Advertise on podcast: Build. Grow. Repeat.

In many companies, nobody can name the person responsible for their AI. So the answer defaults to "no". Nur Aslan founded IADM, the International Academy for Digitalization and Management – a CPD-accredited academy that trains Certified AI Officers. She came to AI governance from HR: for her doctorate she surveyed 397 HR professionals across Europe and kept finding the same gap between AI adoption and AI readiness. In this episode we pick up where the conversation with Dr. Alexa Scheffler ended. Alexa answered "can we use AI for this?". Nur answers the people side: who is responsible, and how do they need to be trained? We talk about why big companies say no to AI for months – in one case, a year and a half – why governance should be an enabler and not a brake, and why a small team doesn't need to hire an AI officer. By the end of the episode, I'm the AI officer at Pirate Skills. Nur's Monday list for a new AI officer: list your AI tools, classify them by risk, name who is responsible for anything high risk, and get technical help where it gets technical. IN THIS EPISODE 00:00 Welcome – a compliance academy incorporated in England 02:50 The gap between AI adoption and AI readiness 06:31 Governance vs compliance – what's the difference? 08:18 Why big companies say no to AI 11:03 Does a small company need an AI officer? 17:02 Training your team: risk classes and the AI inventory 18:29 Building the academy: from live Zoom to AI avatars 24:28 Is there a Claude skill for the EU AI Act? 27:02 The Monday list for a new AI officer 32:02 Why Europe needs to move faster 34:50 See you at Vibe Coding Cologne DEEP DIVE LINKS – The companion article: https://pirateskills.com/insights/navigating-ai-governance – Part one, with Dr. Alexa Scheffler: https://pirateskills.com/insights/navigating-ai-compliance – Pirate Forge – Build phase starts this Wednesday, 30 Sep: https://pirateskills.com/pirate/forge/q4-26 – Meet Nur live at Vibe Coding Cologne, 7 Oct: https://pirateskills.com/events/vibe-coding-cologne/2026-10-07 – Connect with Nur: https://www.linkedin.com/in/nurgul-aslan-70600833b/ – Connect with Ben: https://www.linkedin.com/in/bensufiani/

A native mobile app, built in one weekend and running on my own iPhone – with my kids over for most of it. Two years before Pirate Skills, at Maple Apps, a designer and three full-time developers still needed two weeks to get from finished screen designs to the App Store – and that was a high-performing team. The industry standard is months of work and tens of thousands of euros you'll probably never make back. This time it was one founder and Claude Code, and that was fascinating to me: it's truly possible now. You'll see the Maple Apps rule – no line of code before 1,000 people are on the waitlist – and why an app is not your business but one more door into it. There's the choice between going fully native (Swift and Kotlin) and React Native with Expo, and why that took a weekend instead of a debate. You'll see screen designs made with Claude Design, the first screens on the phone through Expo Go, and the moment the app was really installed, on an hour-long ride to a kids' birthday party. To be honest about where we are: it's not in the App Store yet. The Apple and Google developer accounts are waiting for approval, then come the store rules and the review. Test users are planned for Vibe Coding Cologne on 7 October, with the app live in the stores by November at the latest. The push to finally try came from a first-time builder at our hackathon who shipped her first app without any technical background. If she can do it, and I can do it, you can too. IN THIS EPISODE 00:00:00 – A native mobile app in a weekend 00:01:05 – Maple Apps: 1,000 sign-ups before a line of code 00:03:30 – Two weeks vs. months and tens of thousands of euros 00:04:15 – The app is one door, not the business 00:06:19 – Native or React Native + Expo 00:08:36 – Friday: a protected API and new screen designs 00:11:21 – Saturday: the first screens on my phone 00:12:52 – Installed for real, on the way to a birthday party 00:14:39 – Not in the App Store yet 00:15:46 – Your weekend: the step-by-step guide 00:19:41 – Upcoming events and the Pirate Forge DEEP DIVE LINKS 📖 Full Article + the starter prompt: https://pirateskills.com/insights/mobile-app-in-a-weekend ⚓ Pirate Forge – 30 September : learn vibe coding and vibe marketing by building a real app alongside me: https://pirateskills.com/pirate/forge 🎧 Spotify: https://open.spotify.com/show/7kFl0BJiOYih8yqQTRR9SN 🍎 Apple Podcasts: https://podcasts.apple.com/us/podcast/build-grow-repeat/id1870561487 🗓️ All upcoming events: https://pirateskills.com/events 💬 Connect with me on LinkedIn: https://www.linkedin.com/in/bensufiani/

Your users can now send their own AI agent to get what you offer: find an event, sign up, search your knowledge, even prepare a checkout. They never have to open your website. Most vibe-coded Next.js apps can't do that yet. Your website is a teller who needs to see a face: the code behind every button checks the browser's login cookie first, and an agent calling from ChatGPT, Claude or a server has no cookie. So it gets turned away. Not because it isn't allowed, but because the teller can't see a face. On 13 September I connected pirateskills.com to my own Claude.ai, sent one chat message, and it registered me for an event. Here is how we got there, explained for people who vibe-code, not for senior engineers. First, build the vault once. Each valuable action (discover events, register, search the knowledge, start a checkout) becomes one server function that is told who is asking, with the same AI SDK tool card for every door. Then open thin doors onto it. The Captain, our on-page chat built with the Vercel AI SDK and the AI Gateway. An MCP server with OAuth sign-in for Claude, ChatGPT and Codex. API keys you create and revoke yourself. And a CLI for anyone in a terminal, you or your coding agent. Why not stop at MCP? I personally am annoyed that these types of connections rarely hold. An agent running in the cloud, on OpenClaw, Hermes or a scheduled job, has nobody there to click "allow". OAuth is the convenience. The key is the guarantee. A public SKILL.md is the treasure map: which door, when to use each tool, and an explicit yes before anything that changes something. And for money, the crew prepares the Stripe checkout and you sign. No tool can charge a card. You don't need five doors. One vault and one door is the whole first version, and the article has the six steps plus a prompt to hand your coding agent. IN THIS EPISODE 00:00:00 – Open your vault to agents 00:01:52 – The teller problem: six clicks to join an event 00:02:49 – Demo: signing up for the AI Compass from Claude 00:04:26 – What's in your vault 00:06:02 – Five doors into one vault 00:08:58 – How to build it with Next.js and the Vercel AI SDK 00:11:59 – Your own chat, MCP, and why an API key holds 00:15:45 – The skill: a map for your users' agents 00:17:52 – Agentic commerce: the agent prepares, you pay 00:20:26 – Open your first door, with a live build 00:26:26 – Where to go from here DEEP DIVE LINKS 📖 Full Article (with the four doors, the handshake-vs-key table and the prompt for your coding agent): https://pirateskills.com/insights/open-your-vault 🎧 Spotify: https://open.spotify.com/show/7kFl0BJiOYih8yqQTRR9SN 🍎 Apple Podcasts: https://podcasts.apple.com/us/podcast/build-grow-repeat/id1870561487 🗓️ All upcoming events: https://pirateskills.com/events 💬 Connect with me on LinkedIn: https://www.linkedin.com/in/bensufiani/

Your product can keep fixing itself while you sleep — and so can your marketing. For two weeks pirateskills.com has run on a loop I set up inside Claude Code: a scout that reads every data source once a day, a fixer that works the queue every four hours, and me answering a question only when the system is unsure, changing product direction, or about to do something risky. 142 passes, 38 product fixes, 39 signals watched, a test suite that more than doubled. PostHog proposed the self-driving product. This goes one step further. The same shape watches your marketing, your funnel and your money path, not just your code — which is why I've started calling it the self-driving startup. You'll see my actual screen: the loop's setup and the cloud routines it runs on, the Linear board the agent manages on its own, the Vercel runtime logs and PostHog events it reads, the database advisors and Stripe checkouts it checks, and the skill file that decides how far it may go without me. The shape matters more than the tools. Many sources feed one queue. Two roles, finders and finishers. A verification bar instead of a reviewer. A human who answers questions instead of doing tasks. And re-measurement, so what ships changes what the loop reads next. You don't need my setup to start. One source, one queue, one agent working it is the whole first version — and the last stretch of the video is the seven steps to build it this week. IN THIS EPISODE 00:00:00 – Beyond the self-driving product 00:01:25 – What you don't see happening in your own app 00:04:02 – The shape: one scout, one fixer 00:06:41 – The numbers after two weeks 00:08:56 – Source one: your runtime logs 00:13:18 – Source two: analytics and the money path 00:17:04 – Database, payments and dependency updates 00:19:22 – The same loop for marketing, and moving it to the cloud 00:23:24 – Linear is the inbox, and the rules the agent lives by 00:33:51 – Build your own: the seven steps 00:39:01 – Where to go from here DEEP DIVE LINKS 📖 Full Article (with the live coverage map, the one-day timeline and the free 7-day checklist): https://pirateskills.com/insights/self-driving-startup 📈 The Meta Ads Loop — the marketing half of this shape, in detail: https://pirateskills.com/insights/meta-ads-loop ⚓ Pirate Forge — Q4 cohort starts Wednesday, 30 September: six weeks to vibe code your first revenue-ready SaaS, then six weeks to vibe market it: https://pirateskills.com/pirate/forge 🎧 Spotify: https://open.spotify.com/show/7kFl0BJiOYih8yqQTRR9SN 🍎 Apple Podcasts: https://podcasts.apple.com/us/podcast/build-grow-repeat/id1870561487 🗓️ All upcoming events: https://pirateskills.com/events 💬 Connect with me on LinkedIn: https://www.linkedin.com/in/bensufiani/

Marcus Haas is a mechanical engineer. Paul Anduschus is a former event manager who studied business psychology. Together they run DreamyTales, selling personalised AI-generated books, songs and videos in more than ten countries — and neither writes code. This is the conversation I've wanted to record for a year: how do you actually find a positive-ROI campaign instead of guessing, and how much of that can you hand to an agent? They cancelled Triple Whale and rebuilt blended ROAS themselves. Meta comes in through the Meta Ads MCP, Shopify through an app they built, and three skills merge the data into a dashboard that runs every morning inside Codex. No database. Why two sources? Because Meta only shows you 60 to 70 percent of the truth. The most interesting part is how the automation was built. Paul already had rules for scaling, killing and creating campaigns. He talked them out loud, had AI turn them into a document, and that document became skills. That's the real lesson: the ceiling on your automation isn't your technical skill. It's how much of your own judgment you've made explicit. They haven't gone fully automatic. The agent creates campaigns, but Paul still approves daily budget moves because his rules still have gaps. Marcus supervises the big steps while openly admitting the model makes fewer mistakes than he does. And on creative, AI is good at describing what's in a competitor's winning ad but, in Paul's words, “really bad at describing the aesthetics of it.” So the humans work out why it performs, and the machine generates ten videos per prompt. IN THIS EPISODE 00:00 – What a €90 subscription actually costs the people selling it 01:22 – Meet Marcus and Paul: a mechanical engineer and a jack of all trades 08:08 – Running a whole backend without being a developer 15:17 – Three years of organic didn't work. One euro of paid did. 23:00 – The Triple Whale problem every D2C brand has 28:23 – Meta only tells you 70% of the truth 30:15 – The three skills that replaced the dashboard 32:25 – "We are not operators anymore. We are supervisors." 35:51 – Where they give the agent the leash, and where they don't 49:22 – The creative pipeline: from the Ad Library to ten videos 1:01:08 – The real superpower isn't finding the secret ingredient DEEP DIVE LINKS 📖 Full article — the argument, not a recap, with every quote timestamped to this episode: https://pirateskills.com/insights/vibe-code-ad-reportin 🏴‍☠️ Vibe Coding Cologne — Wednesday 2 September, 18:30–21:00, STARTPLATZ Cologne: Paul and Marcus are on stage on agentic performance marketing. Free: https://pirateskills.com/events/vibe-coding-cologne/2026-09-02 🎁 DreamyTales Marcus and Paul's company: https://www.dreamytales.de 🔗 Marcus on LinkedIn: https://www.linkedin.com/in/marcus-haas-212a77215/ 🔗 Paul on LinkedIn: https://www.linkedin.com/in/paul-anduschus-912060190/ 🗓️ All upcoming events: https://pirateskills.com/events 💬 Connect with me on LinkedIn: https://www.linkedin.com/in/bensufiani/

Dr. Alexa Scheffler builds the software that answers one question for large companies: are we actually allowed to use AI for this? Her reading of the EU AI Act is close to the opposite of the one most German companies are operating on. She spent years inside corporates watching good ideas die at the compliance step — not because anyone could produce the rule that forbade them, but because nobody in the room had read far enough to say otherwise. So she built REGULIGHT: a chat agent that asks only the questions it actually needs, and a workflow that keeps the evidence tidy for whenever an audit shows up. The part of this conversation I keep coming back to is what the Act asks for first. Not a ban, not a registry, not a consultant. It asks that the people in your company are educated with AI — and that you write down who was trained, when, and on what. Two names, a date, a topic. Her words: "Write it down and then you're safe. It's not that difficult." We also get into where the risk actually lives. Fully automated credit decisions and hiring with no human in the loop are high risk, deliberately and for good reason. But move the human back into the decision and the same tool changes category — which means most of the time you aren't asking permission, you're choosing a design. And the question that unlocks the meeting isn't "are we allowed?" but "is this a manageable risk?", because that one hands the call back to whoever owns the outcome. The second half is her own build story, and it rhymes with the first. Computer science degree, then almost a decade without hands-on code, waiting — in her phrase — for "this magical technical co-founder" to make her ideas happen. Vibe coding ended that. We compare setups, argue about terminal versus desktop, and she explains why she sends coding work to Claude and strategy work to Codex using a classroom metaphor I've stolen since. She closes on the cost of waiting, and she has a specific word for it: falling behind. IN THIS EPISODE 00:00 – Why German companies move with the brakes on 02:30 – How the EU AI Act sorts risk, and what's actually high risk 05:13 – The better question: is this a manageable risk? 06:45 – The Act's first requirement is AI literacy 09:09 – The questionnaire that became REGULIGHT 11:24 – A computer science degree and a decade without code 12:32 – No more waiting for the magical technical co-founder 14:55 – From Cursor to the terminal: Alexa's setup now 23:11 – Claude, OpenAI, Gemini: the classroom metaphor 25:36 – Obsidian: the markdown vault your agent reads 38:09 – The only woman in the room DEEP DIVE LINKS 📖 Full Article: https://pirateskills.com/insights/navigating-ai-compliance 🏴‍☠️ Vibe Coding Cologne — Wednesday, 2 Sep, STARTPLATZ Cologne: Alexa is on stage, showing REGULIGHT. Free: https://pirateskills.com/events/vibe-coding-cologne/2026-09-02 ⚖️ REGULIGHT — Alexa's product: https://regulight.io 🔗 Alexa on LinkedIn: https://www.linkedin.com/in/dr-alexa-scheffler/ 🗓️ All upcoming events: https://pirateskills.com/events 💬 Connect with me on LinkedIn: https://www.linkedin.com/in/bensufiani/

Gianna Hodel sells software she built by talking to Claude Code. No developer background, no course, no permission slip — and the first version was written from a converted school bus in Mexico with Starlink on the roof. She's a freelance marketer. She got sick of copying numbers out of one marketing tool and into a spreadsheet, tried for hours to make a chatbot do it for her, failed, and decided to build the thing herself instead. That dashboard became Hebelytics, a product other people now pay a monthly subscription for. What makes this conversation useful isn't that it worked. It's how ordinary each step was. Prompting, in her words, is just briefing — say what the problem is, what you want to see, what the outcome should look like, then give feedback. When Claude told her to connect GitHub, she asked what GitHub was, looked it up, and kept going. Hosting, backend, deployment: all learned one question at a time, in the middle of doing the work. We also get into the part nobody puts in a founder story. Getting Google's read-only API approval, so that any customer could connect their own data, took her roughly six weeks of paperwork — including recording a video to walk the reviewers through the login flow. The code was never the hard part. And her one regret isn't technical either: she'd have shown it to twenty people early instead of letting one or two clients be the only feedback. IN THIS EPISODE 00:00 – It's 2026 — why am I still doing this in Excel? 01:18 – A school bus in Mexico, with Starlink on the roof 07:46 – Teaching yourself to build when nobody's teaching you 09:24 – Prompting is briefing 10:40 – Why nine out of ten go looking for a course instead 16:51 – Building the dashboard for one client 19:25 – The API paperwork nobody warns you about 24:09 – The moment a third person says "I've never seen this" 29:18 – What marketers build that developers don't 44:40 – How much do you actually need to understand? 49:45 – If you're still the one in Excel DEEP DIVE LINKS 📖 Full article — the argument, not a recap, with every quote timestamped to this episode: https://pirateskills.com/insights/build-it-for-one 📊 Hebelytics (Gianna's product — connect your marketing tools, find the leak): https://hebelytics.de 🏴‍☠️ Vibe Coding Cologne — Wednesday 2 September, 18:30, STARTPLATZ Cologne: Gianna is on stage giving a live demo of the thing she built. She came to the last one as an attendee. Free: https://pirateskills.com/events/vibe-coding-cologne/2026-09-02 🔗 Gianna on LinkedIn: https://www.linkedin.com/in/gianna-maria-hodel/ 🎧 Spotify: https://open.spotify.com/show/7kFl0BJiOYih8yqQTRR9SN 🍎 Apple Podcasts:https://podcasts.apple.com/us/podcast/build-grow-repeat/id1870561487 🗓️ All upcoming events: https://pirateskills.com/events 💬 Connect with me on LinkedIn: https://www.linkedin.com/in/bensufiani/

There's a whole level past ChatGPT, and it opens the moment you stop learning tools and start handing over tasks. The same sentence comes up over and over at the vibe coding events I host in Cologne: I've been messing around with ChatGPT, and I'm here to learn what happens beyond. This is the answer to that sentence, recorded for exactly those people. Beginner-friendly. The mindset shift comes first: don't think in tools, think in the tasks you repeat every week. From there it's three levels — Chat is a genuine thinking partner that has never seen your actual work. Co-Work gives an agent a folder and connectors and hands back mostly finished work. Code stops doing the task and builds the tool that does it. Same chat window every time. Nothing to relearn. You'll see all three inside Claude and ChatGPT side by side, the connector list behind the co-work level (Attio, Gmail, Calendar, Drive, Linear, Notion, Slack, Stripe), and the real Claude Code conversations that run Pirate Skills — ads, CRM, campaigns, events, funnel, security review. Plus one starter prompt per level, and the five checkboxes that tell you whether you've hit the chat ceiling already. Not everyone needs level three. Every business has a floor, and level two unlocks a lot on its own. IN THIS EPISODE 00:00:00 – Beyond ChatGPT: who this is for 00:00:50 – Stop thinking in tools, start thinking in tasks 00:02:02 – Chat, Co-Work, Code: the three levels 00:02:25 – Level 1 (Chat): what it's great at, and where it runs out 00:03:53 – The ceiling test: five signs you've outgrown chat 00:05:17 – Pick one recurring task, and name your business model 00:06:44 – The same task at all three levels 00:08:45 – Level 2 (Co-Work): treat the AI like a colleague 00:11:39 – Level 3 (Code): little machines that run without you 00:16:05 – One environment, three levels — nothing to relearn 00:16:33 – AI Compass, the 7-day plan, and where to go next DEEP DIVE LINKS 📖 Full Article (with the interactive Task Walker, the ceiling test and the 7-day plan): https://pirateskills.com/insights/beyond-chatgpt 🧭 AI Compass — Wednesday 26 August, 18:00–19:00 Berlin: free online session, once a month, same spine every time. Bring one recurring task, find your floor, and leave knowing which level you actually need: https://pirateskills.com/aicompass/2026-08-26 🎧 Spotify: https://open.spotify.com/show/7kFl0BJiOYih8yqQTRR9SN 🍎 Apple Podcasts: https://podcasts.apple.com/us/podcast/build-grow-repeat/id1870561487 🗓️ All upcoming events: https://pirateskills.com/events 💬 Connect with me on LinkedIn: https://www.linkedin.com/in/bensufiani/

What if the technical co-founder you've been hunting for the last two years is a chat window you haven't opened yet? Ben Schönenstein isn't a coder. He's an architect who spotted a problem in a café in 2017, assumed he was too late until AI got good enough that "maybe I can just do it myself" stopped being a fantasy. Two years of arguing with models later, he's shipped a laptop-friendly café finder (Kofi) and Matchbox, a problem-first product search engine already holding 17,000+ apps with a growth loop that fills its own gaps. This one goes deep on the honest version of the vibe-coding story. You'll see Matchbox live — type any problem, however big or small, and watch it surface real solutions instead of the heaviest SEO spender. You'll see the personal bug-tracker he built in a day that fixes its own bugs with a single button press and a merged pull request. And you'll get the full stack he'd reach for tomorrow. IN THIS EPISODE 00:00:00 – Two Bens in Cologne 00:00:38 – From architect to accidental vibe coder 00:06:22 – Do you still need a technical co-founder? 00:12:03 – Margin as a moat & the real bottleneck: distribution 00:16:12 – Matchbox live: type a problem, get real solutions 00:18:35 – The bug + idea apps that fix themselves 00:24:03 – "AI isn't a tool, it's a reflection" 00:25:15 – Running parallel work streams in Claude Code 00:31:57 – How to decide what to actually build 00:37:47 – The full stack he'd start any app with 00:40:38 – Just start: the advice, and where to build with us DEEP DIVE LINKS Find Ben Schönenstein on LinkedIn: https://www.linkedin.com/in/ben-sch%C3%B6nenstein-928ab6158/ His app Co-Fe finds cafés where working is actually welcome: https://co-fe.de/en/ Matchbox, have a real problems, and find matching products: https://askmatchbox.com/ 🎧 Spotify: https://open.spotify.com/show/7kFl0BJiOYih8yqQTRR9SN 🍎 Apple Podcasts: https://podcasts.apple.com/us/podcast/build-grow-repeat/id1870561487 🗓️ All upcoming events: https://pirateskills.com/events 💬 Connect with me on LinkedIn: https://www.linkedin.com/in/bensufiani/

AI has never been more capable — and the people winning with it in B2B are the ones who know exactly which work to hand over and which conversations to show up for themselves. Matthew Whyatt runs Tech Torque, sits as a fractional CMO for software companies, and has sold into the biggest mining companies, governments and agencies in the world. He uses Claude every single day — and he'll still tell you that the moment he spots AI slop in your post, he's out. Not because AI is bad. Because nobody curated the output. This conversation goes to both ends of it. Matt walks through his actual setup: a Claude briefing that reads his email and calendar before every meeting, a research analyst project that turns a URL into a one-pager his client can take to the C-suite, and the trick he found last week — talk at Claude via Whisperflow for five minutes, then have it write its own project name, description and instructions. Paste, done, works first time. There's also the Whisperflow reality check: 700,000 words spoken in 57 weeks on Ben's side, 46,000 in five weeks on Matt's. And then the other end. The $300,000 "maximum unviable product" Matt spent three years building and never launched, because he never asked anyone if they'd buy it. Goldratt's bottleneck, and why piling more ideas at the front of the machine slows the machine down. Daniel Priestley's rule: if you can't find a hundred people who'd buy it, don't build it. Ben's version: no code before 1,000 subscribers — two of six apps hit a million users. The through-line: when anyone can ship software over a weekend, features stop being interesting. Features are expected. The speed at which you can build and transfer trust is the speed at which you can build your business. IN THIS EPISODE 00:00:00 – Capable vs. trustworthy: the gap nobody talks about 00:01:44 – AI slop, broken trust, and what has to stay human 00:06:26 – Matt's Claude setup: daily briefings and a research analyst 00:09:26 – Whisperflow: 700,000 words in 57 weeks 00:12:43 – The trick — let Claude write its own project instructions 00:17:06 – OpenClaw, rabbit holes, and "the juice isn't worth the squeeze" 00:21:25 – Goldratt's bottleneck and the $300,000 unviable product 00:27:35 – When anyone can build, trust is the product 00:29:19 – Marketing does the lifting, sales gets one shot 00:35:56 – CRMs: Salesforce to Zoho, HubSpot, and going AI-native 00:39:27 – What to automate first, and how to reach Matt DEEP DIVE LINKS 🔧 Matt's company — Tech Torque: https://techtorque.co/ 💬 Connect with Matthew Whyatt on LinkedIn: https://www.linkedin.com/in/matthewwhyatt/ 🤖 Vibe Coding Cologne — Wed Aug 5: monthly meetup, short talks and real builds, in person in Cologne: https://pirateskills.com/events/vibe-coding-cologne/2026-08-05 🎧 Spotify: https://open.spotify.com/show/7kFl0BJiOYih8yqQTRR9SN 🍎 Apple Podcasts: https://podcasts.apple.com/us/podcast/build-grow-repeat/id1870561487 💬 Connect with me on LinkedIn: https://www.linkedin.com/in/bensufiani/

This week I'll show you how to build an AI company brain that's friendly even for non-techies. The whole idea is to connect all your processes — your CRM, your calendar, your inbox, every tool you use — with Claude Code at the center. You work on the brain, and the brain coordinates your agents and your team. For me and Christina it's been a massive productivity boost, and none of it takes exotic tech to set up. Because here's the quiet truth: everyone's arguing about "agent memory" right now — vector databases, graph engines, courses from Google and Anthropic — but the thing that actually runs my business is a folder of plain text files, and I never wrote a line of database code to build it. We'll go screen by screen through the real setup: the dashboard that tells me what to do next, one ongoing conversation per work stream, and the four kinds of plain files that make up the brain — docs, skills, journals, and dashboards. You'll watch it rewrite its own documentation, and see a live example of turning 20 conversations from a startup night into tracked follow-ups. It's not fully hands-off, and I won't pretend it is — nothing sends, spends, or publishes without me. But none of this is infrastructure. It's a writing habit with tools attached, and you can start this week with a single folder. IN THIS EPISODE 00:00:00 – Intro: an AI company brain for non-techies 00:01:00 – Work out of a coding agent (Claude Code at the center) 00:02:25 – One conversation per work stream 00:04:45 – Every conversation needs context + tools 00:06:20 – Documents & skills, not a vector database 00:07:41 – Journals & dashboards, built to your needs 00:08:30 – A real example: 20 leads from a startup night 00:10:03 – A brain that rewrites itself 00:11:49 – You don't need to be technical 00:12:25 – Permissions, trust & the review gate 00:14:05 – Where to start + upcoming events DEEP DIVE LINKS 📖 Full Article: https://pirateskills.com/insights/company-ai-brain 🧪 Pirate Lab — Wed July 15 (online): we build your company brain together, live — bring one work stream you want to automate: https://pirateskills.com/pirate/lab/2026-07-15 🏴‍☠️ Vibe Hackathon Cologne — Sep 4–6: 48 hours building with AI. Early bird €60 until July 31 (then €90): https://pirateskills.com/events/vibe-hackathon-cologne/2026-09-04 🚀 Pirate Forge — become your own vibe coder & vibe marketer. Next cohort starts Sep 30: https://pirateskills.com/pirate/forge 🗓️ All upcoming events: https://pirateskills.com/events 💬 Connect with me on LinkedIn: https://www.linkedin.com/in/bensufiani/

Two weeks ago a pitch deck wasn't on my agenda. Then an investor evening opened up – and the price of admission was a deck. So Christina and I decided to raise a small pre-seed for Bridesmaid, our agentic wedding planner, and I built the whole thing the way I now do most of my work: one Claude Code conversation on the left, the deck rendering live on the right. Research, investor targeting, the slides, original imagery, a full cohort-based 12-month financial model in Google Sheets, and a print-clean PDF – all in one unbroken session. Work that used to mean a designer, an analyst, a couple of tools and several days. Here's the part I want you to take away, because it's the opposite of most "AI makes your deck" advice: the agent will happily one-shot you a beautiful, generic, wildly over-optimistic deck – and that deck loses you the meeting. The value isn't the generation. It's the steering. In this video I walk through the real Bridesmaid deck, the strategic fights I had with the agent, and where a founder's judgment does the work the model can't. This is the actual deck we're sending to investors. Real numbers, real targets, nothing staged. IN THIS EPISODE 00:00:00 – Intro: the real deck we're using to raise for Bridesmaid 00:02:21 – Start in the repo, not a slide tool (and why HTML) 00:04:34 – The key move: make it interview you 00:05:54 – What Bridesmaid is + the €250k / €1.5M pre-money ask 00:07:54 – B2C or B2B? Reframing the model with the agent 00:09:32 – Choosing the investors: NRW.Bank, neoteq, TS Ventures 00:14:49 – Framing the deck for specific investors: the 10 questions 00:21:10 – Make it attack the deck from the investor's side, cut the b******t 00:22:44 – The money slides: building the model in Google Sheets 00:29:53 – Educated fantasy land: don't trust the AI's first numbers 00:36:44 – Walking the final deck, top to bottom DEEP DIVE LINKS 📖 Full Article: https://pirateskills.com/insights/ai-pitch-deck 💍 Bridesmaid – the working prototype (7-day free trial): https://bridesmaid.love 🧪 Pirate Lab — Wed July 8: this week it's all about the pitch deck. Free live session where we put this insight into practice – bring your build and turn your own strategy into an investor-ready deck with the crew: https://pirateskills.com/pirate/lab/2026-07-08 🗓️ All upcoming events: https://pirateskills.com/events 🎧 Spotify: https://open.spotify.com/show/7kFl0BJiOYih8yqQTRR9SN 🍎 Apple Podcasts: https://podcasts.apple.com/us/podcast/build-grow-repeat/id1870561487 💬 Connect with me on LinkedIn: https://www.linkedin.com/in/bensufiani/

What does it actually take to build AND grow a real software product solo in 2026 — while raising kids and picking up the odd contract to stay afloat? In this episode of Build.Grow.Repeat., Ben sits down with Jamie Murphy, a developer-turned-product-manager who used Claude Code to ship Still Mind, an AI-guided meditation and journaling app that started as a weekend favor for his wife and turned into a growing business. It's an honest, tactical conversation about the full-stack builder life: not just writing the code, but owning the go-to-market, the SEO, the Apple relationships, and the tax-and-admin grind too. No hype, no overnight-success story — just the real mechanics of going zero to one. WHAT YOU'LL TAKE AWAY • Why "generalists" — not pure developers, designers, or PMs — are best set up to win in the age of vibe coding • Scratching your own itch: how Still Mind grew out of a real problem instead of keyword research • Spec-driven development with Claude Code, and why a mature "harness" makes AI feel 10x better • The honest stack call: Flutter + Firebase + RevenueCat, and what he'd do differently today • The relationship edge — booking WWDC editor slots and getting noticed by Apple when everyone can ship an app • Vibe marketing that actually works for a no-authority app: technical SEO, ASO, and AEO • How he built his own AI harness for Answer Engine Optimization — and got 60% more traffic referred from AI • Bootstrapping vs. raising: the spectrum from Pieter Levels to MicroConf to Y Combinator, and the 10k MRR line CHAPTERS 00:00 Meet Jamie: developer → product → solo builder 02:17 Why kids push us to build 04:05 Vibe coding's ideal builder: developer, designer, or PM? 09:32 Still Mind: building from a real itch 13:28 Idea → App Store → first paying customer 17:21 Bootstrapping vs. raising & the 10k MRR line 21:55 How he builds: Claude Code + spec-driven development 25:57 The stack — and what he'd change 28:22 The Apple relationship edge: WWDC, editors, Design Awards 35:18 Vibe marketing: technical SEO, ASO & content 39:47 Building an AEO harness (+60% AI traffic) ABOUT THE GUEST Jamie Murphy — builder of Still Mind, an AI-guided meditation & journaling app. 🧘 App: https://getstillmind.com/ 🌐 Website: https://jamie-murphy.com/ 🧵 Threads: https://www.threads.com/@making.jamie ABOUT BUILD.GROW.REPEAT Build.Grow.Repeat. is the Pirate Skills podcast for founders and builders who want to build products people actually want, grow them with AI, and do it as a small-but-mighty team. Hosted by Ben Sufiani. 🏴‍☠️ Pirate Skills: https://pirateskills.com 🎧 Listen on Spotify: https://open.spotify.com/show/7kFl0BJiOYih8yqQTRR9SN 🎧 Listen on Apple Podcasts: https://podcasts.apple.com/us/podcast/build-grow-repeat/id1870561487

A/B testing used to mean a paid tool, a developer, and weeks of setup — and the best tools kept leaving the market. Google killed Optimize; VWO and AB Tasty moved their free tiers upmarket toward enterprise. Startups got left behind, guessing and redesigning on a hunch. Now it's one Claude Code session through PostHog's MCP. You say one sentence, and the agent builds the whole experiment — feature flag, 50/50 split on a single URL, server-side with no flicker, consent-safe, internal traffic filtered — and you keep the judgment calls. Things I used to do with a lot of trickery on the site, I didn't have to touch. This is move 6 of 6 in the Growth Codex series — conversion optimization, the finale. You'll see the real Bridesmaid test: the conversation-vs-checklist variant, a directional lift of roughly +14% completions / +11% CTA clicks that never crossed statistical significance at our traffic, and the honest pirate posture — throw spaghetti early when traffic is thin, get scientific as the stakes scale. Plus the copy-paste prompts to set up and conclude your own test this week. The honest part: at our traffic the result was directional, not a proven winner — it would take the better part of a year to reach full significance, so we shipped on the totality of the signal and logged the call. None of that is hidden here. IN THIS EPISODE 00:00:00 – Intro: A/B testing got complicated, and the best tools left 00:00:37 – The tools we kept losing: Google Optimize → VWO → AB Tasty 00:01:32 – PostHog + Claude Code to the rescue 00:02:24 – What an A/B test is: one URL, two variants on Bridesmaid 00:03:16 – Behind the scenes: built via the PostHog MCP, nothing by hand 00:04:13 – The setup prompt: goals, consent, no flicker, internal filtered 00:06:15 – Don't guess: watch the session replays 00:08:17 – Throw spaghetti, then get scientific 00:10:40 – Reading the dashboard: the signup-completion drop 00:14:12 – Six moves, one self-optimizing engine 00:16:49 – Your spaghetti test: run one this week DEEP DIVE LINKS 📖 Full Article: https://pirateskills.com/insights/vibe-ab-test 🧪 Pirate Lab — Tue June 23: free weekly live session where we put this week's insight into practice — watch me set up a full A/B testing and conversion stack live with Claude Code and PostHog, no code, no dashboards: https://pirateskills.com/pirate/lab/2026-06-23 Pirate Night - Celebrate 10 years of Pirate Skills in the coop I have been waiting for, with no other than Manuel Koelman, the founder of Pirate X and host of the legendary Pirate Summit, join us here: https://piratesummit.com/pirate-night/ 🎧 Spotify: https://open.spotify.com/show/7kFl0BJiOYih8yqQTRR9SN 🍎 Apple Podcasts: https://podcasts.apple.com/us/podcast/build-grow-repeat/id1870561487 🗓️ All upcoming events: https://pirateskills.com/events 💬 Connect with me on LinkedIn: https://www.linkedin.com/in/bensufiani/

What if setting up a proper Meta ads campaign took minutes in a chat instead of hours clicking through Business Manager? Meta's official Ads MCP connects your ad account to Claude or Codex. You describe the campaign – objective, audience, budget, creative angles – and the agent builds the full structure in minutes, named and tracked from the start, with everything paused until you launch it yourself. Experienced marketers finally ship the complex setups they never had time for; beginners get a far gentler on-ramp. The optimization layer is where it gets unfair. The agent reviews performance against Meta's numbers AND your own database – and that's how we caught an ad with the best click-rate in the whole account and zero actual signups. Paused it, moved the budget to the ad that converts. You'll see the real Bridesmaid campaign: the fake winner, the ghost-signup problem, the leak in the signup form itself, and the copy-paste prompts to run all of it yourself. The honest part: a few finishing steps still live in the Ads Manager UI, and at a small budget the numbers come with caveats. None of that is hidden here. This is not theory. It's a live campaign, real spend, real misses. IN THIS EPISODE 00:00:00 – Intro: putting my Meta ads on a loop 00:00:46 – 10 years of clunky Business Manager – meet the Meta Ads MCP 00:02:51 – Two unlocks: pros ship faster, beginners ramp gentler 00:03:16 – How it works: describe the campaign instead of clicking 00:05:39 – Inside the real Bridesmaid campaign + creatives 00:09:43 – Meta vs your own database: clicks aren't customers 00:17:30 – Put the review on a loop – on a leash you control 00:19:48 – What the loop found: fake winner, ghost problem, sign-up leak 00:28:58 – What's next: the A/B test + free-trial → paid 00:31:13 – Where to learn more: Pirate Lab, Hackathon & Forge DEEP DIVE LINKS 📖 Full Article: https://pirateskills.com/insights/meta-ads-loop 🧪 Pirate Lab — Wed June 17: free weekly live session where we put this week's insight into practice – bring your project and run an agent-driven ad setup with the crew: https://pirateskills.com/pirate/lab/2026-06-17 🎧 Spotify: https://open.spotify.com/show/7kFl0BJiOYih8yqQTRR9SN 🍎 Apple Podcasts: https://podcasts.apple.com/us/podcast/build-grow-repeat/id1870561487 🗓️ All upcoming events (free): https://pirateskills.com/events 💬 Connect with me on LinkedIn: https://www.linkedin.com/in/bensufiani/