Preflight — a pre-shipping checkpoint for AI builders

Preflight Score

/100

Validate first

Your next move: The user and problem are sharply framed and the riskiest assumption is cheap to test, but there's almost no evidence yet that builders will pause to use it. Run the intercept test before investing further in the build.

This score reflects how clearly the idea is framed, not whether it will succeed.

Summary

A checkpoint that helps AI builders surface their riskiest assumption before they build the wrong thing. The central risk is behavioral: builders who can ship in an afternoon may not pause to run a structured check first, even if they agree it's useful.

Riskiest Assumption

Solo AI builders will stop to run a structured check in the moment before building, instead of opening Cursor and building immediately.

Why it's the riskiest: Everything else in Preflight only matters if builders actually pause to use it. Shipping is faster and more fun than reflecting, so the default behavior works against the product — this is the belief most likely to be wrong and cheapest to test this week.

24–48h Validation Experiment

The mid-idea intercept test

24–48 hours

Catch 8 AI builders at the exact moment they have a new idea and measure how many run a Preflight Check before they write any code.

1.Find 8 builders actively shipping with AI tools (Cursor/Lovable/Bolt communities, indie hacker Discords).
2.When one mentions a new idea, send them the Preflight link and ask them to run it before building.
3.Watch whether each one completes a check before writing code, or builds first.
4.For everyone who completes it, ask: 'What, if anything, will you do differently now?'
5.Log the count that ran it pre-build and the count that named a concrete changed action.

Success signal

At least 5 of 8 builders run a check before writing code, and at least 4 of them name a specific action they changed because of it.

Evidence Map

The assumptions to test, risks to manage, questions to ask, and actions to take next.

Assumptions

Solo AI builders will pause to run a 5-minute check instead of opening their editor and building immediately.

Confirmed if a builder mid-idea completes a check before writing code; weakened if they say 'useful' but skip it and build anyway. This is the behavior the whole product depends on.

Risk: highEvidence: weak

Builders find the riskiest-assumption output sharp enough to change what they do next.

Confirmed if a builder can name a concrete action they changed after reading it; weakened if they call it 'generic' or 'what I already knew'. Generic output makes Preflight feel like a ChatGPT wrapper.

Risk: highEvidence: none

Builders will run the 24–48h experiment Preflight hands them rather than just reading it.

Confirmed if a builder reports actually running the experiment within two days; weakened if the check is read once and closed. Value only lands if the experiment gets executed.

Risk: mediumEvidence: none

Builders will share a public check with a teammate or advisor to align on what to validate.

Confirmed if public share links get opened by a second person; weakened if checks stay private. Sharing is the main way Preflight spreads without paid growth.

Risk: mediumEvidence: none

Risks

Preflight reads like generic AI prose and feels like a ChatGPT wrapper.

Mitigation: Force every assumption to name a specific user group, behavior, and the evidence that would confirm or weaken it. Reject vague output like 'users may not want this.'

Builders treat the score as a verdict on whether the idea is good, then dismiss the tool when they disagree.

Mitigation: Frame the score as clarity of framing, not likelihood of success, and label recommendations by validation readiness ('Validate first'), not idea quality.

The check is read once and never acted on.

Mitigation: Lead with one riskiest assumption and one copy-pasteable 24–48h experiment, so the next action is obvious and small.

Questions

1.Walk me through the last time you had an idea and started building — what happened between the idea and the first line of code?
2.Did you check anything before building? What, and why that?
3.When was the last time you built something and realized later it was the wrong thing?
4.What would have to be true for you to spend 5 minutes before building instead of starting right away?
5.Who would you send a check like this to before you started?

Actions

1.Run the mid-idea intercept test with 8 builders this week.
2.Tighten the prompt so the riskiest assumption is never generic — name the user, the behavior, and the evidence.
3.Instrument the share flow to see whether public checks get opened by a second person.

Building something with AI tools?

Run your own Preflight Check

Paste an idea, answer four questions, get your riskiest assumption and a 24–48h experiment.

Run your own Preflight Check

Not an idea validator. A pre-shipping checkpoint.

PF-SAMPLE · Created with Preflight — the pre-shipping checkpoint for AI builders