MIT's NANDA initiative reported in 2025 that 95 percent of enterprise generative AI pilots delivered no measurable impact. The models are not the problem. The programs are. Enterprises are repeating a failure mode Eric Ries documented in 2011: one large bet, fully specified upfront, shipped a year later. The Lean Startup was written to kill that pattern. Most organizations never learned the lesson the first time.
The article makes a case that generative AI has made lean discipline more urgent, not obsolete. Building a working prototype now takes an afternoon. That collapses the cost of experimentation, which is exactly what Ries said was the constraint. The piece draws on CB Insights startup post-mortem data showing poor product-market fit as the leading cause of failure, connects it to Toyota's genchi genbutsu principle, Teresa Torres's continuous discovery framework, and the Google Ventures five-day sprint method. The argument is specific: demo-driven roadmaps are theater, not rigor, and the teams crossing the AI divide are amplifying work people already do, not building for boardroom applause.
The full piece is worth reading for its four action frameworks, which are concrete and operational, not advisory. It also makes a pointed argument about long-form design thinking: months-long IDEO-style engagements built for a slower world are now mostly ritual. The laws of UX did not change. The front-loaded ceremony around understanding users did not survive the afternoon prototype. That argument alone is worth the read.
[READ ORIGINAL →]