The AI GTM stack for solo founders.
The actual systems I run to do go-to-market without a team — outreach, qualification, content, strategy. Built in Claude Code. Not prompts you paste once. Not theory. The real projects, with the numbers attached.
Most “AI for founders” content hands you a prompt and wishes you luck. Most GTM advice is written by people who’ve never had to run it alone. I’m a founder doing my own go-to-market with no sales team — so everything here is a working system I actually depend on, not a tip. Four principles hold all of them together.
Systems, not prompts
A prompt is a one-off. A system is a directive file Claude reads every session, calling deterministic scripts in order. The difference is whether it compounds or resets each time you use it.
Research before writing
Every system inverts the usual order. It gathers the specific context first — then generates. That single inversion is why the output reads human instead of templated.
Claude orchestrates, Python executes
The decisions are Claude. The deterministic work — API calls, parsing, file ops — is Python. That separation is why these don't compound errors the way pure-LLM pipelines do.
A human gate before anything ships
Every system has one manual review step where you can still say no. Automation gets you to the 90% draft; the gate keeps the 10% that would've sounded like a bot from going out.
Four systems. One operating motion.
Each one owns a stage of the founder GTM motion. The first is downloadable today — the whole project, file by file. The rest are documented here and shipping as resources next.
The Research-First StackFree download
Cold outreach that reads like you spent 20 minutes on each company — because a system did. Apollo pulls leads on growth signals, Tavily researches every company, Claude writes from the brief, you send manually, a 90-second Loom closes. The whole project, six files, free.
138 messages → 11 Looms → 1 client · 9% close rateGet the 6 files free →AIOS Lead EngineDocumented
Five-stage LinkedIn qualifier. Scrape profile + posts → score ICP-fit → analyse the conversion gap → draft the DM → batch-upload to the outreach tool. The scoring step is what stops you wasting Looms on people who were never going to buy.
Production prompt-pipeline · audited end-to-endX Growth AgentDocumented
A nine-agent system that runs an X account end-to-end — Sentinel watches, Scribe writes, Judge evaluates before anything publishes, Publisher ships, Engager replies, Analyst reports. A Debugger self-heals on three-strike failures. Runs daily on a €6.80/mo budget.
Running daily · 707-line specAuroraDocumented
A multi-agent founder strategist. Feed it the company dashboard and the CEO memos; it red-teams the strategy and generates a CEO-ready pitch deck from raw data. Built in a hackathon to prove how far agentic delegation goes when the structure is right.
Pitch deck from raw data in under 3 hoursEvery system is a 3-layer Claude Code project
Not a chatbot workflow. A directive file tells Claude what to do. Python scripts do the deterministic work. Claude reads the directive, calls the scripts in order, handles errors, and updates the directive when it learns something. The same architecture runs outreach, qualification, content, and strategy — only the directive changes. That’s the whole idea: one agentic pattern, pointed at every part of the funnel.
Building the version of this every founder can run.
These systems are how I run my own GTM. The product I’m building — Founder Copilot — is the version you won’t have to assemble yourself. Start with the free outreach stack; that’s the system that opened the door for me.
Start with the Research-First Stack →