Reports
Cover illustration for When AI Fails: 2026 Report..
Reports

When AI Fails: 2026 Report.

MIT found 95 percent of genAI pilots fail. Our own data on near-misses shows why, and the four structural fixes that prevent the failure.

Coming online in 2026Notes filed April 2026

The MIT 2025 finding that 95 percent of generative AI pilots fail is mostly explained by three patterns we see in field reports: deployment without a deterministic verification step at the highest-risk action, scope expansion past the model's trained domain, and operators expecting AI to fix a culture problem. The report compiles the HVAC quote bot, the law firm chatbot, the CPA tax summary near-miss, and three additional engagements not yet published, against the four guardrail patterns that hold. Ships in June 2026, with early access to operators on the field-notes list.

While you wait

Coming online in 2026. The roadmap below is committed, the writing is queued.

Related reading