The dangers of the wrong AI.
The HVAC bot that quoted $400 on a $4,000 job. The law firm chatbot that fabricated a Florida statute. The CPA tool that emailed the wrong client. Each is one missing guardrail away from a clean build.
Three near-misses from inside our own portfolio. The HVAC quote bot in Cedar Park that priced a 4-ton system replacement at $387 because a customer misread a sticker. The Boca Raton estate planning chatbot that confidently cited a 2022 Florida elective share amendment that does not exist. The Houston CPA's AI tool that attached the wrong Michael Chen's tax summary during a 40-meeting peak week. None of these failures was about the AI getting smarter. Each was about a missing deterministic verification step at the highest-risk action: the outbound send. The full longform, with the four guardrail patterns we deploy by default, ships as part of the When AI Fails 2026 Report in June 2026.
Coming online in 2026. The roadmap below is committed, the writing is queued.
Related reading
- 01
The HVAC quote bot that quoted $400 on a $4,000 job.
Austin, TX
The misread sticker, the $150 deposit, and the $500 human-in-the-loop gate that fixed it.
- 02
The law firm chatbot that made up case law.
Boca Raton, FL
Why scope containment beat smarter AI.
- 03
The accountant whose AI sent the wrong tax summary.
Houston, TX
Two clients named Michael, one outbound email, a regulatory event.