gonna restart Claude. A really good way to do that is you just close Claude, and then you just type resume, and then you can go back to that same conversation. So that's a nice way where if you need to restart Claude but you don't wanna, like, completely lose the conversation, you can just type leave, hit resume, and then that will bring you back into it.

    16. AGAakash Gupta48:00

        That's actually one thing that I dislike about Claude Code greatly versus Claude is, like, discoverability of my prior chats. Is there any way to go back to, like, go see something? Like, just today I picked up a chat from two weeks ago.

    17. CVCarl Vellotti48:13

        Mm-hmm. Yeah, I mean, sometimes you wanna do that. The, I think the only thing that you can do that's helpful with that... Okay, so it knows how to use it because of this, but it didn't use the skill again. So I think this is something, unfortunately for this demo, we're not seeing use those skills, but it is still, like, using the tools that we've given it.

    18. AGAakash Gupta48:28

        Yeah. So just, I would say, folks, if you wanna use the skills, for now just specify in your prompt to use the skill. That's what I had to do.

    19. CVCarl Vellotti48:35

        Yeah, I think that's, that's good advice. I'm sure that that trigger will get better with some future version of Claude, but not working well. But the, the, the concept is still the same here. To your point around, like, you know, having specific sessions that you wanna find, one thing I just discovered that's kind of cool is you can, um, you can actually name your sessions. Oh, you, so you can do rename, and then let's say that this is, like, um, we'll just call it test name. But what's cool is then when you type resume, you actually see, instead of just showing you, like, whatever the first prompt from that conversation was, you'll, you'll have it here. And then what's cool is you can actually search. So now I can find that session in here. So that, that's one thing. You, you kind of have to, like... It'd be nice to auto-name them, which, um, is kind of a surprising feature. And I bet there's a chance that by the time people watch this video it will be a feature. But for now, you can name sessions, and then you can search for them and get back to them. So, and it has basically every session you've ever done, which is pretty cool.

    20. AGAakash Gupta49:22

        And while we're on the topic of things that it's deficient compared to the UI, there's still no way to, like, select text from earlier in your prompt, right? Like, if I just type something, I only know how to, like, go backspace if I make a mistake.

    21. CVCarl Vellotti49:32

        Oh, yeah, yeah, yeah. It's very annoying. Like, for example, yeah, like, I can't, uh, edit this.

    22. AGAakash Gupta49:38

        Yeah.

    23. CVCarl Vellotti49:38

        Like, it only goes... Yeah. That, that is definitely a, a deficiency of being in the terminal, for sure.

    24. AGAakash Gupta49:42

        Hmm. Okay, so that's still a deficiency.

    25. CVCarl Vellotti49:44

        Yeah. Okay, cool. So we now have this user journey. Honestly, looks pretty good. And then what you could do is you could take this and you could put it straight into your document.

    26. AGAakash Gupta49:53

        If you guys become this AI native PM, that thing is gonna be one of the most annoying things [laughs] that you guys have to deal with.

    27. CVCarl Vellotti49:59

        Yeah. Uh, there's ... The only thing that helps is, like, you can, if you ... Like, just getting really comfortable with, like, if you hold Option, it jumps words. If you hold con- Command, it, like, jumps entire lines. That can make it a bit faster, and then you can go up and down. But yeah, it's just, just an annoying thing about being in the terminal for sure.

    28. AGAakash Gupta50:15

        Option is a pretty big hack, actually, right? Because then you can edit there.

    29. CVCarl Vellotti50:19

        Yeah, you can do this, and then that also lets you, like, delete whole words at once.

    30. AGAakash Gupta50:22

        Yeah.
13. 51:00 – 55:30

    Hooks Feature

    1. CVCarl Vellotti51:00

       Um, have you done anything with hooks? Have you played around with them at all?

    2. AGAakash Gupta51:03

       No.

    3. CVCarl Vellotti51:04

       Okay. So the concept of a hook is that every time you have Claude do something, there's kind of, like, a bunch of different steps within that entire, like, life cycle. So there's as soon as you do the prompt, there's when the prompt ends. I'll kind of look at, like, all the different possible triggers for a hook. But the cool thing about a hook is that you can, like, set automatic actions for Claude to take when you're set some point in that life cycle of, of when you give it a request. So here we're g- just gonna do a really basic example. What we'll do, and this one's actually very fun, is, um, we're gonna create a hook that n- that gives us a notification and a sound as soon as it finishes an action. So where that's useful is, like, I'm gonna have it create a presentation next, and presentations take quite a long time to make. And, uh, let's say I don't wanna just, like, sit there, or even within this demo, I think if you're following along, there's a couple times that I was like, "Oh, we'll come back to that terminal," and then I, I actually forgot. [laughs] And that's because there's no notification, right? There, there's no way to tell. So we're gonna actually create a hook that gives us a, a notification, and it's really easy to do. All we're gonna do is we're just gonna say, "Create a hook that makes a Mac notification," and we'll say sound as well, "when you finish a task." And Claude kinda knows how to create its own hooks. Oh, actually, it's, uh, or it's gonna be smart and it's gonna look up when it doesn't know how to make a hook. So it's, it's looking it up. What's cool is I think because we have the plugin for those, the Anthropic plugin, it, like, fetched from its own, its own documentation without actually having to do a search, which is kind of, like, another cool thing that it's just very smart around knowing where to look for this stuff.

    4. AGAakash Gupta52:32

       Mm-hmm.

    5. CVCarl Vellotti52:33

       Okay, so it's seeing if there's a settings file. So while this is happening, um, what I'll do is I'll go ahead, and I have kind of made this little graphic around, um, all the different places that you can have a hook. Okay, so within Claude Code hooks, there's, like, a bunch of different ones, and I'll cover, like, a couple examples where we might wanna use this. Like, as Claude works, this does, like, a, a ton of stuff. So there's, like, when you start a session, you could have a hook that says, "Hey, as soon as you start a session, pull in all linear tickets for this project." And that might be helpful if you have, like, a specific repo that is just around managing tickets, or, like, a subfolder. Then as soon as you start, it pulls in those tickets. Or as soon as the session is over, or right before it uses a tool, or right after it uses a tool, you might wanna have, like, a, another agent. Or if it's, like, a specific tool around writing a document, you might have a- another agent that triggers as a hook right after it does that, so that it can review that work. Or right at, let's say you have, like, a long-run coding task, then you might have something where every time it finishes writing some code, it also runs the test suite against that code. So these are very advanced and, like, I don't think they're... They're definitely kind of mostly more for coding use cases, but if you're creative, then there's, like, sort of some really interesting things that you can do with hooks. Another good one is, like, so as you, if when you use Claude Code for a long time, eventually it will compact, and you'll kind of, like, lose all the session history before, and it's, it can be kind of actually hard to get back to unless you are ... Like, Claude will literally clear it from the UI. So one thing that you could do is you could say, "Every time before you're gonna compact the conversation, I want you to automatically export it and put it into, you know, a folder with a, as a markdown file," so that if it messes something up after it compacts, you can still get back to that conversation history before it did the compaction.

    6. AGAakash Gupta54:06

       Hmm. Very interesting.

    7. CVCarl Vellotti54:08

       Yeah. So these are definitely more, this is, like, real advanced and I haven't used these too much, but it's kind of, like, a good feature to know, and I think that there'll be some, like, really interesting setups with these hooks if you're, like, really trying to create, like, a very tightly controlled, um, Claude Code session.

    8. AGAakash Gupta54:21

       Mm-hmm.

    9. CVCarl Vellotti54:21

       Okay, cool. So let's test this. Now, a- any time it finishes a task, which is really, like, any time it finishes any sort of, like, message, it should make a notification. So let's see if it works. Oh my gosh. Man, Claude is not helping me today. Sometimes live demos... I tested this before and it worked, and then now, now not. Um-

    10. AGAakash Gupta54:38

        [laughs] That's the fun.

    11. CVCarl Vellotti54:39

        Yeah, that's the fun. Okay, let me, let's, uh, we're gonna resume that session. This, we're gonna do some live debugging here. So let's resume that previous session. And it says, it w- saying that it already created, so let's tell it to delete that. Okay. So while that's happening, we're gonna do some classic Claude stuff. We're gonna go back to our other PRD that we had written, which is here. So we can just give it the name of this. So please create a presentation. So this is gonna use the presentation skill. Using your presentation skill for PRD in Google. And then we know, use my email. Okay. So one thing, as you use Claude, you kind of wanna get very comfortable just jumping between sessions as it works, 'cause there is always a little bit of downtime between them. So it's gonna pull in that PRD. Uh, actually, it says

14. 55:30 – 1:01:49

    Creating Presentation from PRD

    1. CVCarl Vellotti55:30

       it found it locally, but it also found it through MCP, and then it will create a presentation. So we'll, we'll come back to this, and then let's see what's happening with our hooks.

    2. AGAakash Gupta55:36

       The hook file is deleted, so now we'll ask it to recreate.

    3. CVCarl Vellotti55:40

       Yeah. Okay, so I'll restart Claude, see if this helps. We'll-

    4. AGAakash Gupta55:42

       And so that, what you're doing quickly is you're just doing Control + C up.

    5. CVCarl Vellotti55:46

       Yes. Okay. So now we're gonna give Claude another chance here. We'll see. It's gonna look it up again, and then it will create it. And then let's see if this other one... Okay, so, okay, perfect. So this time it looks like it is figuring it out. So it knows that as... Okay, there we go. So it actually did use this skill. So this is the first time in this demo we've seen it sort of show a skill correctly. And so I think your tip of saying to just use this skill is probably the best way to use it for now. And so it'll look like this. It's just looking to /documentskills PowerPoint. And now I have pretty good confidence that we'll actually do it correctly, 'cause w- we know it's actually, like, in that skill and reading the instructions. And so now it's creating the slides using its own thing. And it starts that as an HTML, and then it converts it into a, a PowerPoint. Okay. And now it's going. And what it's doing here, it's kind of cool, classic, like, kind of Claude Code stuff, is we see that it is intelligently starting, like, multiple sessions for each slide instead of working on all of them sort of in sequence. It's just, it's figuring out how to start them all in parallel. So we're seeing some of that nice sub-agent features here.

    6. AGAakash Gupta56:44

       And regular Claude can't do that, right?

    7. CVCarl Vellotti56:46

       No, regular Claude can't do that. That's definitely only a Claude Code feature.

    8. AGAakash Gupta56:49

       Yeah. So that's one of the big benefits, is the parallelization.

    9. CVCarl Vellotti56:53

       Okay. The only other thing I can think of, of why this isn't working is that we might need to, like, close all the Claude Codes for it to pick this up. But yeah, in general, hooks are something that's good to explore, and it's kind of like a good coverage, uh, when you're thinking about what are all the Claude Code features. Hooks are a good one, and you can add, like, you can actually add sounds, like the whole UI. Some people have really funny ones. Or you could, like, prank someone where they have it, like, make, like, all fart sounds ev- basically any time it takes any action. So that is, like, a kind of a good power feature to, to be aware of, um, because you'll probably see that, like, on Twitter and, and things like that. Let's, um... While we're waiting, maybe we can... Sometimes you can just... Okay, so it's creating all the slides here. It's not working. Do we need to close all the... Okay. We are getting, we are making good progress on our overall demo. So let's, um, we'll just kinda jump ahead in time. We'll look at these slides after they're created. But let's say that we've presented these slides, and now we've gotten feedback, and then now we need to go ahead and incorporate that feedback into our PRD. And then we'll move into the very last part of our demo, which is using that. Basically, now we have our full PRD, we've presented it to stakeholders, we have the information, and now we wanna create the actual engineering tickets. And, uh, from the very beginning of this whole session, we set up Linear. We'll see that come back into play, and it'll be able to create really good tickets for us that go straight into Linear. And then as kind of like our very final step, what we'll do is we will, after those tickets are created, you know, stakeholders, they, you never really want them to be in Jira or, or Linear, like, kind of checking out what's going on in your project. You wanna give them sort of a source of truth. And so what we'll do is we'll take it, once it's in Linear, we'll pull it back into our Google Doc and have, like, a status update section. And then as a very final thing, I'll show you how to, you can make that update automatically.

    10. AGAakash Gupta58:35

        Sweet.

    11. CVCarl Vellotti58:36

        Okay, cool. So let's go ahead and imagine that we have our feedback here. And what, for that, I cr- I think I pre-created stakeholder meeting notes. And then what we'll do is we will, um, have it incorporate that into the PRD. Okay. So we saw that PRD from earlier, and now we will go ahead and update it with those meeting notes. Okay, so it found it locally, uh, and then it will find it also in Google Drive. And now it's also gonna find the stakeholder meeting notes. So that's one thing that's nice is, like, you can just really control so many parts of this, um, whole sort of workflow just straight from here. You can just kind of tell it, "Do this and this and then do this." And if you give it the right tools, it will really figure out how to do all of that. And when it makes a mistake, it will kind of figure out how to fix itself.

    12. AGAakash Gupta59:16

        Mm-hmm.

    13. CVCarl Vellotti59:19

        Okay, so still working on those slides, so it's kind of a, a long-running task, so this is a good one where, you know, you'd wanna definitely go do some other work while this is happening like we are now. Close this one. Okay. So it is looking at its own tools and figuring out what it can do.

    14. AGAakash Gupta59:34

        Yeah, improving your own ability to do, like, two things at once is so powerful here because there is, like, these micro context switches you need to keep making.

    15. CVCarl Vellotti59:42

        Okay, so in this case, for some reason, it's not able to use that MCP, so let's just get the stakeholder meeting notes. So in this example, this is, like, notes that you may have taken while you were in a meeting. And so, um, basically, what are the key decisions? So you know, you've presented your PRD, you've done a lot of thinking, hopefully you've met with your engineers before, and then, you know, when you're talking to your main stakeholders, they say, "Okay, we really need to make sure it's compliant," or, "We cannot ship this without SSO." So maybe you as a PM, you're trying to get it out quicker. You thought maybe you could do just, like, regular login, but then, you know, your head of engineering or your, you know, head of sales or something is saying, "We cannot go out with this feature unless it has SSO." So this kind of gives us more requirements for us to include when we, uh, sort of update our PRD. So-

    16. AGAakash Gupta1:00:26

        And hopefully you're using an AI note-taker too, so that you're truly AI native. [chuckles]

    17. CVCarl Vellotti1:00:31

        Yeah, yeah, exactly. Yep. That's a good example, like Granola or one of those. Do you have a favorite AI note-taker?

    18. AGAakash Gupta1:00:37

        I like Granola because it doesn't follow you into the meeting.

    19. CVCarl Vellotti1:00:39

        Oh, how d- how does it work? I actually haven't used it. Does it kind of sit on your computer?

    20. AGAakash Gupta1:00:44

        It's just the audio.

    21. CVCarl Vellotti1:00:44

        Mm.

    22. AGAakash Gupta1:00:44

        So Granola's the way. But it will just r- it's, like, it's, like, always there, so if you're worried about your privacy, don't use Granola.

    23. CVCarl Vellotti1:00:51

        Ah. That's, um... I know I've seen some pretty funny, like, memes where, you know, you, you join a, someone will join a meeting early on purpose, and then everyone else's AI note-takers will join ahead of them. And then they'll just say, like, super crazy stuff. And then as soon as people join the meeting, they'll stop talking. So then the, the, [chuckles] the transcript or the summary of the meeting will be, like, completely insane, and people will be confused why. I don't think anyone has done that, but I actually do think that would probably work.

    24. AGAakash Gupta1:01:15

        Oh, yeah. That would be a good chaos monkey sort of April Fools' joke on your team.

    25. CVCarl Vellotti1:01:20

        Yeah, exactly. Cool. Okay, so now we see that it, it was able to kind of update everything here with what it found in those meeting notes. And then as our final step here... Great. Okay, so it updated the PRD. It's kinda telling us what it updated. So now what's cool is at this point, you know, this PRD is really good. Like, you've worked with your team to write it. You've gone to the meeting with meeting notes. You probably talked to your engineer about those meeting notes. You've updated it. So now at this point, you really have, like, a great PRD with all the requirements. So the final step is let's create the tickets.

15. 1:01:49 – 1:08:08

    Creating 19 Engineering Tickets in Linear

    1. CVCarl Vellotti1:01:49

       Creating tickets as a product manager depends on how your team is set up. Sometimes engineers do it, sometimes PMs do it. But just in general, it's always, like, such a painful process, 'cause you know the information, and you just have to, you know, go through the UI and click all these buttons and, like, kind of make sure everything's all organized. Whereas now with Claude, it can just literally create all those tickets for you. So let's say now create the tickets for this PRD. In the project. Let me see if I already have a named project.

    2. AGAakash Gupta1:02:17

       And if your team had a really specific way about w- how they wanted tickets, you could create a skill around that.

    3. CVCarl Vellotti1:02:21

       Yep, exactly. Okay, cool. So we have a Taskflow Calendar Integration project. So, and we still have that Linear, uh, MCP set up from before. So now it is just starting the task, so it's gonna break it into the engineering tickets and then create them in Linear. Great. And w- what's so cool is that because we have that Linear MCP set up, it knows exactly the tools that it has, and it is just starting to work on it here.

    4. AGAakash Gupta1:02:46

       Yeah, we didn't say Linear, right? It just, it knew Linear is the system.

    5. CVCarl Vellotti1:02:49

       Yeah, exactly. I have all the information I need. Let me create a to-do list, and then create the engineering tickets. So we're seeing that nice agentic-ness of Claude Code.

    6. AGAakash Gupta1:02:57

       And that's because the MCP was on for Linear.

    7. CVCarl Vellotti1:03:00

       Yeah. Okay. And then we saw our slides that we were having it create from before. Let's see how these turned out. Of course, this is another area where, like, in that slide skill you d- could provide, like, more context. But what's cool is, like, this is, like, a real, this is, like, real Google Slides. These are not images.

    8. AGAakash Gupta1:03:13

       Whoa.

    9. CVCarl Vellotti1:03:14

       Anything-

    10. AGAakash Gupta1:03:15

        Wait, that's actually sick.

    11. CVCarl Vellotti1:03:16

        Yeah, right? And it, like, designed it kind of nicely.

    12. AGAakash Gupta1:03:19

        I don't think people know about this.

    13. CVCarl Vellotti1:03:22

        Yeah. And then anything that you wanna change, like, it won't get everything 100% perfect, but it can be pretty darn good, and this is only improving. And so-

    14. AGAakash Gupta1:03:29

        Wow

    15. CVCarl Vellotti1:03:29

        ... we have, like, a pretty, a pretty nice deck here, I would say. Like, this is, you know, you might wanna make some edits, but this is not a bad first pass at all.

    16. AGAakash Gupta1:03:39

        This is saving you [laughs] I just, this is my entire career. I would have to spend time, like, we'd create a PRD, then we'd create a deck. It would take an hour or two. Just turn that into a prompt, and if we had made this a proper skill that it knew about your template and how you like your slides, it would be even closer to ready.

    17. CVCarl Vellotti1:03:55

        Yeah, exactly. So that's kind of the, the final payoff. And then going into our example, we had those meeting notes from these slides, which we're now making the tickets from. But yeah, what's so cool is that, like, this is, like, a real presentation. Like, this is all editable. Uh, it made all the little... Like, that's kind of why it was taking so long, is 'cause it was defining all these things in, in, like, HTML to see how big they should be. And one thing that's cool, and I think this will be coming soon, is you could actually have Claude, like right now, it can't really, like, see its work very well, so it knows, like, it knows how the code should be, but there's gonna be a pretty cool feedback loop soon, or I bet there's people who have already built this, where you could have it use, like, uh, Puppeteer or Playwright are ones where it can kind of take a screenshot of its own work, and then you could have it iterate. So this was its, basically its first pass creating these, but it would be pretty cool to just have it actually look and then make sure that things were aligned. Although it did do a pretty good job with all this.

    18. AGAakash Gupta1:04:45

        You mentioned those tools. I think people might have heard about them and not know them. What exactly are Puppeteer and Playwright?

    19. CVCarl Vellotti1:04:49

        Yeah, so those are really good ones to be aware of, like, if you're doing anything where you're, like, creating any sort of, like, visual UI, where you want Claude to be able to see its own work. And so, um, they're basically just tools for it to create a, a page, like this one, or we had, um, we had the one earlier where we were looking at hooks. It can build that page, and then it will screenshot it and then, like, send that screenshot to itself, so it can actually see, like, what its work is, and then it will, can, can make edits on its own. And so that's, like, a, a really good way when you're doing any sort of front-end sort of development at all. Give it those skills so it can check its work. 'Cause sometimes, you know, it'll say it's done, and it'll all be all functional, but it'll look terrible because it just, like, doesn't really know, or things will be aligned weird. If it can see its own work, it usually is, like, very good at noticing those big discrepancies and fixing them on its own.

    20. AGAakash Gupta1:05:37

        I can't emphasize how cool that feature is that we just showed, guys. I mean, did we just kill Gamma for PMs with this podcast episode? Like-

    21. CVCarl Vellotti1:05:44

        Yeah, right?

    22. AGAakash Gupta1:05:45

        Pretty sick.

    23. CVCarl Vellotti1:05:47

        Okay, cool. So let's see. So it, it took our PRD, and now it created 19 engineering tickets, and it also gave them priority, which would be based on what was in that PRD. And it's saying that they are in this project. So let's go ahead, moment of truth.

    24. AGAakash Gupta1:06:00

        Did it do a good job creating these tickets?

    25. CVCarl Vellotti1:06:03

        Okay. So we, we can see all these here. They were all created today. And then we see them, P1, P2. These are pretty nicely sort of, like, labeled, like this is how I would do it as well. So let's, like, look at one of these P1s. Okay, overview, requirements, technical considerations, and then the acceptance criteria. And so these are still, like, slightly high level. I think that's 'cause the PRD itself was high level. This would be sort of a great thing that you could hand off to your engineers, where they might actually, like, add more information. But this is, like, I think as an engineer, you're really getting a lot of that information around what exactly do you need to be able to build without necessarily... What would be cool is you could even, like, have Claude, if you connected it to your code base, 'cause here it's not connected to the code base, so it really, it's just saying what the requirements are without saying, like, how to build it or anything. But if it was connected to your code base, you could even have it include a lot more technical details so then an engineer, when they're reviewing, they already kind of have, like, a great place to start off as they're going through these tickets.

    26. AGAakash Gupta1:06:55

        And a lot of people might be wondering about that. How do you do that? How do you connect it to your code base?

    27. CVCarl Vellotti1:06:58

        Yeah, so in this case, that'll be kind of the last thing we look at here, actually. So we kind of just saw, like, some two awesome payoffs, right? We saw the ability to actually, like, where we're seeing is being in Claude, but then actually doing work, like, out in the world. We have these slides. We've had these documents. We pulled them in. We've, we've created the tickets. The last thing to kind of talk about here is, like, how can we actually have Claude, like, work within a code base? So I think as a product manager, in order for that to happen, you'll need to get GitHub access, which I think, you know, two years ago, a product manager's getting access to GitHub would be, like, weird. Like, what, what do you really need to do looking in the code? But now, as more and more PMs are becoming AI native and, and using these tools and wanting to ex- be able to explore the code base, I've, I've already heard, uh, I went to a Cursor for product managers sort of little meetup, and one of the questions that Cursor asked was, "How many of you have access to your code base?" And it was basically all PMs that were at this meeting, and more than half said that they did have access. So that's one thing that if you, you might have, you might just need to ask at your company. Otherwise, I think it'll be a thing that more and more product managers actually do have that access. And the easiest way to do it, to give Claude access to your code base, is you would just get the GitHub, you would just get the link to your repo, and then you would just point it at that repo, and it could explore it.

16. 1:08:08 – 1:19:39

    GitHub Integration & Remote Worker Setup

    1. AGAakash Gupta1:08:08

       Cool.

    2. CVCarl Vellotti1:08:08

       Okay, very last thing, and this is very cool to kind of capability of Claude Code, and a way that- As you're building out your systems at work, or you're building out your systems, like, personally with your, you know, Claude code at home, is actually sort of connecting, like, your local files to GitHub and then adding Claude to your GitHub so that you can have it take actions for you even when you're not at your computer. And this is what is just so cool, and, um, I wanna just kinda demo 'cause I use this all the time. So let's go ahead and kind of move into our very final phase here. So what we're gonna do is we will go ahead and let's, let's have it sync everything up. So right now, all of this, uh, all of this code and everything is already in GitHub. If you need it to be in GitHub, all GitHub is, is it's really just like a storage of files exactly like this. Like, if, if I were to download this project from GitHub, it would just have all of this information in it. And so I'm gonna say, please sync this repo, which just is, like, short for repository, to GitHub. And it kind of already knows that it, it is, like, which GitHub it is connected to. Okay, so it's gonna check and then sync, and then the very last thing we're gonna do is we're gonna actually add Claude as an app to our GitHub repo, and I'll, I'll show you why this is, like, so cool. So right now it's saying there's nothing new in the repo, so we're gonna be go ahead... We're gonna be good to just go ahead and push everything into it. So what's cool is all of the work that we've done locally, it's otherwise only on this computer that we just looked at all this on. But by pushing it to GitHub, now it is, like, all of those files in the current state are saved in GitHub. So if we went to another computer, we could just pull all that same information down, work there, push it back up. So it's a good way for you to, you know, transfer work between different computers or if you we- had something collaborative, they could download it. So that's why it's so awesome for engineers. Okay, so it's now synced, so we can go ahead and open this. And then, um, this repo I will make accessible to everyone who's watching this video, so you can... It has all the files that we used throughout the whole te-, whole time so that you can follow along. And so if you were on another computer, you could just copy this into... You could just s- tell Claude, "Hey, let's work on this repo," and it would pull all those files in. Okay, so the last thing we're gonna do here is we're gonna go ahead, and we're going to add Claude as a, um, app. And then I'll show you something cool that you can do with that. Okay, so the way it works is you go to github.com/app/claude. So there's all different kinds of apps you can add to your GitHub repo. We're gonna configure this, and then, um, I already have mine turned on so that Claude basically has access to all the repos. But then there's still one more step that you have to do, which is you have to go, um, into that specific project. And so in order for Claude to be able to work in here, you have to add your API key. So what we're gonna do is we're just gonna add my Claude API key, which is Anthropic key. And then, again, I'll, I'll go ahead and delete this key right after the video, but we will include it here. Okay, so now Claude has been added to this repo. So what's cool is we can open up a new issue. So let's go ahead and test this. Create a new issue that just says, "Say hello." Great. Then we're gonna make a comment @claude, "Say hello." And now what's gonna happen is Claude has been assigned that. So I could be-- And I could do this from my app. I could do it from anywhere else. Right away, we kind of see this. Whoa. We get, like, a, a real emoji reaction here. Claude is, is telling us that he has seen it, and then now he's gonna work this ticket, and it will be a new comment. And the way that you can see these is if you go to, um, actions... Oh, it actually picked that first one up too. So we go to our actions in GitHub, and there's, there's a lot. You know, uh, as we kind of saw with Hooks, like, when you get into some of more the engineering world, there's, like, a lot of customization and setups you can do. But this is just sort of the simplest thing where we're gonna add Claude as an app so that it can take actions for us in the repo. So now we can open one of these and, uh... Oh my gosh. Could not... Man, all these, these demos are not working as well as I would've hoped today. Okay, well, one thing that you can do also is, like, with this repo, because we have Claude also here, we can say, "Look at the open issues. Why was Claude not able to work them?" Okay, so, and now our Claude that we have, like, on our own device is gonna investigate, and it will be able to look at the details, and it will tell us. So this is, like, another good thing is, like, um, you can have-- And this is, like, a good workflow in general is, um, let's say, like, as a, as a real world scenario. So let's say that you as a product manager, you have your PRD on your own device locally, and then you also have it in Google Workplace. But let's say you're at lunch, and, you know, you're talking to your engineer, and something comes up where it's like, "Oh, did we consider for that calendar integration feature, did we consider how we should handle time zones?" It's like, "Oh yeah, I forgot. Ar-Arizona actually doesn't have daylight savings." Then, you know, that's one of those things that you kind of normally you just have to remember when you're back at your desk, start a Slack conversation about it or something. But what you could do instead is you could actually just go into your GitHub repo on your phone or something. You could create an issue that says, "Make sure that we account for time zones, uh, in Arizona." You could create that, and then you could @Claude, and then by the time you get back to your desk, it will have already worked that ticket so that you can then just pull it back down, and then you can see what it updated. And you can do all this stuff when you're, like, not even at your computer. One thing I personally use it for is, uh, I think I'm kind of a, a crazy person for doing this, but I have, like, my whole journal is, like, on GitHub. Luckily, it's a private repo. And I will just open an issue that's, like, a voice transcription of my day, and then I'll, like, @Claude, and then Claude will add that to my journal for me. So there's all these kind of different things you can do when you are away from your computer with Claude, like, a-accessible in the cloud. And that's just kind of the last thing I wanted to show is there's, like, step one of, you know, using Claude code while you're at your computer. But there's a lot more automations you can start to do if you add Claude into your repo. And then there's other stuff you can do too where, like, you can have it automatically run every day and then maybe check a couple different sources and synthesize them and then add them into your repo so that you can pull them down when you're at your computer. So even though we covered fairly advanced features here, there's still, like, a lot more to kind of keep exploring with Claude and these automations when it's in the cloud.

    3. AGAakash Gupta1:14:00

       And since we've put people on GitHub, can you just give them the basics of using Bi- GitHub? 'Cause I think a lot of people aren't gonna know, like, pushing and pull requests and branching. What do they need to know about these terms that they've heard if they're gonna use GitHub?

    4. CVCarl Vellotti1:14:12

       Yeah. So GitHub is, it's just the best place to kind of, like, store your files. It's just the easiest way with, for Claude to work with these files. I was reading something, and it said that Claude is fiercely competent with Git, and that's kind of, Git is, like, the language of using GitHub. And so, like, literally any time you need to do anything with GitHub or with Git, you can kind of just ask Claude, and it will basically do it correctly. But the main things to know are, so we have our, this is a repo or a repository with all of our files. If we want to get those files onto our own computer, then we're gonna pull them. And then if we've done work on our computer and then we want it to go back into the repo, then we're going to push them. That's the really, that's like the absolute basics. And then one thing that Claude will do, for example, let's say that you have, um, done work on your computer, but you also want to pull work in, it will make sure that... It- it's very, very good. It is fiercely competent at Git, so it'll make sure no work is, is lost. That's r- that's really the basics. If it's like a private repo where it's mostly just your own stuff, then it's really just pulling and pushing. The one other thing to know is, like, let's say you're prototyping or you're building your own feature, then you might want to, before you actually, like, just push that code, like, all the way into, like, the main repo. Like, let's say that we had a website that was reading from this, and we wanted to make sure that the website was, like, working with the new code rather than just sort of like pushing it onto, like, our main live site, then what you can do is you can open up a branch. And what that is, is it's basically, like, a branch of that code that has new stuff, and you can push the branch, but that won't make it go to, like, whatever is actually deployed. And then you can test that branch. And then the last thing to know is that once you have said, "Okay, this branch looks good," then you can merge it into main, which is sort of like your main repository. But I would say for most use cases of, of product managers, you kind of just need to know that you need to pull files down, and then you need to push them up, and that's how you're able to sync them.

    5. AGAakash Gupta1:16:01

       Cool. So simple enough. Pull it from your GitHub, push it back when you're done. If you do this at the beginning and end of your session, now you can use GitHub. Now you can have Claude as your remote worker when you're out on the go.

    6. CVCarl Vellotti1:16:13

       Yeah, exactly. And, and we won't cover it here, but, um, once you have Claude in your GitHub repo, then you can just have, like, your local Claude set up all kinds of cool stuff. You can have it set up workflows where it checks a certain website for you every day and then sends you an email with updates from that website. You can do kind of, like, all types of things. And once you kind of get into that GitHub world where Claude is up there, then you can work with your cloud Claude and your local Claude to do just even more stuff.

    7. AGAakash Gupta1:16:39

       Powerful stuff. Awesome. So if we had to summarize, right, what we've taught you guys today is from the beginning to the end of the feature how you can get your survey responses, write your PRD, create your slides, create your tickets, even potentially service some of those tickets with Claude as an agent on your code base. So these are the big use cases of Claude Code. Is there anything else people should be exploring when they're thinking about Claude Code for PMs?

    8. CVCarl Vellotti1:17:04

       I think the main thing is just it's, it's around everything that we've covered today. I kind of hope that it was really just, like, a good kind of inspirational example of seeing you can use Claude Code locally, it can pull stuff in from anywhere, you can work on it locally, and then you can push it back out. And any time you're starting, like, a new task or, or something else, just see, "Can I do it in Claude Code?" is my best advice. Just, "Can I do it in Claude Code?" And Claude is very good at understanding how Claude Code works, and it will kind of help you figure out how to do that. And so the more you're able to do this, the more you'll build these workflows. And every time, the first time it'll be done from scratch, but the next time you do it, you'll already have that workflow from the first time you did it. So just really have, like, a mindset of experimentation, and know that the first time it won't be perfectly smooth. We didn't have everything work here perfectly today. But even just right now from that demo that we just tried, now it actually just worked. Now Claude is saying hello. So it might not work perfectly the first time, but Claude is kind of there to help you, and just keep that mindset of experimentation and keep trying to see how much more you can do because really, these tools are just gonna get more powerful. They're just gonna get more powerful. We have Claude 4.5. By the time you see this, there might be Claude 5 or Claude 5.5, and it's gonna be even more competent. And this is just the way that work is gonna happen in the future. So I would say if you haven't started using Claude Code, start using it now. And if you aren't using it for everything, see if you can use it for everything.

    9. AGAakash Gupta1:18:25

       That's our challenge to you guys. Try to do Claude Code for more. And if I'm honest, you know, I don't use Claude Code for everything. I use the UI for a lot of stuff, but I use Claude Code for all of my big tasks, my tough tasks, the ones where I want to connect into various MCP servers, the ones where I want to use various context files, the ones where, like I said, I want it to pull down 200 web pages. So for anything powerful you want to do, try to use Claude Code. Bring that into your way of working because the people who get ahead of these tools, the people who are really learning how to use these tools first, first of all, all your companies are now evaluating you on how you use AI, so you're gonna do well on that. Second of all, you're gonna become known as that person amongst your PM team, your product leadership. So when your product leaders have a question about Claude Code, they'll start coming to you. And then finally, you yourself will become more productive. Some of these things, as we saw, they don't make you more productive. So I want you to be careful about where you use it, but figure out the intuition for when you should and shouldn't use it. Bring Claude Code in. My guess is the way PMs work, what they spend time on five years from now will look nothing like what they're doing today, which already looks nothing like what it was five years ago. So stay ahead of this. And Carl, thank you so much for bringing us along the bleeding edge of it.

    10. CVCarl Vellotti1:19:36

        Yeah. Thanks for having me. It was great.

    11. AGAakash Gupta1:19:38

        Bye, everyone.

17. 1:19:39 – 1:20:24

    Outro

    1. AGAakash Gupta1:19:39

       I hope you enjoyed that episode. If you could take a moment to double-check that you have followed on Apple and Spotify podcasts, subscribed on YouTube, left a rating or review on Apple or Spotify, and commented on YouTube, all these things will help the algorithm distribute the show to more and more people. As we distribute the show to more people, we can grow the show, improve the quality of the content and the production to get you better insights to stay ahead in your career. Finally, do check out my bundle at bundle.aakashg.com to get access to nine AI products for an entire year for free. This includes Dovetail, Mobbin, Linear, Reforge Build, Descript, and many other amazing tools that will help you as an AI product manager or builder succeed. I'll see you in the next episode.

Episode duration: 1:20:34