Build a Growth Operating System in Claude Code, Step by Step

1. 0:00 – 5:21

   Intro

   1. JMJames McAulay0:00

      AI has just completely unlocked what growth people can do. We no longer need to wait for engineers to build growth experiments. You can simply ship to production with permission from your engineers yourself. We're seeing these full-stack growth hires achieve incredible amounts as a single person in a startup. I guess you could call this like, yeah, an agentic growth operating system. You could use this if you were a full-stack growth person, first growth hire. If you're gonna be doing work inside of Claude, I think this operating system will, like, help you get better growth results, and it's my sort of take on how to work with Claude as a growth person.

   2. AGAakash Gupta0:33

      Meet James McAulay, who helped ElevenLabs double from a hundred million to two hundred million ARR in nine months and helped AI startups like Jack&Jill and Wonder Studios with their growth.

   3. JMJames McAulay0:43

      When I was at ElevenLabs in twenty twenty-five, we were definitely ahead of the curve with AI, but actually our launch processes were mostly custom GPTs, and that's because skills hadn't been formalized across the AI ecosystem.

   4. AGAakash Gupta0:54

      How the hell did you scale up this business to 100K revenue in just four months?

   5. JMJames McAulay0:58

      So my first startup was a marketplace. It actually took us six years to reach a hundred thousand pounds of monthly revenue. My current business hit that goal in ninety-five days. I haven't hired a single full-time employee.

   6. AGAakash Gupta1:10

      So you've taught over four hundred people now. What do most people get wrong as they're learning this stuff? Before we get into today's show, please take a second to check that you're subscribed on YouTube and following on Apple and Spotify podcasts. If you want access to all of my favorite AI tools, I've gotten them to give you an entire year of their paid plans. Check out bundle.aakashg.com for an entire year of Bolt.new, Airtable, Speechify, Descript, Magic Patterns, Linear, Dovetail, Arise, and Mobbin'. And now, into today's show. We've been talking a lot about how you use Claude Code for product work. But what about all the important growth work? SEO, ads, product marketing. These are really important skills, and if you are on a growth team, you might have been wondering, "How do I apply everything in this channel?" If you're a PM at a startup who has to handle growth work, you might have been wondering, "How do I apply this to my growth work?" Today, I've brought in an expert, somebody who's not only grown his own business to a hundred thousand dollar monthly revenue, but he has helped grow other companies. James, welcome to the show.

   7. JMJames McAulay2:24

      Thanks for having me. Really excited to be here.

   8. AGAakash Gupta2:26

      How do people know what they're gonna see right now actually works?

   9. JMJames McAulay2:31

      Yeah. In a nutshell, I ran my own company as CEO for ten years, and I was responsible for growth there as the CEO is. And then after we were acquired, joined ElevenLabs and spent a year at ElevenLabs. And for me, that was a masterclass in growth, uh, one of the fastest-growing AI companies in history. And I learned a hell of a lot while I was there. After I left, I sort of spent a month going deep on Claude Code. I did some growth consulting for a few other AI companies, and now I'm running my own business teaching people how to use AI agents to grow faster. So I think my track record checks out, and I think I've, I've seen enough companies, and I've seen a company like ElevenLabs from the inside, uh, to, to hopefully know what I'm talking about.

   10. AGAakash Gupta3:15

       So what are we gonna learn today?

   11. JMJames McAulay3:16

       So I'm gonna show you the system I use to run my business. Twelve months ago, I was still using Claude Chat, like a lot of people. But today I run everything inside the terminal, and I'm gonna walk you through the system that I'm using. I'm gonna show you some skills. I'm gonna show you some tools that I use, some growth experiment discipline for anyone to kind of approach growth in a, a rigorous way. Um, and also this system I use for making sure that I'm, like, actually working towards what really matters. I think with AI, it's possible to get distracted. It's possible to go down rabbit holes, build anything. Uh, but I've sort of deliberately engineered this whole harness so that I'm constantly kept on track, and Claude or any agent that drops in here is sort of laser-focused on the growth goals. So, um, I've been using Claude Code for this, and if you go into the link at the bottom of this video, there's gonna be, like, a way to download this. It'll be a GitHub repo, uh, and you can just install this in, like, a couple of commands. This works in Claude Code, Codex, Hermes, OpenClaude, if anyone's still using OpenClaude. Um-

   12. AGAakash Gupta4:22

       [laughs]

   13. JMJames McAulay4:22

       But I guess, yeah, twelve months ago, I was, you know, like a lot of people in Claude Chat. I had a project, and I started putting, like, context into this project, and I realized that with the right context, as we all know by now, um, an AI agent can be incredibly helpful. And so I'd heard about Claude Code. It was pretty new at the time, but pretty much this time last year, after I left ElevenLabs, I, I actually, like, flew to Mallorca, uh, for a week, went cycling, and then when I wasn't on the bike, I was just in the terminal, like, every, every waking hour for, like, two weeks, uh, just becoming obsessed with this stuff. And so I'll show you, like, the system that I've spent basically a year tweaking that I run my business through, um, you know, with, with goals and also with skills and with tools. And hopefully, by the end of this video, you'll know how to put it to work, uh, and why I think it's, it's worth implementing either my system or something very similar to this. Aakash, where would

2. 5:21 – 6:52

   The Growth Agents repo

   1. JMJames McAulay5:21

      you like to start?

   2. AGAakash Gupta5:21

      I guess, how do we set it up? What's the scaffolding here that makes it all work?

   3. JMJames McAulay5:25

      Okay. So, um, I've packaged this up as a GitHub repo. It's called Growth Agents, and there's, like, a couple of, like, instructions down here that will get it running. Essentially, what this is, is I've packaged up, like, some of the tools that I think are incredibly helpful for, for working agentically. Um, I only work with tools that have, like, good APIs or good CLIs. Um, so everything sort of feeds back into whichever agent you're using. I've also packaged up a bunch of skills. We can go through some of these. These are skills genuinely that I use, you know, every day. Uh, morning standup skill, monthly retro, but also, like, a good PRD skill, an SEO audit, et cetera. We can get into that. So it's, it's a collection of skills. It's a recommended set of tools. I've got some, uh, weekly routines or, like, cron jobs, depending on how you want to set it up, that basically improves the harness every week. So it looks at the way you're working, and it suggests new skills, or it goes looking for new CLIs. So- I guess you could call this like, yeah, an agentic growth operating system. You could use this if you were like, you know, a full stack growth person, first growth hire. If you're a quite specialized product marketer, SEO lead, um, you can basically onboard to this system, and it will ask you about your goals and then sort of customize all of it to you. Maybe we start with this 'cause I actually think that this is quite interesting unless you wanna go somewhere else, Aakash.

   4. AGAakash Gupta6:51

      Let's do

3. 6:52 – 7:21

   The goal cascade

   1. AGAakash Gupta6:52

      it.

   2. JMJames McAulay6:52

      Okay. So this time last year, I sort of wrote down, you know, my Q4 goal, and I edited the project instructions in Claude and said, "Everything we talk about has to move me towards that goal." And it was really helpful. Like, Claude told me when I was getting distracted, helped me prioritize, uh, really helped me with, like, career decisions, actually. You know, I was tempted to go back into full-time work, but I had told Claude that I wanted, you know, this freedom to run my own company again.

   3. AGAakash Gupta7:18

      I've been using Ari from Ariso for a while

4. 7:21 – 12:01

   Ads

   1. AGAakash Gupta7:21

      now, and the thing that hooked me isn't one feature, it's the memory. Most AI tools forget you the second chat ends. Ari doesn't. It connects to my email, calendar, Slack, and it remembers everything. Every commitment I make, every thread I'm on, every person I've talked to and what we discussed. It tracks and organizes my work automatically, and it shows up before every meeting with exactly what I need. I never walk in cold anymore. That's the individual side. The bigger side is what Ariso does inside companies. It coaches managers, keeps teams aligned with the actual strategy, and turns all that scattered activity, the meetings, messages, and docs into a clear picture of how execution is really going. Leaders see drift before it costs them a quarter. One brain for you, shared clarity for the whole org. Try Ari at ariso.ai/aakash. That's A-R-I-S-O.A-I/A-A-K-A-S-H. Every growth team has a funnel they should be watching more closely. Usually, it's checkout. Amplitude custom agents will watch that funnel for you on the schedule you set. Let me show you. You start by describing the job. Watch my checkout drop off. The agent builder will ask a few questions to shape it, and then the agent connectors plug it into your stack. That part matters. The agent reads the PRD, the tickets, the team conversations, so its analysis actually carries the full context of what's happened. From there, it monitors the funnel on a schedule, and when it spots a problem, it acts. Right here, it files the linear ticket itself. The loop between knowing and doing closes on its own. That's the shift. Set up the agent once, and it runs the work. Custom agents are free for all Amplitude customers in open beta right now. There's a library of pre-built agents too, including the exact checkout drop-off investigator. Start at amplitude.com/ai-agents. The agent library is linked right there.

   2. JMJames McAulay9:11

      And Claude helped me, like, hold the line. Um, and so this year, you know, I've got a big punchy goal for my company, and this goes into, like, a mission file. I then run quarterly OKRs. Don't need to explain those. I'm sure everyone working in product knows what OKRs are. Um, and again, these are documented in the system. They tie back into the mission. I then run a monthly retro, and so at the end of every month, Claude pulls data from all the tools that matter. For me, it's Stripe, PostHog, Xero, Attio, and basically sort of ladders this up and says, like, "Did you hit the OKRs?" And like, "Are we on track for the mission this year?" And helps me sort of like plan the following month. But then week to week, I'm also doing like a Friday retro. And I'm, I'm not joking, I spend about an hour on these weekly retros, usually just with Whisperflow, just kind of like speaking into Claude about what went well this week and what didn't. And you'll notice here I've said that it takes my personal reflections on, like, how I feel the week went, and then it takes quantitative measurements from all the tools that are connected to it. So instead of me saying, "Oh, I feel like, you know, sales were maybe a bit slow this week," it just goes to Attio and says, "We're on track," or, "We're not on track." If I say, "Uh, I'm not sure if the, you know, main landing page is converting," Claude just, like, gets the hard, cold numbers. So I do that every week, and then I have a skill called daily standup. You know, if you're in a company, um, I think daily standups, if they're done well, if they last about five minutes, and if you're actually standing up, they're just really helpful. But when you're working by yourself, um, you don't have anyone to stand up with. So instead, I do my daily standup with Claude, and it sort of checks that I'm working towards, you know, the biggest goal this week, uh, which ties into the month, which ties into the OKRs. So this kind of cadence, I find really helpful. Um, you know, growth is all about prioritization. We do not wanna get distracted. We don't wanna waste time running the wrong experiments. And so I think this system is basically what's kind of like keeping me or you, if you install this, on track. And I've set up a session hook so that when you start Claude, every time, it's basically reminding you of your goals. So these are-- everything I'm gonna show you is sort of fictitious numbers, just as an example. When you start Claude, instead of getting distracted and going off and using Astra to build some 3D game that doesn't matter, [laughs] uh, it's gonna remind, it's gonna remind you, right? That like, "Look, here's the mission, and this is what you said we would work on this quarter," and it's gonna push back. Um, basically, when you install this, your CLAUDE.md will get, um, sort of customized. And so if Claude sees you getting distracted, um, it's gonna sort of pull you back on track. So yeah, I guess you could think about this as having your own, like, growth lead who is just, you know, repeatedly saying, "Is this a 10X bet? Is this going to move

5. 12:01 – 17:04

   Are AI retros actually useful?

   1. JMJames McAulay12:01

      the needle? Why did you get distracted this week?" Is there anything you wanna get into on, on any of that?

   2. AGAakash Gupta12:05

      Well, I have to admit, I have tried this, the weekly review, the daily standup. I haven't tried the quarterly, monthly part of it, but what I found was that, like, AI would, like, miss certain context about something that changed rapidly, or I'd have to be correcting it. And so unlike a human, where I felt like the standup and the retros were useful for alignment, I have kind of stopped using these skills, just sticking to execution. But you're not the only guest who has emphasized how important these are. So can you convince me, like, why should I be using AI as a thought partner for all these retros and standups?

   3. JMJames McAulay12:46

      I have iterated on these skills. You know, the one I use most often is weekly retros. Like, I sometimes forget to do a standup. I'm gonna be honest, like the standup is maybe like the sort of, uh, least valuable one, and I often forget to do it. But the weekly retro, I am religious. I think I've done like ten, fifteen of them over the last ten, fifteen weeks. So For me, it basically notices patterns, and I'll give you an example. Like a lot of technical or engineering-minded people, I, I sometimes overbuild. I'm, I'm sometimes, like, keen to just go and add more features, and that's not what is actually gonna move the needle. And so Claude noticed after, like, two or three retros, because it can see my to-do list, I've got it connected up to Todoist. It could basically see that, like, there were certain outreach or certain, like, marketing tasks that were just stale every week. And then it would look at GitHub, and it would say, "Hang on, you've, you've shipped so much that, like, we did not need to ship. You've, like, refactored part of your platform for no reason other than it was fun to refactor." So I guess for me, I think it's helpful just in, like, spotting blind spots and, and basically being held to account on what you say matters. 'Cause I, I do think this is one of the biggest risks with AI now, is that we can just get-- we can build anything, right? Like, it's not an exaggeration to say that if you or I sat down on, you know, tomorrow morning, we could have a working sort of MVP or, like, a feature by the end of the day. But if nobody wants that feature, if nobody wants that product, we've wasted our time. Um, so I think it's, I think staying on track and I think, like, spotting your own patterns of maybe avoidance or procrastination or doing fun stuff instead of doing the stuff that's really gonna move the needle. I have found-- personally, I've found Claude very good at, like, holding me to account on that.

   4. AGAakash Gupta14:34

      Okay. Well, maybe I'll have to give it a try with these updated skills. Maybe my skills just weren't very good. And maybe you can show us later what the output looks like.

   5. JMJames McAulay14:43

      Yeah, of course.

   6. AGAakash Gupta14:44

      All right. So how do we spin up, like, our first agent?

   7. JMJames McAulay14:47

      Okay, so let's try a live demo. I've basically set up a sort of folder of context for a fictitious product marketer. Um, let's get into Claude.

   8. AGAakash Gupta14:56

      And you're in iTerm2. Why iTerm2?

   9. JMJames McAulay14:59

      I just quite like the, some of the themes in iTerm. Uh, it feels like quite a fast and snappy terminal, but I mean, it's basically the same, it's basically the same as Terminal. I'm gonna get people in the comments saying, "It's not the same as Terminal." But, um, I think I installed it, like, fifteen years ago when I was, uh, doing computer science at uni, and every computer I've had since, I've just installed iTerm and then moved on to iTerm. Um, so-

   10. AGAakash Gupta15:22

       Have you tried like a VS Code or something? Like how should people be thinking about-

   11. JMJames McAulay15:26

       Yeah. Yeah, yeah

   12. AGAakash Gupta15:27

       ... where they run Claude Code?

   13. JMJames McAulay15:28

       When I was learning Claude Code last year, I was doing it inside Cursor, and so I had, you know, my files in the middle, agent on the right. And maybe I'm gonna get hate for this [chuckles] from engineers as well, but I realized very quickly that I wasn't reading the files. Like, I was setting up tests, I was inspecting all the front end. I was basically managing, like, the coding agent, um, but trusting that, like, with the right testing harness, you know, it couldn't ship something that would break. And eventually I thought, well, I have this full screen, right? I've got my, like, full desktop screen, but I'm only using this, like, column on the right. I'm ignoring all this. Why don't I just run Terminal as my sort of like preferred way of running Claude Code? I also think that the Cursor pricing-- I was running Claude Code through Cursor, and I, I remember very quickly, you know, they were trying to move me up the pricing levels. And I think just running Claude Code two hundred dollar a month plan inside the terminal was, was the fastest way to do it. And so I'm gonna show you a demo here, right, where I've got, like, one screen. I've got this nice, uh, plugin called Claude Heads Up Display. Just keeps an eye on my context, reminds me which branch, which work tree I'm in. But when I'm at home, I have, like, super wide screen, and I will often have, you know, like two, two windows like this, uh, or maybe five. [chuckles] It's definitely not the most, like, aesthetically pleasing way to, to work with agents, but I find it just, like, snappiest, fastest. I, I don't like the Claude Code desktop app. I find it-

   14. AGAakash Gupta17:00

       [chuckles]

   15. JMJames McAulay17:00

       ... just a little bit clunky compared to this.

   16. AGAakash Gupta17:02

       Cool. So let's get this agent spun up.

6. 17:04 – 21:47

   Live onboarding demo

   1. JMJames McAulay17:04

      Okay, so hopefully this is gonna work. I've set up, like, a demo folder. It's got a bunch of context on, like, what this person's trying to achieve. I've designed this in a way that it will interview you about your goals. It'll ask you where your goals are. You know, are they written down somewhere in Asana? Are they in Google Docs? Are they in your head? And it will try to sort of like crystallize all of this for you. Um, and then basically, um, we'll sort of pressure test your goals and set up like a folder for growth experiments. If I come back to here while this is just working. I've set up a bunch of rules around, like, how to run growth experiments. Um-

   2. AGAakash Gupta17:41

      Hmm

   3. JMJames McAulay17:41

      ... I don't know if you've tried, I don't know if you've tried this, but I've actually found that Claude is a very good, uh, sort of analyst of data. And, you know, we want to make sure that our A/B tests are, you know, not only statistically significant, but, like, set up correctly. We wanna make sure that we don't prematurely read the result of an experiment. Um, so I've sort of packaged all of this up into, like, a set of rules, and then I've also set up this, like, loop that basically makes sure all your experiments get closed. So I don't know if this happened to you, but it's, it's very easy to start a growth experiment and then get distracted by shiny object syndrome and basically forget to close it. And so, like-

   4. AGAakash Gupta18:20

      Yeah

   5. JMJames McAulay18:20

      ... you try something, and you never close the loop. Um, or, you know, you get to the end of the year, you have run like a bunch of growth experiments, but they're not documented. So, you know, you've got a bunch of PostHog experiments, but you haven't actually documented the learnings. So this system is constantly asking you, "I'm looking at the data," you know, "Can we conclude this? What did we learn? Let's document it. Let's, like, use that to inform the next growth experiment." So it's trying to, like, it's trying to hold you to account, basically. If you start something, make sure you finish it and make sure that whatever you learn is documented so that the next experiment is better.

   6. AGAakash Gupta18:58

      How often do we forget that, right? [chuckles] It's so easy, and that's actually where-

   7. JMJames McAulay19:02

      It's-

   8. AGAakash Gupta19:02

      ... that alpha is, is figuring out, okay, this one failed But here's why. And like you said, I think Claude is pretty good at having like a rigor around these experiments. Like once I ran an experiment, it was like, "Oh, wait, but here's this confound, so we have to actually look at this media mixed model in order to look at it." So it's actually pretty smart at that point.

   9. JMJames McAulay19:22

      It's-- Yeah, it's really good. Um, I've seen this play out, you know, in my first company. We'd run a bunch of experiments, and we wouldn't document them very well. And then we'd hire someone new, and they would say, "Oh, I wanna experiment here, you know, this part in the funnel." And we'd say, "Ah, sorry, like already tried that." And they'd say, "Cool, like where can I see the results?" We'd be like, "Um, [chuckles] well, we can tell you that it didn't work, but we, we don't actually have it documented." So, um, I think with AI now, there's kind of no excuse for, for not documenting stuff. Okay, so we've, we've run this, we've run this setup. Um, and basically the agent has sort of looked through the folder that it's been like started in. So I imagine most people watching your show have got their own, maybe it's a second brain, maybe it's like one folder on their computer that's got all the context they need. But I'm not trying to like get anyone to create a new folder. This is basically gonna like run inside your existing folder, and it's read through, and it's saying, "Okay, here I can see that you've got some OKRs already. This is good. You know, you want six million net new ARR. Um, you've got some OKRs on Q2, but I don't see anything for Q3. I don't see anything about like what your focus is this month. Uh, you've written some notes about this week and maybe some Q4 planning. Uh, couple of tests I can see that you ran, uh, some information on the business, an export from product analytics." Uh, so it's like done an audit, and it's like, "Okay, cool. I have like some sense of what you're trying to achieve." Uh, so now it's asking me like, "Do you wanna do kinda rapid setup where I just like ask you quick questions, or do you wanna do just like full setup where I interview you about like everything to do with your role, and I get like full context?" Um, for the purposes of this demo, we should probably do rapid. But when I, when I teach people like how to use Cowork or Code properly, the first exercise I do is getting them to do an interview with Claude that takes up to like two hours, and it feels like a slog. It feels, you know, like you're just answering loads of questions. But in doing that and having Claude basically sort of structure your context in a second brain, every answer, every question from there on out is better. I was just doing the full onboarding and just like switch on WhisperFlow, just talk out loud. Uh, but for, for the sake of the interview, let's do some rapid onboarding.

7. 21:47 – 26:51

   Ads

   1. AGAakash Gupta21:47

      Agents and assistants are browsing, shopping, and transacting alongside humans, while fraudsters use AI to build faster, cheaper, and more sophisticated attacks. Technical teams need device intelligence that evolves as quickly as these attacks do. Fingerprint is exactly that. Accuracy is the number one reason customers choose Fingerprint, because their device intelligence goes deeper than identification alone, resolving visitors with precision that holds up through spoofing and evasion attempts for months at a time. Fingerprint collects more than one hundred signals across devices, network, and behavior to build a single identifier, then provides signals like bot, agency, proxy, and VPN detection to give technical teams a complete picture of risk. They ship hundreds of product updates a year, so they adapt first as browsers update, privacy requirements change, and fraud tactics change. With one API call, teams can get this running in minutes. Over six thousand companies rely on Fingerprint every day. Join them now at fingerprint.com. I used to think I had a retention problem. Turns out I had a messaging problem. I was sending the same onboarding emails to every new user, whether they activated on day one or never logged in again. I had no idea who was slipping or why. Customer.io changed that. Every message I send is now based on what users actually do in the product. Someone hits a key activation moment, they get nudged to the next one. Someone goes quiet, they get a different path entirely. Their AI agent makes it fast. I describe the campaign I want, and it builds the full journey for me: triggers, timing, copy, even branching logic. And when I want to know how something is performing, I just ask the agent directly, and it tells me what to do next. They also have an MCP server, which means AI tools like Claude can see directly what's happening in your Customer.io workspace. Your segments, your customer data, your attribution, all of it. So instead of explaining your business context every time you need help, Claude already knows it. Notion used Customer.io to personalize their onboarding and hit nearly fifty percent open rate, improved conversion by six to seven percent with localized campaigns, and pushed open rates up another twenty percent through AB testing. The idea is simple. Customer.io helps you deliver more impact from every message you send. If you're a PM or founder and your onboarding is still one size fits all, try Customer.io at customer.io. Imagine learning AI product management, AI product strategy, AI product leadership, advanced PM with Claude Code, all from frontier leaders at OpenAI, Anthropic, and Google for less than ten dollars a day. Actually, you don't have to imagine it. That's exactly what Product Faculty's AI Builder Fellowship gets you. One yearly membership, six live cohorts. Not recorded videos or another content library you never finish. Real live sessions with frontier AI leaders, hands-on AI build labs, executive insight sessions, capstone projects, unlimited retakes, and continuous access to new programs as the AI landscape evolves. You'll also get every new certification they add while you're a fellow at no extra cost. The goal is simple: help you master frontier AI skills to become an AI native product builder and operator, not just someone who talks about AI. Purchased separately, these programs are worth more than nineteen thousand two hundred and forty-five dollars. The lowest priced certification alone starts at twenty-seven hundred dollars. But founding members can join the entire fellowship for less than thirty-six hundred dollars per year. That's less than ten dollars a day. Soon, the price increases to five thousand dollars a year. So if you're serious about becoming the person your company turns to for AI, this is one of the highest ROI decisions you can make this year. Join the AI Builder Fellowship at productfaculty.com.

   2. JMJames McAulay25:30

      Okay, so it's saying, yeah, you're a growth lead. I don't really know like sort of what your quarter's objectives are, and it's saying here, do these exist already? Like are these in Notion? Are they in Asana, Linear? Um, and then it's also noticed, you know, you exported some data from analytics, like is there a tool I can connect to? Um, so I'm just gonna say quick demo to make sure that it doesn't like go too deep into this, but like, uh, no Q3 objectives actually written down [chuckles] and, uh, data from PostHog. So yeah, it's gonna like piece together all of this stuff, and then you're gonna have-- Uh, let me just come back to here. Um, it's gonna say, "Okay, cool, like we got, we got these skills, you know, do you wanna do a daily standup?" Probably not. Um, and the skills I've like sort of put together, we've got these like content skills that I use all the time, like a tone of voice skill that sort of interviews you, asks you for examples of your tone of voice. I'm sure people have talked about that on your show before. I do a lot of like YouTube as you do, uh, for growth. So I've got a skill that basically does the package, the title, description, uh, the keywords, the research. I've got a skill that looks at competitors. You know, there's some stuff in there for anyone doing YouTube.

8. 26:51 – 32:00

   The SEO audit skill

   1. JMJames McAulay26:51

      Uh, you mentioned SEO at the start of the call, so I've basically remixed a really good skill by a guy called Corey Haines. Do you know Corey Haines? Uh-

   2. AGAakash Gupta26:59

      No.

   3. JMJames McAulay26:59

      So he has done, he's done this amazing, uh, set of skills called marketing skills. I've-- I definitely think you should get him on the show. Um, and his most popular one is SEO audit. Um, so I, I took this. I've sort of like put my own spin on it. I noticed, for example, that like if you ask Claude to scrape a website, it often hits like Cloudflare blocks or like doesn't, doesn't read the website accurately. Uh, so one of the tools that I've suggested installing for anyone doing SEO is Firecrawl, just like turn any website into an MD file. And then also Data for SEO, which is this like dirt cheap SEO API. I can't believe how cheap it is. Uh, you give Claude code the API key, and it can just research like any term for like a few cents. So I've like-

   4. AGAakash Gupta27:48

      Mm.

   5. JMJames McAulay27:48

      -kind of wired those into the SEO audit.

   6. AGAakash Gupta27:50

      So you don't recommend like a whatever hundred dollar Ahrefs subscription or something when you can just do it in cents for whatever you need from Data for SEO?

   7. JMJames McAulay28:00

      No. Uh, I've, I've been-- You know, my company used Semrush, and then at ElevenLabs I think we used Semrush. Um, but I noticed, you know, I've, I've done years of SEO work. They were basically like the same tasks. I just wanted to see like my performance over, you know, certain keywords. I wanted to know our keyword gaps. I was often wanting to like put together a keyword universe. I, I found it so clunky in those tools to actually kind of like put together a keyword universe. Whereas with Claude, I said, you know, "Analyze my website, analyze competitors, use Data for SEO to create a keyword universe, and then we will like attack the keywords that matter." So yeah. Do you have a website? Like do you have a, a-

   8. AGAakash Gupta28:41

      Yes.

   9. JMJames McAulay28:41

      -like a personal website?

   10. AGAakash Gupta28:43

       Yes.

   11. JMJames McAulay28:43

       What is it?

   12. AGAakash Gupta28:45

       www.news.aakashg.com.

   13. JMJames McAulay28:53

       Okay. Um, so yeah. I think this skill is amazing, and not only is it going to analyze, you know, your backlink profile, not only is it gonna look for basic stuff like are your H1s correct, are your links correct, um, it's also gonna do a page speed analysis. So I don't know how much you know about SEO, but there's a, a Google tool called Lighthouse, measures like how quick your website loads. You used to have to like go to PageSpeed, their website, to kind of like get a test, export the test, save it, give it to the agent, and then ask the agent to kind of like make the change. Uh, but this-

   14. AGAakash Gupta29:29

       Mm.

   15. JMJames McAulay29:29

       -uses the Lighthouse CLI. So it basically just like tests it for you, pulls all the data, and looks at, you know-- If it's inside your repo, so I'm doing a sort of like outsider test on your website, but if I run this inside my own code base, it immediately finds ways that we can like speed up our website. Uh, which actually matters quite a lot for SEO and, and helps with conversion. I think there was a study that said, you know, every, every second of load time that you can like reduce has this kind of like proven gain in conversion. Um, so anyway, I think this is like a really good skill. You can see here it's just kind of like crunching through. Maybe we can come back, or I can send you that afterwards. But this is one I-- This is genuinely one I use all the time. I used it on my own website a few months ago. I used it this morning. I literally deployed a new page and just ran SEO audit to check that again it was like done correctly. Uh, so yeah. One of my favorite skills if you pair it up with Data for SEO and Firecrawl.

   16. AGAakash Gupta30:30

       Mm. Can we go back to the agent that we spun up and that was pulling your context and just close the loop on how to build sort of like an autonomous growth agent?

   17. JMJames McAulay30:39

       Yeah. Of course. So I guess I wanna like manage expectations here that these are not like fully autonomous, uh, growth agents. Like-

   18. AGAakash Gupta30:49

       Mm.

   19. JMJames McAulay30:49

       -you can set up these skills to run on a schedule. You can take, you know, any of these skills, and you could create like an agent using something like Vercel's Eve framework, OpenClaw, Hermes. My system is not really like, you know, this agent is kind of like twenty-four seven sitting on my Mac Mini doing work for me.

   20. AGAakash Gupta31:11

       Mm.

   21. JMJames McAulay31:11

       It's like I am in the driving seat, and at any time I'm sort of like sending this agent off to do a piece of work that might take an hour, and then I'm like sending this one off that might take, you know, another hour. I've-- You know, for example, the SEO audit, I have a Claude code routine every Sunday that like tests my whole site and tells me if anything's gotten slower. So if I've shipped something that like had a negative impact, then it sends me an email and it says, "There's been a regression." And that was an agent that sort of woke up Sunday, ran the test, and then like sort of shut down. Um, but yeah, basically what I've-- this is more of like an operating system. Like if you're gonna be doing work inside of Claude, I think this operating system will like help you get better growth results. And it's, it's my sort of take on like how to work with Claude as a

9. 32:00 – 36:12

   Are autonomous agents overhyped?

   1. JMJames McAulay32:00

      growth person.

   2. AGAakash Gupta32:00

      So are autonomous agents overhyped, or are they just not there at quality? Or why do we just have kind of like this schedule on skills?

   3. JMJames McAulay32:08

      That's a really good question. Overhyped in that I reckon like a lot of the people telling you on LinkedIn and X that they have, you know, autonomous agents doing stuff for them, they're either like burning tokens on sort of like kind of useless stuff or, you know, these agents are maybe like going out to scan a competitor and sending a briefing back. But in my experience, like people are using these kind of autonomous agents to basically like summarize information once a day or like once a week. You know, I think we've both seen Instinct go kind of viral on X over the last few weeks. I wouldn't call Instinct autonomous in that, you know, I don't think it kind of just decides to go and do a piece of work for you. Like, these agents are like at your beck and call, right? You give them a command, and they can go and spend like two or three hours, twenty-four hours on a problem. I'd say that's, you know, like relatively autonomous. But the idea that we have these agents that we sort of create, and then they just go off and like print money for us without us sort of doing anything, I think is a little bit of a fantasy, um, and maybe some kind of overhyping on Twitter. What about you? Like, do you have any agents that are sort of doing stuff without you really sort of steering them? Or is it, is it like this kind of direct the agent, have them work for a few hours, come back?

   4. AGAakash Gupta33:28

      I am painfully bad at focusing on growth stuff. All I focus on is content, content, content. [chuckles] So the only thing I have is like a couple schedules. Like, I look at something religiously called like my LinkedIn post lab, which basically-

   5. JMJames McAulay33:44

      Ooh

   6. AGAakash Gupta33:44

      ... I have a schedule which pulls from 100 creators who I follow on LinkedIn every morning. So I don't actually browse LinkedIn really. I just browse my LinkedIn post lab, and I actually-

   7. JMJames McAulay33:55

      That's cool

   8. AGAakash Gupta33:56

      ... host it on my website. And so-

   9. JMJames McAulay33:57

      Okay

   10. AGAakash Gupta33:58

       ... that's just a schedule. And then I have a couple loops running. The idea with these loops is that they're gonna go look at, um, the feedback that I give them to become self-improving. So they're not autonomous in the sense that they're not doing anything without my review. But I do have, for example, a loop that's focused on helping me figure out what I'm gonna post on LinkedIn. And so it'll go look at my post performance yesterday, and it'll automatically update itself and say, "Okay, you tried this." Like for instance, realistically, yesterday I tried, "Hey, what if I post two posts instead of one post per day, one link post and one image? What's gonna happen?" And so this morning it came back to me and said, "Hey, your net impressions were actually lower than one post a day, so maybe that experiment didn't work. And so here's what I'm proposing you write about today. You just published this deep dive. You should write about that." So that's the only sort of-

   11. JMJames McAulay34:52

       I mean, you said you're not doing much growth, but that sounds very scientific. It sounds like you're doing-

   12. AGAakash Gupta34:58

       True

   13. JMJames McAulay34:58

       ... you're just, you're doing content-like growth, right? And like you're, you're trying to iterate, you're trying to run experiments. Uh, I would-- I think you're, you're underselling what you've built slightly. And, and, and that sounds cool. I think we could, we could call that sort of proactive agent. You could say it's autonomous, you know, if you were sort of using the term liberally. But as we both know, like it's, it's probably a skill that runs on a schedule, does the same thing sort of every day, and that, that can be powerful. Uh, but yeah, I'm not claiming that, you know, if anyone installs this, they can sort of like retire and just have agents like, you know, run their business for them. I think this is like-- Uh, w- I mean, I, I feel like I had an analogy for this recently, but yeah, my, my day-to-day is like five or six tabs sending off different pieces of work to Claude Code. Sometimes I use like GLM or Kimmy if like Claude is down. Um, but the whole harness around this is basically what is like, you know, sort of nudging me towards the goals and sort of making sure that we like document and learn from what we do. And then these skills, you know, you can spin these up as subagents, you know, but you could set them up as a, a schedule. But I'm, I'm not sort of-- I'm not claiming that this is like an agentic team,

10. 36:12 – 39:14

    SEO audit results on my site

    1. JMJames McAulay36:12

       yeah, that like runs itself. Not yet, anyway.

    2. AGAakash Gupta36:15

       So maybe can we circle back to the SEO audit? I think there we'll find that I don't focus on SEO quite enough. [laughs]

    3. JMJames McAulay36:23

       [laughs] Okay. So yeah, one thing I forgot to do here is I forgot to ask for this as like an HTML report, but we got some issues. Uh, so, well, it thinks that the issue is a lot of your posts are paywalled, uh-

    4. AGAakash Gupta36:35

       Mm-hmm

    5. JMJames McAulay36:35

       ... which is probably, probably intentional. Apparently, a lot of posts don't have SEO descriptions, and there's a bunch of stuff about, like, crawlability. Some of your titles are quite long. There's, like, some duplication going on. You know, like, there's quite a lot of good stuff here, and I remember 10 years ago paying SEO agencies, like, quite a lot of money to come back with a report like this, like two or three weeks, you know, after we'd engaged them. And then I remember giving a report like this to my developers and waiting two or three weeks to have it all sort of like prioritized, backlogged, built. Whereas now, you know, if I sent you this report, you gave it to whichever agent is inside your repo and said like, "/goal test and refine until all of these are solved," it would do it. And so, you know, in about a couple of hours, we've just like tidied up your SEO presence quite a lot. Does this match what you were expecting?

    6. AGAakash Gupta37:28

       Well, I'm actually literally in the process of looking at what happened to my SEO [chuckles] because in May I was getting like 25K visits from SEO, and it's down to like 10K. So these are all great next steps for me, and I'm literally going to go do those after this.

    7. JMJames McAulay37:45

       Yeah, definitely. Um, I don't have your email to hand, but I will like-- I'll send this to you straight afterwards. I mean, a couple of things that I think are interesting. Um, so we could talk about My PRD skill. I mean, effectively everyone in your audience knows what a PRD is. I don't think we need to go into this in too much detail. But again, like I said earlier, mine just really grills you on like why you want to build something. So, you know, there's the famous, um, five whys test. I think it was like Toyota that came up with this, where they were actually using it to like get to the root of a problem. So something happened and they'll ask like, "Well, why did that happen? Okay, well, why did that happen?" And eventually you get a kind of root cause analysis. I've taken that and I've basically like pointed it back at the user and said like, "Why do you want to build this?" So for example, I want to, you know, let's say I want to like build a new page on my marketing site and Claude will say, "Okay, cool. Like this seems like it might be a good idea. Tell me why. Like what metric is it going to move?" And I'll say, "Well, I think, you know, it might rank for this term in SEO." And then it will sort of look at the data and say, "The, the volume isn't actually that high. Are you sure you want to build it?" And so sometimes it's trying to talk you out of building something, but then basically once it's got the PRD, it's set up in a way that it's got all these checkboxes in the PRD, and it just basically sort of ticks off its work as it goes. And so if you hit a context limit or you need to like kill the session and you continue after that, the agent just looks at the PRD and it sees, you know, I'm halfway through and then just, just keeps

11. 39:14 – 45:38

    Multi-agent refine

    1. JMJames McAulay39:14

       going. So I think this is like a good PRD skill, but we don't need to sort of dwell on it too much. This is something that I use all the time. Um, multi-agent refine. I don't know if this visual is like the best visual, but basically when I've written a PRD, when I've created anything, when I've built something that I think is like quite critical, like payments infrastructure, I'll run this skill that basically spawns subagents, could be like three or four, sometimes up to ten, and they all just take like a slightly different lens. And so one of them is maybe like, you know, looking for technical correctness. One of them is like asking if we missed anything. One of them is looking for like simplicity. Um, so if I'm building something, then it's kind of these like subagents in an engineering sense. But if I'm planning a piece of growth work, then this multi-agent refine basically takes, you know, like a few different approaches on like growth strategy and then comes back and says, "Okay, your plan is strong in this area, but like the critics, the kind of like panel of critics found these weaknesses." And as soon as I kind of like created this skill and started using it, I, I don't think I've stopped using it on like a daily basis because even Fable, like even Astra can, can miss like quite, quite big things. I think the subagent that I find particularly helpful, I don't know. Yeah, here, this one missing. Often we will sort of like have an idea. We'll go to an agent and we'll say like, "Let's build this." You know, maybe interview me before we build it, write the PRD. But it's kind of like we don't know what we don't know. And so if you ask an agent like, "What's missing from this plan?" It very often surfaces things that you hadn't thought of and that your primary agent hadn't thought of. Um, so yeah, I, I love this skill. Um, the one thing I would say about this is make sure that you're using Sonnet subagents. I accidentally s-spun up like 20 Fable subagents yesterday and just immediately hit my limits. [laughs] So yeah, make sure you're using Sonnet or Opus subagents for this one.

    2. AGAakash Gupta41:21

       Can you show us that in action?

    3. JMJames McAulay41:23

       Yeah. Let's, uh, see what we got here. Okay. Let's, uh... So we got this-- I don't know if this is gonna open at the moment, but we've got there like Q2 growth OKRs, right? And so I'm gonna say multi-agent refine OKRs. And so I use this quite a lot for like, yeah, strategy planning documents. And so it's going to spin up maybe like four-- Actually, let me do this. Let me just switch to Sonnet so that all our demos happen faster. Okay. Um, okay, so it's going to spin up, I think, five subagents to kind of like critique our OKRs. Um, now, as I've mentioned in, in like the GitHub repo, you know, there's some stock subagents that I use all the time, like what's missing? Is this feasible? You can also, if you want, I've seen people set up their kind of like panel of advisors, you know, like Paul Graham, uh, Hormozi, like whoever you look up to, you can basically train an agent on their writing. And then when you run-

    4. AGAakash Gupta42:32

       Yeah.

    5. JMJames McAulay42:32

       -multi-agent refine, it will sort of like approach it as they would. So what it's gonna say is, it's gonna say, "Do you want autonomous improvements?" So like it's gonna run a first pass, the master agent is gonna like review the results and then just go and fix it for you. Or do you want to be involved? Like, do you want to kind of like see the critics' results and then, like decide what you want to change? I, I use both, but for this one, let's say like we'll sort of check in. It's, you know, notice that we are critiquing Q2 goals in Q3, but this is, this is just a demo. So yeah, it's looking for like, are these OKRs measurable and rigorous? Are they feasible and do we have capacity to do them? Are they aligned with like the mission? Have we missed anything from these? And then, yeah, those are the four. So they're running here. You can see these are subagents. Uh, this one, you're a skeptical critic reviewing this company's growth OKRs. You are only-- Oh, okay, it's back. [chuckles] Sonnet is fast. Uh, so it's found a lot of problems here, right? Um, this document is stale. I mean, that's a, that's an obvious problem. Uh, there's a note in the retro that undermines these numbers. Bunch of like technical things about like qualitative judgment instead of quantitative judgment. Like I've seen all of this happen before, right? When teams write OKRs, it's not set up, you know, to be measured correctly, or they've kind of like hedged some of the numbers because the wording of the OKR was wrong. Um, so yeah, these agents have just finished actually. I forgot how fast Sonnet 5 was. And so now our main agent is like synthesizing what they all found. And it's gonna come back to me and say, okay, like P1s or even P0s, like you need to fix these things in this document. P1, P2, and then like P3, you can sort and leave that out. And again, just like adversarial review, this is called adversarial review, always, in my experience, surfaces like far more than just getting one agent to kind of like review its own work. So it's come back here and said, you know, numbers are stale again. Uh, it's not actually linked up to company OKRs, so we don't actually know if this person's OKRs are, are helping the company OKRs, and et cetera. So you can run this, and then it will say, "Which of these do you wanna change?" And you can sort of go through. So I, I think you have to try this skill out to kind of like experience the magic, but trust me, this is a very good skill.

    6. AGAakash Gupta44:52

       So should you be running like multi-agent review on the SEO audit, or when should you be running that?

    7. JMJames McAulay44:58

       Let's go through this example. Yeah. So it's basically, it's found these issues. If this was my repo, I would then turn this into a PRD, and, you know, the agent would go and research the code base, it would research the audit, and it would put together like a phased PRD. And then I would multi-agent refine the PRD. So-

    8. AGAakash Gupta45:19

       Mm.

    9. JMJames McAulay45:20

       Audit PRD, five subagents like attack the PRD. And usually, yeah, they'll kind of find improvements and maybe like things that the first agent missed. So I'm often using multi-agent refine in this planning phase just before I then hand the plan over to, you know, the build agent or,

12. 45:38 – 47:27

    Experiment discipline

    1. JMJames McAulay45:38

       or agents.

    2. AGAakash Gupta45:38

       And then once you have multi-agent review, how do you get it to do the kind of the disciplined experiment process?

    3. JMJames McAulay45:45

       Yeah. So this is all sort of built in. If I just come in here. So we've got this like template, uh, here, you know, our hypothesis, uh, if it's tied to a key result, like the baseline, the primaries, you know, any confounders, what might make us stop this experiment. Like this alone, I wish I'd had this, you know, like many years ago when I was sort of running growth in a small startup. Just this alone will force you to be rigorous in your experiments, right? And then we've also got like a playbook on A/B testing, uh, maybe with like specific stuff on, on your site. So if I was in this demo, let's just like close this off. Um, help me set up an experiment for top of funnel landing page conversion. Want to test, uh, exit popup. So with all these rules, with all of these kind of templates, it's just gonna make sure that your experiment is like documented correctly. And then all the skills I showed you, like weekly retro, even daily standup, they look at the list of running experiments and kind of ask you to, you know, they'll say, "This one's finished. Like, shall we do a review of it?" Or, "This one's gone stale. It was supposed to finish two weeks ago, and you haven't documented it." So the whole system is kind of like looking to make sure that these experiments get wrapped up. So in this example, you know, it's asking me for baseline conversion rate, traffic, et cetera. Nobody should be answering these questions manually, right? This is all data that should come from PostHog, Mixpanel, Amplitude. So in this demo folder, I haven't got PostHog set up, but I've referred in this like set of tools. PostHog is like the first tool

13. 47:27 – 52:44

    PostHog + Claude Code

    1. JMJames McAulay47:27

       that I recommend everyone set up. Have you used PostHog yourself? Have you used it in product work?

    2. AGAakash Gupta47:31

       I use it, yeah. It's so good. And in particular-

    3. JMJames McAulay47:34

       Yes

    4. AGAakash Gupta47:34

       ... I feel like there's so much on the free plan, which is so nice.

    5. JMJames McAulay47:37

       You're literally the second person to say this to me in 24 hours. Like, they're very generous with their free plan. And in fact, I think I only upgraded last week because I wanted, you know, retention past 12 months. But they have built just like a stunning MCP, API, CLI. Um, they've got like a sort of PostHog wizard now that you just like one line into Claude Code, and PostHog will sort of, you know, run all your experiments for you. I mean, just to, just to show you mine, um, I created this PostHog account at the same time as I was like discovering Claude Code. And so I haven't created like any of these dashboards. Claude Code has done all of this through the API. Similarly, you know, with my experiments, I haven't created these in the UI. Basically, Claude's API key has permission to do this. It's set up the measurement. We're waiting for our test. So far, our test isn't quite working as we wanted it to, but it's still got loads of time left. So yeah, like PostHog plus Claude Code is, to me, just like a growth person's dream. I also think, you know, if you're a growth person, try to get-- try, try, try to get at least read-only access to your company, like Git repo.

    6. AGAakash Gupta48:47

       Mm-hmm.

    7. JMJames McAulay48:47

       Like, your team might not let you ship stuff, but if your agent can see all the event definitions in PostHog, then, you know, it can basically sort of help you measure absolutely anything. But if your kind of PostHog system is a bit of a black box, you know, if, if you're a growth person that only has the UI and cannot see the code, you're often gonna be like pinging engineers on Slack saying like, "When do we fire this event?" And like, "When was this event last updated?" And, "Why have we got two of this event?" Whereas if your team just give you a read-only key for Git, you can just answer all those questions yourself.

    8. AGAakash Gupta49:23

       What are some tips and tricks to get the most out of PostHog as a growth person?

    9. JMJames McAulay49:28

       Okay. So definitely what I just mentioned about having access to your code base if you can. Even if you're not gonna be writing code or shipping PRs, just get access to Git so that you can understand like how events are firing. If you are shipping PRs, whether you're a sort of product person, growth engineer, I have set up a bunch of tests in my CI, so like continuous integration. When I push a PR and all the tests run, one of the tests is that everything has to have a PostHog event. So I cannot merge code into production unless it has all of the PostHog tracking set up. Because I know that I'm gonna forget this. Like, I'm gonna be excited about a feature I'm gonna create a new landing page, and I'm just gonna forget about PostHog. Whereas like right in, you know, day one of setting up this repo, I thought, "Okay, what is the system that means I never forget PostHog events ever again?" And the system was, you know, a test when we submit a PR. So I'd highly recommend doing that so that you don't need to think about, you know, adding new events. And then something else I did, I'm trying to remember where it is in the UI, and again, I don't use the UI [chuckles] that often, so I'm-

    10. AGAakash Gupta50:35

        It's mainly like through Claude Code, right? That's how I use it.

    11. JMJames McAulay50:40

        Oh, yeah. I'm-- Yeah, yeah, yeah. This is me showing the UI because we're on screen, and it's slightly more interesting. One thing I did last week that was really helpful is I basically wrote very good definitions for every event. So I was good at creating the events and, you know, every event existed. I hired an agency who are doing some like A/B tests across my funnel, and they were in Cl- they were in PostHog chat, and I noticed that one or two things were getting misinterpreted, and it was because we had event names, but we didn't have event definitions. And so I basically wrote a PRD where I said, "Okay, like we're going to define every single event. I want you to like scan all of them, make the plan." Then I multi-agent refined the PRD just to check that we were like taking the right approach. And now I can see every event. It says, you know, when this is fired, where it's fired, use this for, you know, measuring these types of things. And every single event in my entire sort of list of 350 is documented. Now, pre-AI, [chuckles] you'd only have this if you were working with like a god-tier product manager, right? Like I just-- I don't think anyone's data would be sort of this well-annotated. But this was a task I did in about two hours last week across like five different things I was doing. So yeah, highly recommend just making sure your data-- It's a bit boring, but make sure your data is like well labeled, well annotated.

    12. AGAakash Gupta52:05

        What's like the right prompt here to Claude Code? Like audit all of my telemetry and event definitions, figure out gaps where we should create new events, figure out places where our definitions are wrong and help update. Something like that?

    13. JMJames McAulay52:19

        That was perfect. Yeah. [laughs] Basically. I mean, if, if you like, I can, can go back into Claude Code and find the exact prompt, and we could show it on screen. But what you've just described would absolutely work. And just when you do that, just make sure, you know, I'm not gonna go into my API key settings right now, but just make sure that that key has the ability to like read and write to-

    14. AGAakash Gupta52:41

        Mm.

    15. JMJames McAulay52:41

        ... your like event definitions.

    16. AGAakash Gupta52:43

        So

14. 52:44 – 1:00:18

    Mistakes and connecting every tool

    1. AGAakash Gupta52:44

       you've set up this agentic growth OS. A lot of people, they might just be getting access to the tools to set it up. You've made the mistakes along the way. What are the big mistakes and learnings that people can just speed up their process by learning from your mistakes?

    2. JMJames McAulay52:59

       So when I started working with Claude Code, I think like everyone, I felt like a caveman that had discovered fire. And I kind of like fell under the spell of Claude Code and thought that it could do anything and I could solve any problem. And I realized that, you know, we've all experienced this, but like Claude will confidently give you information that's just wrong, and it will introduce bugs even when you're supposed to have tests. So I guess one thing on, you know, if, if you're doing this in any sort of technical capacity, I now use Cubic as like AI code reviewer, uh, which just like makes sure I'm not shipping bugs. 'Cause I had one or two bugs that were like, they didn't cost me money, but they were like quite annoying. Uh, so definitely, yeah, Cubic. Um, [lip smacks] and I also use Sentry for error reporting. So like highly recommend if you're gonna be writing code, you know, don't trust Claude on its own. I've said here, you know, you can use any agent here. I'm primarily in Claude Code, but like this is all set up in a kind of agent-agnostic way. So like I do also open Codex sometimes and just kinda like keep working in this folder. I think my biggest piece of advice is like get everything connected as fast as possible. So like this is another diagram, right, of like my setup, and I've got, you know, Sentry, Cubic that I just mentioned here. Playwright is great for kind of like testing stuff, but I think a lot of your audience aren't even writing code. So, you know, I've also got all my like content tools, right? Like ElevenLabs, Epide-Epidemic, Mistral. Um, on analytics, I've got like VidIQ, Data for SEO. Like I would recommend just trying to plug everything into Claude Code. Like giving it full context across every ecosystem you work in is, is what kind of like helps it move at an incredible speed. You know, yesterday, I don't know if I've got it here. Yeah, I use Kit for my mailing list, and I was thinking, "Oh, wouldn't it be cool to get like an alert in Slack every day that shows me people who engaged with an email?" Right? So I've like sent an email to like 1,000 people. Show me the people who've engaged. And then I thought, "Well, actually, I don't want just everyone. I want people who haven't bought from me before, and also I want them to be enriched." So I want to use something like Apollo to see their LinkedIn, to see their profile. Uh, and I want to use Attio, my CRM, to kind of check who's paid and who hasn't. Now, if I hadn't connected all those tools, that would require an engineer. Like that would just be a painful exercise. But that was a prompt. Like yesterday, I sent that prompt saying like, "I want a daily Slack alert that sends me non-booked leads enriched with Apollo using, you know, with a link to Attio." Uh, I'm not gonna show you. I could show you on Slack, but it's got, it's got emails, so I'm not gonna do that. And that flow is like so helpful for me and for someone who's helping me with sales at the moment. Um, and it's only possible because everything is plugged in to Claude Code, whether it's, you know, an official MCP, an official CLI, or just an API key. You know, if you give Claude Code an API key, it will usually work wonders for you.

    3. AGAakash Gupta56:05

       Wow. So we talked a little bit about this at the beginning, but I want to circle back this. How has this stuff evolved since you were at ElevenLabs?

    4. JMJames McAulay56:16

       Yeah. I was at ElevenLabs in 2025, and we were definitely ahead of the curve on AI. And In 2025, being ahead of the curve meant using custom GPTs. You know, skills, if I'm correct, you can fact-check me on this, but I'm pretty sure that like skills as a concept were really formalized in like autumn last year. So for the majority of last year, you know, people weren't using agent skills. We had a few different custom GPTs. We had one for just making sure that like copy was on brand, for example. Like ElevenLabs has a sort of like fairly well-defined like tone of voice. And so it was a custom GPT, you could run anything through it, and it would just sound on brand. And so then we started to extend that. You know, we had a, a big spreadsheet, like a launch checklist, and every launch you would like duplicate this sheet, you would like adjust it, whether it was like a tier one product launch, tier two product launch, and then you'd just work through the sheet. And we started to build out like custom GPTs that would like take on parts of that process. I am 100% sure now, you know, I left ElevenLabs last year, but I'm 100% sure there is like a launch skill now. Um, so I think it's, it's come on a long way. I would also bet that the majority of people in the growth team at ElevenLabs will be like working in the way that I just showed, like inside Claude Code plugged into all the tools. Twelve months ago, you know, even the developers were just discovering Claude Code. Uh, I don't think growth people and product people were inside Claude Code, you know, this time last year. So yeah, as a, as a growth industry, as like a product industry, I think we've, we've been given these superpowers in the last 12 months that, you know, really do unlock so many abilities for like non-technical people.

    5. AGAakash Gupta58:05

       It's actually wild. What do you see as like the next frontier? You mentioned Instinct. Is that part of it? What's like, what's coming up in the next couple months that everybody in growth is going to be adopting?

    6. JMJames McAulay58:18

       Well, I think I've heard people talk about this idea of the autonomous company, like the company of agents that kind of grows itself. And I've seen like a few people on X kind of claim to have built something like this, and I have had attempts at building something like this, you know, where I want to kind of like close my laptop at 6:00 p.m. and just have agents, as we talked about earlier, like autonomously pick the right task, work on it, check their own work, and then in the morning, I sort of wake up, review it. Maybe, maybe I don't even review it. So yeah, to me, I think it's like this notion of like a self-improving 24/7 company of agents. And right now, you know, my system works really well, but it still kind of waits for my prompt. I've got some scheduled tasks like you have. But I would love to reach a point where an agent came back to me and said, like, "I've wrapped up this experiment. It was a success. I've kicked off the next one. You didn't do anything." Um, and I think just this, this idea of like an agent running 24/7 is just not quite ironed out yet. Like Vercel have built a great framework called Eve. Highly recommend trying that out. Um, people bought Mac Minis and just sort of like set up OpenClaw on them. Instinct is kind of the latest personal agent and like Meta's Muse assistant this week as well. Just this idea that like there's an agent not bound to your computer, like running in the cloud somewhere, kind of getting stuff done for you, maybe even preempting what you want or, or need. Whoever figures that out for companies, I think, is, is gonna make a lot of money. And then I think, you know, growth people will be tasked with, you know, pointing these agents at the right experiments. Like you've got to understand your customers to, to run good experiments. There's no point in running, you know, A/B tests

15. 1:00:18 – 1:07:19

    From one workshop to a cohort business

    1. JMJames McAulay1:00:18

       on a page that no one cares about. So yeah, that would be my sort of prediction.

    2. AGAakash Gupta1:00:21

       I want to talk about you for a second. How the hell did you scale up this business to $100K revenue in just four months?

    3. JMJames McAulay1:00:27

       Okay. So I was doing consulting, I was doing growth consulting, and I got asked to do like a workshop teaching execs on AI. And actually, like 12 years ago, I was teaching execs coding through coding bootcamps. Uh, so I, I did this one-day workshop, and I got really good feedback, and people were like, "Oh, you're, you know, a good presenter. Um, feel like I transformed how I use Claude in two days." And so I thought, okay, like maybe, maybe I'm onto something. And I'd been like plugging all my MCPs into Claude. I had this pretty good system that was using like Apollo, Granola, Todoist. And I ran a workshop online, and I said, "Hey, I'm gonna show you like how I'm running this business." It's, it's still consulting. Like at this point, I'm still just doing growth consulting. And I got something like 150 people attend that, just mostly from my own network. And so again, I thought, okay, there's appetite. Like I've got product market fit here. People want to know how to actually use AI to grow faster. And so I was doing consulting like four or five days a week, and I thought, okay, I'm gonna pull that back to three days a week, and I'm gonna give myself two days a week to run some sort of cohort. Like can I teach, you know, like a bunch of companies, maybe like 20 people, can I teach them over four weeks how to like get all this stuff set up? And I got 60 people signed up. And so that was like the busiest month of my life, was like three days a week of consulting and then sort of two to four days a week of like running this course. Um, and yeah, I think like, you know, I put my heart and soul into it. I really made sure that everyone got good results. Uh, and the word of mouth has just been kind of crazy. So I incentivized them with a referral program. I said, you know, "If you've enjoyed this and you know someone that you want to, to kind of like get trained, here's your link, and, you know, it gets them a discount, it gets you some money." Again, Claude Code set it up for me. I used something called Rewardful with an API that was like very easy to use. And then I got people referring, you know, like companies that wanted to train 20 people at once. Uh-

    4. AGAakash Gupta1:02:24

       Wow.

    5. JMJames McAulay1:02:25

       And with my pricing model, like, you know, 20 people at once is pretty good. So I've basically been sort of like building my own learning platform, running cohorts like this every couple months, uh, and doing like private workshops as well. So I've like just got back from Vegas, uh, went to Greece in the summer, uh, to do these like intensive, like either hackathons or sort of like exec training sessions. And yeah, I think I've just-- I've found a format that, you know, people, people want, uh, and I've used Claude, you know, fully as, as much as I can. I've like maxed out that Claude two hundred dollar a month subscription. And I just feel like I've applied everything that I learned, like running a company, being at ElevenLabs, doing growth consulting. So yeah, I think just good timing and sort of like the right stage of my life to apply all the lessons I've learned.

    6. AGAakash Gupta1:03:14

       Hmm. You've basically like taken everything you've done in the past and you've put it into like a business built around that skill set, which is genius. So you've taught over four hundred people now. What do most people get wrong as they're learning this stuff?

    7. JMJames McAulay1:03:27

       I think people don't realize that you can ask an agent to improve itself. Like so much of what I do, and actually I've seen maybe, yeah, one of the things I learned was like, I was at this company, Jack and Jill, and they're an agent company. I don't know if you've heard of them. Jack kind of t- speaks to candidates and Jill speaks to recruiters, and together they try to like match people. And this is like a AI native, you know, like agentic first company. And I would hear these like-- I'd hear conversations in the office where someone would say like, "I asked Jack how he would improve his own system prompt, and he came back with like, you know, this ten-step list, and I did it, and now, now he's better." I was like, "What?" Um, you know, we're-- I'm not talking about, you know, recursive self-improvement in the sort of AGI sense, more just that like you can point an LLM at its own system prompt or at its own skill and run evals or just say like, "How can we improve this?" And I think that's a skill that like a lot of people haven't even realized is possible. You know, quite often someone comes to me with a question. They're like, "How can I improve my CLAUDE.md?" Now, I can give them my take on it. I can say, "Okay, here's some best practices." But what I usually say to them is like, "Well, have you asked Claude? Like, have you said to Claude, 'Based on all of our recent interactions, everything you know about me, and by the way, go and read the Anthropic docs, how could we improve my CLAUDE.md?'" So yeah, I think, I think people just haven't yet sort of figured out that they can-- these agents can improve themselves. I think as well that the reason for that is Claude Chat isn't very good at like editing itself, right? Like it's in the cloud. It can't really like set up second brain. It can't edit its own system instructions. But once you get into Claude Code and Cowork, you know, they are able to like reflect on their own work and basically edit any file on your computer. And I think when people make that jump from like cloud-based chat to kind of like file-based agent, they start to realize, oh, like Claude can fix anything about itself. So yeah, maybe that's like the biggest, biggest breakthrough for a lot of people.

    8. AGAakash Gupta1:05:40

       So I've got your LinkedIn here. People got the GitHub link below. You're on YouTube. You have a website. You obviously have the Agent Accelerator. Is there anything else people should know if they want to work with James or learn more from you?

    9. JMJames McAulay1:05:53

       I think you've covered it. I mean, I post like daily tips on LinkedIn. I am posting like a good video every week on YouTube. Uh, if you want to book a call with me, there's like a link on the Agent Accelerator website to kind of like find out if we would be a good fit. But yeah, I think you are more prepared than I was there with all the links, so I think that covers it.

    10. AGAakash Gupta1:06:15

        All right, guys. We'll put those in the description or the email where you found this, and see you in the next episode. I hope you learned as much from today's episode as I did. If you can do one thing that's totally free that would help the show, it would be to check that you're following on Apple and Spotify podcasts. Check that you've left ratings and reviews on those platforms. Check that you're subscribed on YouTube. Leave a like and a comment on this video, and then share it with your friends. We're trying to make better and better podcasts. After two years, we think we've gotten something pretty good going. So let us know what we can do to make it even better, who else we should interview, and we will put on the best shows we possibly can. Finally, don't forget my offer for the bundle. You get an entire year of my paid newsletter, plus my favorite AI tools, bolt.new, Airtable, Speechify, Descript, Magic Patterns, Linear, Dovetail, Arise, and Mobbin. That's twenty-seven thousand dollars worth of value for just a hundred and fifty dollars. So check that out at bundle.aakashg.com if it interests you, and I can't wait to share our next episode soon.

Episode duration: 1:07:29