NOT VIBE CODING

Build a software startup without writing code.

A proven team workflow, run by AI agents. You make the decisions.

A free template that gives you the first hires of a startup as AI agents: a project manager, an architect, developers, a tester, security, legal, growth and social media. They plan, build, test and launch with you, and nothing goes live without your yes.

Use this template Read the README

github.com/Fadi88/one-man-startup

Made for Claude Code, Codex, Gemini CLI and Cursor. MIT licence.

Excerpts, word for word, from the recorded setup dry run with a fictional founder: she asks to go lighter on the gates and the setup agent recommends strict and explains why; it recommends deleting receipt photos once read; setup ends with nothing committed, published or sent, and the test sweep green.

Replayed word for word from a recorded dry run with a fictional founder.

This is not vibe coding

Vibe coding

Describe what you want, accept whatever the AI writes, ask again when something breaks. Fast for a demo. It tends to end in an app that half works, that nobody understands, and that gets more fragile with every fix.

This template

The structure established software teams use: written down before it is built, designed before it is coded, tested by someone other than whoever built it, and released only when you approve. The agents fill the roles; the workflow is the traditional one.

What happens here	The established practice
Before anything is built, you are interviewed about who needs it, the smallest useful version and what happens when it goes wrong.	Requirements elicitation
Everything the product must do is a numbered requirement, and every change names the requirement it serves.	Requirements traceability
Building can't start until the spec is confirmed and no question on it is still open.	Definition of Ready
An architect designs the system and records each hard-to-reverse choice with its reason.	Architecture Decision Records
Developers write the test before the code.	Test-driven development
A separate tester checks the work against the requirements and can stop a release.	Independent QA
A security reviewer looks for ways data could leak or be misused.	Threat modelling
Every change re-runs every check, and a round ends only when they all pass.	Continuous integration
Nothing goes live, costs money or deletes data without your approval, one step at a time.	Release approval
Decisions, and the ideas you turned down, are written down with the reason.	Decision log

What changes is who does the work, not how it is done.

From idea to launch

Five steps, each a standard software practice: set up, get grilled, build, check, and you say go.

Your AI team

You talk to one agent, the project manager. It brings in the others when they're needed.

You talk to the project manager agent, which hands work to the plan, build, check, grow and save groups.

What stays in your hands

Agents read, draft, write code and test on their own. Going live, spending money, deleting, sending messages, prices and legal risks wait for your yes.

How it's built

Plain files in your own GitHub repository. Nothing is locked into this template or any one AI tool.

Your AI coding tool reads the instructions, writes the paper trail in docs/ next to your code, and runs the checks.

Getting started

Install your AI coding tool, then ask it to check that Git and Python 3.11 or newer are installed.
Click Use this template and name the repository after your product.
Open your copy in your AI tool.
Type /setup in Claude Code, or ask any tool to run the setup interview from skills/setup.
Ask for your first thing, for example a landing page with a waiting list.

The agents still make mistakes, and the tests and approvals won't catch all of them. The legal agent writes checklists and drafts for a lawyer, not legal advice. Support for Codex, Gemini CLI and Cursor is written but not yet tested; see what has been tested.