Build one core workflow (no dashboard, settings or polish) in weeks 1-2, give interested people free access in weeks 3-4 asking them to break it, then fix and ignore off-core requests in weeks 5-8. If they don't return a second time, keep iterating; target 5-20 paying users.
Take the landing page that collected the most emails and ask what you can build in exactly one week (not two weeks or a month); the limit forces you down to the core value prop. Ship the rough version to real people.
He used early GPT-3 to cluster and categorize viral tweets into a startup-tweet library viewed in a CLI, used it to post, then launched a single search-bar feature on Twitter and Reddit.
His 8-step stack: DALL-E/Canva/Figma for creatives, Carrd or Webflow for the site, Bubble for the app, Stripe or Paddle for payments, Notion+Zapier for ops, Product Hunt + Submit media list for launch, Ahrefs/Testimonial.to for marketing, Tally/Hotjar for feedback.
When it applies: testing many ideas quickly, non-technical or speed-first builders
When a prominent person complains publicly about a problem (Pieter Levels on fake MRR screenshots), build a dead-simple fix within hours (one headline, two buttons, a leaderboard) and launch the next day by quoting the original tweet so you ride its context.
When it applies: You can ship an MVP in hours (boilerplate/AI coding)