三位工程师 72 小时用 GrokBot 做出付费产品

Carnage @0xCarnagee原推发布于 2026.09.18

在 X 上打开原推

Three SpaceXAI engineers built a company in 72 hours with GrokBot They started with a blank repo and no idea what to build - Three days later it was taking payments I rebuilt their whole process into 30 prompts - one complete loop from blank page to first paying customer: FIND THE SIGNAL 1. Ask me 5 questions to pin down one reachable buyer, one recurring problem, and a deadline I cannot move. 2. Read these sources and extract every repeated complaint, current workaround and verbatim quote. Link each one to where it came from. 3. Rank the problems by urgency, how easily I can reach the buyer, what they already spend, and how long a prototype takes. 4. Turn the strongest one into a single testable claim: who needs it, what outcome they want, why they would pay. 5. Write 7 interview questions about the last time they hit this problem - what they tried, what it cost, what they did instead. 6. Compare the alternatives: existing products, manual work, doing nothing. Find the one gap a prototype could prove in a week. 7. Shape one offer: buyer, deliverable, price hypothesis, delivery window. Make the promise specific enough to be wrong. 8. Find 10 real prospects from public sources, explain why each fits, and draft the opening line for every one. 9. Design the kill switch: a validation test with a deadline, a success number, and the condition under which I stop. 10. Give me a one-page verdict - build, revise or stop - and name the assumption I still have not tested. BUILD THE SYSTEM 11. Design the smallest agent team that covers this. Each one gets a single output, its required inputs, and a definition of done. 12. Write the shared brief: customer, problem, offer, deadline, constraints. Mark every line as confirmed fact or assumption. 13. Set up the workspace so I can see in ten seconds what shipped, what is moving, and what is blocked. 14. Build the shared memory: facts, preferences, decisions and open questions, each dated and sourced. 15. Define the handoffs. What research passes to design, what design passes to engineering - artifact, acceptance criteria, unresolved questions. 16. Close one loop end to end: plan, act, observe, evaluate. Set the retry limit and the condition that escalates to me. 17. Draw the boundaries. What each agent can touch, and which actions wait for my approval before money moves or anything ships. 18. Write the review checklist that catches invented facts, broken behavior and work that only looks finished. 19. Produce a daily update: shipped, learned, blocked, next. For every blocker, name the exact decision I have to make. 20. Audit the whole thing for duplicate work, unnecessary handoffs, and automation that costs more than it saves. SHIP THE OFFER 21. Cut it to one customer journey and one useful outcome. Decide what to build, what to do by hand, what to postpone. 22. Draft the landing page: one benefit, real evidence, one call to action. Only claims we can back up today. 23. Specify the shortest path from visitor to promised outcome, and name the three places it will break first. 24. Make signup work: essential fields only, a confirmation that says what happens next, and it has to be clean on mobile. 25. Design the version we deliver manually, so we learn which parts customers actually care about before automating them. 26. Prepare to take money: the offer, the amount, delivery terms, receipt. List what must exist before the first payment. 27. Walk the full journey as a stranger - discovery, signup, payment, delivery. Rank anything that stops completion. 28. Write 5 personalized outreach messages that say honestly where the product is and ask for one concrete next step. 29. Track the funnel: visits, real conversations, signups, paying customers, successful delivery. Find where people leave. 30. Review the evidence. Pick one improvement, one feature to postpone, and one assumption to test next. PLAN → ACT → VERIFY → LEARN → REPEAT The 72-hour version costs $20 a month. Start with the one question your business still cannot answer.

这条是引用着另一条写的：x.com/0xcarnagee/status/2090102148344262659