Plan Any App Idea With AI

Most people open AI and ask: "How do I build X?" That's backwards. The magic move: make AI interview YOU first. You know things about your idea you've never said out loud. These three prompts pull them out, attack them, and turn what survives into a plan.

Prompt 1, The Interview (AI asks, you answer)

Paste this into Claude (or any strong model):

You are an experienced product strategist. I have an idea I want to build. Do NOT give me advice yet. Instead, interview me, one question at a time, waiting for my answer before the next question. Cover: who exactly this is for, the moment they'd reach for it, what they do today instead, why existing options disappoint them, what "wildly successful" looks like in 12 months, and what I'm uniquely able to bring to this. Push back when my answers are vague. After ~10 questions, write a one-page summary of the idea AS I DESCRIBED IT, no embellishment.

Why it works: answering out loud exposes the gaps politely-nodding friends never will. Expect question 4 or 5 to sting a little. That's the good part.

Prompt 2, The Attack (pressure-test)

New chat. Paste the summary from Prompt 1, then:

You are a skeptical investor who has seen 1,000 ideas like this fail. Attack this idea seriously: the 3 most likely reasons it fails, the hidden costs I'm not seeing, who else does this and why users stay with them, and the single riskiest assumption I'm making. Then, and only then, tell me the ONE version of this idea you would actually fund, and why. Be specific, not brutal for sport.

Why it works: you want the eulogy before the funeral. The "one version I'd fund" line at the end is often worth the whole exercise, it forced my app from "outfit randomizer" toward "premium AI stylist."

Prompt 3, The Build Plan

Same chat, after the attack:

Now turn the surviving idea into a build plan for one person working with AI tools. Break it into phases where EACH phase ends with something I can see working. For each phase: the goal, what exists at the end of it, what I can fake/mock instead of building, and what I must NOT build yet. Assume I will over-scope, cut aggressively. End with Phase 0: what I can do in the next 48 hours.

Why it works: "each phase ends with something I can see working" is the anti-overwhelm rule. And letting AI tell you what NOT to build yet is the discipline nobody has alone.

The 48-hour rule

Whatever Phase 0 comes out, do it within 48 hours. Ideas have a half-life. (Mine was: get one AI call to classify one photo of one shirt. That single working call became an entire app.)

One honest warning

My first AI-generated plan said 18 days. It took 4 months, because the plan revealed what quality actually required, and I chose quality. Plans aren't promises; they're maps that improve as you walk. Re-run Prompt 3 whenever reality diverges. That's not failure. That's steering.