Check out the conversation on Apple, Spotify and YouTube.
What We’re Covering Today (0:00)
Aakash: My first episode with Carl Velotti got 30,000 views. It’s been my most popular episode of the last couple of months. So we have Carl back for a masterclass in advanced Claude Code usage.
Carl: Ever since our last episode about 2 months ago, I’ve continued to use Claude Code, continued to explore more of its 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.
We’re at the point now where every time 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. There are certain things where it would have taken me 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.
Why Claude Code Hit $1 Billion ARR (4:11)
Aakash: There’s this crazy news about Claude Code—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?
Carl: I think what it really comes down to is, you have the main frontier labs—OpenAI, Google, and Anthropic—and they’re all taking their own bets on how to approach using their LLMs.
For OpenAI through ChatGPT, you really see general usage. It’s good for search, it’s good for images, it can do some coding. So they’re much more general.
For Google, they really have taken big bets on being the absolute king of multi-modal. They can ingest video, they have the best image generation now with Imagen 3. But they haven’t really gone in-depth to the super deep workflows around coding.
With Anthropic, they’ve really focused on how can this be the best tool for coding and for work. I spend all day in Claude Code because it can do more and more parts of my workflow. I don’t need a general tool and I don’t really need it to be multi-modal. I need it to be really good at text and code.
What they’re getting is maybe fewer users, but they are absolutely just power users. As soon as you start using it, it almost becomes addictive to where you just keep doing more and more stuff in it.
Aakash: I can say from my personal usage, after our episode, I had to switch to the Claude 5X plan, $100 a month. Then I started using Claude Code and I hit the 5X limits quickly, 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. I told it to pull all 90 newsletter articles I wrote this year and analyze how my own style has changed. Normal Claude choked at it. Claude Code said “all right, let’s do it.”
Opus 4.5: The AGI Moment (8:00)
Carl: Another thing since our last podcast, Opus 4.5 came out, which is what I’ll use today for our demo. A lot of 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 walk away from your computer and it will just keep going forward in the task. If it hits a block, it’ll find a way forward.
Aakash: 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?
Carl: The big difference between a beginner Claude Code user versus an advanced Claude Code user is that for an advanced user, you’re really paying attention to what are the workflows that you’re doing multiple times, and then how can you define files so that you can go step by step and really define those workflows so that the next time you do that, it can run through all of that in one shot.
It’s just thinking about the whole system that you’re building with Claude rather than just having to do individual tasks.
Setting Up MCPs: Connecting to Your Tools (11:04)
Aakash: Should we start with MCP servers?
Carl: Yes. One thing to call out—everything we cover today stacks with that first video. The only thing I’ll mention is I pretty much always use Claude with the dangerously skipped permissions turned on because I just haven’t had any issues with it.
Aakash: Also known as YOLO mode.
Carl: The first thing I want to cover today is: 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 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 if I have stuff in Notion or Google Docs or Confluence, how do I bring that into Claude Code in a way that’s not just copying and pasting it?
The main answer is we’re gonna use something called MCPs. This stands for Model Context Protocol. It’s something that was invented by Anthropic, and it’s basically just the standard for how LLMs connect with services.
Setting Up the Linear MCP (13:17)
Carl: We’ll start with setting one up for Linear. Linear is a really modern version of Jira—it’s a way for engineers to manage their tickets.
A good mindset that you should have if you’re a Claude Code user is: I want Claude Code to be able to connect to this thing. You can ask Claude Code, “does Linear have an MCP server?” and it will search for you. Linear does have an MCP server.
[Shows setup process]
So, we just ran one command. Many companies have it set up this easily. When we come back into Claude, we usually have to restart Claude for it to pick up whatever changes there were to its settings.
Now when we go to MCP, we see that we’re connected to Linear. What’s cool is when you click on any of these MCPs, they basically come with lots of tools that Claude Code will automatically understand how to use. By connecting Claude Code to the Linear MCP, this is all of the tools that it gets.
There’s a ton here—you can get documents, you can get issues, you can create issues, you can get a list of projects. Just by connecting to this MCP we have 25 different tools that Linear has built to allow LLMs to work with Linear.
Aakash: This is basically step one. 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?
Carl: A good mindset is: I have to do this thing and I want to try to live in Claude Code all day. What are the MCPs that already exist for it?
Most of the work that a product manager does is gonna be in the document world. So Google Drive, set that up right away. Notion, Confluence—that’s the first one I recommend because that’s where immediately you’re gonna be able to do work on your computer and push it out so that people can see it.
Slack has a good one. All the major document systems like Notion and Confluence have their own MCP. There’s one for Google Search Analytics. Most companies, if they want an LLM to be able to interact with them, they’re gonna have their own MCP.
Aakash: As a PM you really need to care about insight and data from inside your company—Salesforce, analytics, BI, product analytics, user insights like Dovetail, your support desk like Zendesk or Intercom. Nowadays all these products have an MCP, so you should really think about connecting your Claude Code to all of those.
MCP vs API: Understanding the Difference (19:52)
Carl: MCPs are the new way for an LLM to interact with different tools, but there’s the more traditional way, and that is with an API.
For an MCP, you get all the tools automatically. An LLM right out of the box just knows how to use that MCP. Whereas with an API, it doesn’t know what to do, so it needs to have context around how to actually use that API.
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 the two ways that you can connect your LLM to the world, and they’re good for different use cases.
Skills: The Secret to Powerful Workflows (27:51)
Aakash: So we’ve been hearing from a lot of people, skills themselves are amazing. What are some of the things you use skills for?
Carl: For this podcast, I have a podcast production guide skill which is my lifesaver. 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. I have a skill for creating takeaways from a podcast using a transcript. I have a skill for coming up with new newsletter topics.
Aakash: It didn’t take as long as you might think. Usually what I’ll do is give like 2 or 3 examples. Here’s version A and version final 3 times. Notice how I edited it, how I improved it. Take that down, then you do it, and you can just run vibe checks like 2 or 3 times. I’ll just have it edit 2 or 3 posts. I’ll edit the skill 2 or 3 times, and it’s pretty much good to go after that.
Carl: That’s awesome. And it’s cool because this stacks too. If you had agents, you could give the agents access to different skills, and then when they’re working, they’ll have those skills.
Where to Get Skills (32:39)
Carl: There are a couple of different places that skills can come from. There are a lot of these—people share them all the time on Twitter, on LinkedIn, there are big GitHub repos with tons of skills.
But you have to warn everybody: you guys are very susceptible to prompt injection hacks if you’re downloading a skill and running in dangerously skipped permissions. So be really, really careful about that.
Aakash: Good tip. Maybe it’d be good to even have Claude investigate your skill. Maybe you could build your own skill that’s a skill investigator whenever you add a new skill.
Carl: Literally not a terrible idea.
The Skills Marketplace (35:05)
Carl: Another thing you can do is Anthropic has built what are called skills marketplaces. This is a new advanced Claude Code feature called plugins. You can discover plugins—these are bunches of packs of skills that companies have built.
There’s one that’s really good that is built by Anthropic themselves called document skills. So we’re gonna /plugin install document-skills.
This is a collection of document processing including Excel, Word, PowerPoint, and PDF. If nothing else, I definitely recommend at least getting the Anthropic official skills because these are things that as Claude works, it will automatically create pretty nice things for you.
[Sponsor Break: Amplitude – 9:37]
[Sponsor Break: Pendo – 10:16]
End-to-End Demo: From Survey to Tickets (22:21)
Carl: We’re gonna work with a fictional company called Taskflow. You can think of it as a to-do app, and we’re gonna work through a feature for a calendar integration.
We’re gonna start with user research. 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’ll 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 engineers start that work.
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.
Step 1: Creating the User Research Survey (22:26)
Carl: We’ll create a user research survey for a calendar integration based on our feature context document, and we already have a template.
[Claude processes and creates the survey]
Now the cool thing—let’s see this in Google Drive. So it gave us the link here. One thing, whenever it pushes to Google Drive, it pushes straight markdown. So kind of a pro move here is you can cut it and then paste from markdown and it will be nice.
Aakash: Whoa.
Carl: 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. You just right click and then do paste from markdown and it will take that and make it really nice. Now we have our user research survey.
So the first time we’ve seen it, we’ve actually pushed something out into Google Docs.
Step 2: Analyzing Survey Responses (24:41)
Carl: Now let’s say we send this out and we get responses from it. I have already created sort of a response—we got some survey responses from this.
There’s sort of a step in between here where you’d actually want to draft that survey in a way that users could respond to it. Depending on the thing that your company was using, Qualtrics is a very common user research platform.
Aakash: So then you could basically automatically create the survey in MCP once you’ve gotten the feedback from a UXR team.
Carl: Exactly. There is in fact a Qualtrics MCP server. So that is the mindset. It’s like, OK, 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.
[Claude synthesizes the survey responses]
Step 3: Creating the PRD (44:21)
Carl: So now let’s go ahead and create a PRD based on those survey results.
Aakash: Step 1, you send out your survey for feedback to Google Docs. Step 2, you had to create that survey in Qualtrics using Qualtrics MCP again in Claude Code. Step 3, you have it analyze the results and now step 4, create a PRD.
Carl: And probably in this case, it would make sense to potentially have a PRD writing skill which would have your template in it. It just makes it nicer to use so you’re not having to remember what was the name of this markdown file that I had with the template.
[Claude creates the PRD]
Now we have our calendar integration PRD. It’s the moment of truth. Is it any good?
Aakash: That’s Opus 4.5 with no instructions, so much better than GPT-4.
Carl: Yeah. And it’s pretty nice. I mean, I’d have to read this more carefully, but if you give it a lot of context, then these things can be actually pretty good on their first pass. It’s really showing that user research in here, which 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.
Step 4: Creating the Presentation (50:46)
Carl: Now we need to get ready to actually present this to stakeholders. We’ll go ahead and create a slide deck.
[Claude creates PowerPoint presentation]
Presentations take quite a long time to make, which is where hooks come in handy for notifications.
[Shows completed slides]
Aakash: I just—this is my entire career. I would have to spend time, we’d create a PRD, then we create a deck. It would take an hour or two, just turn that into a prompt.
Carl: And if we had made this a proper skill that knew about your template and how you like your slides, it would be even closer to ready. This is a real presentation, this is all editable. That’s why it was taking so long—it was defining all these things in HTML to see how big they should be.
Aakash: I mean, did we just kill Gamma for PMs with this podcast episode?
Step 5: Creating Linear Tickets (57:49)
Carl: The final step is, let’s create the tickets. Creating tickets as a product manager, sometimes engineers do it, sometimes PMs do it, but in general, it’s always such a painful process because you know the information and you just have to go through the UI and click all these buttons and make sure everything’s all organized. Whereas now with Claude, it can just literally create all those tickets for you.
[Creates tickets in Linear]
So 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.
Aakash: Wow.
Carl: Let’s look at one of these P1s. OK, overview, requirements, technical considerations, and then the acceptance criteria. These are still slightly high-level. I think that’s because the PRD itself was high-level. This would be a great thing that you could hand off to your engineers where they might actually add more information.
As an engineer, you’re really getting a lot of that information around what exactly do you need to be able to build.
[Sponsor Break: Jira Product Discovery – 30:01]
[Sponsor Break: Miro – 30:56]
GitHub Integration: Working Remotely (1:07:08)
Carl: The very last thing, and this is very cool, is actually connecting 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.
Let’s have it sync everything up. All of this code and everything is already in GitHub. If I were to download this project from GitHub, it would just have all of this information in it.
By pushing it to GitHub, 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 transfer work between different computers.
Adding Claude as a GitHub App (1:10:10)
Carl: The last thing we’re gonna do is we’re going to add Claude as an app. You go to github.com/app/claude. We’re gonna configure this. In order for Claude to be able to work in here, you have to add your API key.
So now, Claude has been added to this repo. We can open up a new issue. We’re gonna make a comment @claude and now Claude has been assigned that. I could be, and I could do this from my app, I could do it from anywhere else.
Real-World Use Case (1:12:23)
Carl: A real-world scenario: let’s say you as a product manager have your PRD on your own device locally and also in Google Workspace. But let’s say you’re at lunch and you’re talking to your engineer and something comes up where it’s like, oh, did we consider for that calendar integration feature time zones? Oh yeah, I forgot, Arizona actually doesn’t have daylight savings.
Normally, you just have to remember when you’re back at your desk, start a Slack conversation about it. But what you could do instead is you could go into your GitHub repo on your phone. You could create an issue that says, make sure that we account for time zones 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 pull it back down.
Aakash: One thing I personally use it for is I have my whole journal on GitHub in a private repo and I will just open an issue that’s like a voice transcription of my day, and then I’ll @Claude, and then Claude will add that to my journal for me.
Carl: There’s all these different things you can do when you are away from your computer with Claude accessible in the cloud. There’s step one of 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 in your repo.
GitHub Basics for PMs (1:14:00)
Aakash: Can you just give them the basics of using GitHub? What do they need to know about these terms like pushing and pull requests and branching?
Carl: GitHub is just the best place to store your files. It’s the easiest way for Claude to work with these files. I was reading something and it said that Claude is fiercely competent with Git. And Git is the language of using GitHub. Literally anytime you need to do anything with GitHub or Git, you can just ask Claude and they will basically do it correctly.
The main things to know: we have our repository with all of our files. If we want to get those files onto our own computer, then we’re gonna pull them. If we’ve done work on our computer and we want to go back into the repo, then we’re going to push them. That’s the absolute basics.
The one other thing to know is if you’re prototyping or building your own feature, you might want to open up a branch. What that is, is it’s basically a branch of that code that has new stuff and you can push the branch, but that won’t make it go to whatever is actually deployed. You can test that branch, and then once you said this branch looks good, then you can merge it into main, which is your main repository.
But for most use cases of product managers, you 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.
Aakash: So it’s 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.
Key Takeaways (1:16:57)
Aakash: So, if we had to summarize, 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 codebase. Is there anything else people should be exploring?
Carl: I think the main thing is just around everything that we’ve covered today. I hope it was a good 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.
Anytime you’re starting a new task or something else, just see: can I do it in Claude Code? That’s my best advice. Claude is very good at understanding how Claude Code works and it will help you figure out how to do that.
The more you’re able to do this, the more you’ll build these workflows. Every time the first time it will be done from scratch, but the next time you do it, you’ll already have that workflow from the first time. Just really have a mindset of experimentation and know that the first time it won’t be perfectly smooth. It might not work perfectly the first time, but Claude is there to help you.
Final Thoughts (1:18:25)
Aakash: 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 5 years ago. So stay ahead of this.
If I’m honest, 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 I want it to pull down 200 web pages.
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. When your product leaders have a question about Claude Code, they’ll start coming to you. And finally, you yourself will become more productive.
Figure out the intuition for when you should and shouldn’t use it. Bring Claude Code in.
Carl, thank you so much for bringing us along the bleeding edge of it.
Right, everybody, see you later. If you want to learn more about how to work this way, check out our full conversation on Apple or Spotify podcasts. And if you want the actual documents we showed, the tools and frameworks and public links, be sure to check out my newsletter post with all of the details.
Finally, thank you so much for watching. It would really mean a lot if you could make sure you are subscribed on YouTube, following on Apple or Spotify podcasts, and leave us a review on those platforms. That really helps grow the podcast and support our work so that we can do bigger and better productions. I’ll see you in the next one.
