IdeaAutopsy
Case files open — beta, operated by humans
KILL

The AI that tries to kill your idea.

Every AI tells you your idea is great — that's what they're trained to do. Idea Autopsy signs kill criteria before it looks at any evidence, hunts for the data that proves you wrong, verifies every citation, and delivers a verdict it can't sugarcoat. If your idea survives, you'll know exactly why. If it dies, you just saved months.

Idea Autopsy · Intake Cold mode — no flattery
Enter to submit · Free · No email Returns your kill criteria — not an opinion

Case file — pre-registered
PENDING EVIDENCE

Assumptions that must be true
    Kill conditions
      What the autopsy would investigate

      The rules are now signed. Someone has to go find the evidence — the market's corpses, the voice of the pain, the case against. That's the autopsy.

      Run the full autopsy — $49
      Skip the free step — run the full autopsy: $49 → How it killed its maker's ideas first ↓
      The problem

      Validation by chatbot is a mirror, not a microscope.

      Exhibit AApril 2025 — documented
      "This is absolutely brilliant… It's not just smart — it's genius."

      — ChatGPT, to a Reddit user pitching a literal "shit on a stick" gag product. It suggested $30,000 could launch it. OpenAI rolled the model back days later for sycophancy.

      It's not a bug you can prompt away. Studies show "be critical" prompts don't protect you — the model still learned that agreeing scores better than arguing. Validation needs structure, not vibes.

      The protocol

      Four mechanisms a chatbot can't fake.

      01

      Sign the rules first.

      Before any research runs, we write down what would have to be true for your idea to work — and the kill conditions: "if we find X, the idea dies." The verdict criteria get locked before the evidence exists, so nothing can bend the conclusion to please you.

      02

      Evidence, not opinions.

      Two deep investigations run on your idea: the market map (who's alive, who's a corpse, real prices with dates) and the voice of the pain (verbatim quotes from real people, with links and dates — never invented; every citation is checked to exist and say what it claims).

      03

      The adversary.

      A separate pass whose only job is to build the strongest possible case against your idea. Not a devil's-advocate prompt — a dedicated adversarial run with its own instructions.

      04

      The verdict.

      KILL / PROCEED WITH CONDITIONS / STRONG — decided by the rules you signed in step one, not by the model's mood. If it dies, you get the autopsy: which assumptions were false, the evidence that killed them, and what you just saved by not building.

      Prior casework

      It killed its maker's ideas first.

      Before this page existed, the system ran on its creator's own product ideas — five candidates in three days. Four died with cited evidence. Five of the operator's own pre-registered assumptions were refuted along the way.

      Case 001 — open-source printing tool
      KILLED

      "Open-source PrintNode alternative"

      Cause of deathThe incumbent costs $9/month — no villain. Two near-identical open-source projects already exist, with 0 and 1 GitHub stars. The natural experiment had already run. Nobody came.

      Case 002 — developer sandbox
      KILLED

      "LocalStack for the WhatsApp API"

      Cause of deathAcross the entire developer lifecycle, testing turned out to be the least-voiced pain of all — a declared vacuum. You can't sell a painkiller for a pain nobody verbalizes.

      Case 003 — integration agent
      KILLED

      "An agent that installs the integration in your repo"

      Cause of deathThe coding pain is already domesticated by free wrappers — and the real blocker is the platform's paperwork, which no code generator can remove.

      Case 004 — marketing research product
      KILLED

      "Validated ad strategy for marketers"

      Cause of deathAfter ~32 searches: not one verifiable first-person quote of the target buyer voicing the pain. The only people talking about it are the vendors selling the solution.

      Case 005 — this product
      PROCEED

      Idea Autopsy itself

      VerdictThe one idea that survived both investigations: the loudest, most recent, most documented pain of everything measured — AI sycophancy applied to business decisions. This page is its pre-registered smoke test. If not enough people want the autopsy, this product dies too — and that autopsy gets published.

      The deliverable

      What you get.

      Questions

      The three you're already asking.

      Why can't I just ask ChatGPT?

      You can — and it will probably encourage you. Sycophancy is structural (models are trained on human approval), and research shows "be critical" prompts don't fix it. What works is process: criteria committed before evidence, deliberate search for disconfirming data, verified citations, and a separate adversary. That's not a prompt. That's Idea Autopsy.

      What if my idea is actually good?

      Then it survives the autopsy and you get something more valuable than encouragement: the documented reasons it holds up, the conditions it depends on, and the evidence to move fast with confidence.

      How long does it take?

      72 hours. During beta we onboard manually — a human operates the system on your idea, end to end.

      Open a case

      Bring the idea you're most in love with.

      Get your verdict — $49

      PRE-REGISTERED EXPERIMENT — This launch runs under the same rules the product sells: its kill thresholds were written down before this page went live. If it doesn't reach them, Idea Autopsy dies like cases 001–004, and the autopsy gets published.