Start Here
Founders rarely fail for lack of effort; they fail by pouring months into the wrong thing for the stage they're at. This skill's only job is to tell you where you are and the single next thing to do, then get out of the way. It also runs the one test that decides whether you're building a company or a feature.
The method
Diagnose the stage, run the Remove-the-AI test, route to one skill. Full build-arc map, the test in
full, the symptom-to-skill table, a worked diagnosis, and anti-patterns:
references/stage-diagnosis.md.
Step 1: Run the Remove-the-AI test
Ask: take the AI out of your product, does it still work? If yes, you have an AI feature, not an AI-native company. Before anything else, find the closed loop that would make it break without the intelligence. If there isn't one, that is the work.
Step 2: Locate the stage
Place the founder on the build arc by what they can't yet answer:
| If they can't answer… | They're at | Run next |
|---|---|---|
| "What's the sharp, falsifiable bet?" | Frame | frame-the-hypothesis |
| "Who's the market, and where's the regulatory/competitive wedge?" | Frame | map-the-terrain |
| "What's the system, before any code?" | Architect | architect-before-code |
| "How do I build it with agents without it collapsing?" | Build | agentic-build-loop |
| "How is this safe enough to act on a human?" | Trust | eval-and-safety-harness |
| "What's the smallest agent that proves the loop?" | Ship | design-the-mva |
| "Is it actually working?" | Grow | measure-ai-native-pmf |
Step 3: Route to the earliest unvalidated assumption
When two stages fit, send them upstream. The highest-value work is almost always the riskiest belief they haven't tested yet, not the most exciting one.
Output
- The result of the Remove-the-AI test, stated plainly (native or wrapper, and what's missing).
- The founder's current stage and the one skill to run next, with a one-line reason.
Constraints
- Route to exactly one next skill. A list of five is the same as no answer.
- Don't skip the Remove-the-AI test to be encouraging: a wrapper diagnosed early saves months.
- Stay theme-agnostic; the founder supplies the domain.
Dictionary
the Remove-the-AI test · AI-native vs wrapper · data flywheel
Copy-paste version
For non-coders: paste into any chatbot (Claude.ai, ChatGPT, Gemini).
Act as my AI-native build coach. My idea: [ONE_LINER]. Who it's for: [ICP]. Domain: [DOMAIN].
1. Run the "Remove-the-AI test": if I take the AI out, does my product still work? If yes, tell me
I have a feature not an AI-native company, and help me find the closed loop that would make it
break without the intelligence.
2. Diagnose my stage: Frame, Architect, Build, Trust, Ship, Grow, or Defend - based on the riskiest
thing I can't yet answer.
3. Tell me the ONE next thing to do, and why it's the highest-value work right now. Just one.
Don't flatter me; route me to the earliest unvalidated assumption.