A prototype is an instrument with an end date

"Did you really throw everything away, or only the code?" Someone asked that on a French forum, under a post from a developer who vibe coded an app for a month, about 20,000 lines of code, and then threw it all away. AI-built prototypes look finished, so the usual move is to call them version one and keep going. But in Veracode's 2026 report, AI-generated code had an average security pass rate of 56%. This episode is about what a prototype is actually for: learning what's worth building, before a development team builds it for real. What you'll hear: Why the handoff in product discovery is knowledge of what to build Build before you build: vibe code to prove it's worth building, then let a development team build the real thing The four-line contract to put at the top of every prototype When a prototype can do a real job, and when it can't Try this week: before your next build, write the four-line contract: the test, the one path, what it must never hold, and the signal that ends it. Next episode: Four people alone had twice the ideas of four together. I'm Marcin Klimaszewski and I run Product Build Day. Send me your question on LinkedIn, and it might become an episode. Sources: ⁠LinuxFr: J'ai vibe-codé une appli pendant un mois, puis j'ai tout jeté (11 September 2026)⁠ ⁠SweClockers forum thread (29 July 2026)⁠ ⁠Veracode: GenAI Code Security Report (30 July 2025)⁠ ⁠Veracode: 2026 GenAI Code Security Report (28 July 2026)⁠ ⁠WIRED: Thousands of vibe-coded apps expose corporate and personal data on the open web (7 May 2026)⁠ ⁠Security Boulevard: the RedAccess scan of about 380,000 apps (7 May 2026)⁠ ⁠Product Build Day Warsaw, July 2026