Insights
Where generation outruns validation.
Writing on AI verification and agent autonomy — from a production mobile app run partly by unattended AI agents, to what the same failure modes mean for pharma insights and market research. A short series, posted as each piece is ready.
The $440,000 sentence nobody read
AI drafting has outrun review capacity everywhere — pharma research and production software both. Last year that gap cost Deloitte a $440,000 refund.
I built the check on my own system first
Two weeks running AI agents with real autonomy on a live production app. Three rules that made it survivable — and why none of them is "trust the AI more."
Three ways an AI-drafted insight is wrong
And why "it cites its sources" catches none of them. A veteran's-eye view of what breaks first in market research.
Why "it cites its sources" isn't verification
The obvious objection, and why it's wrong — plus what the EU AI Act's August disclosure rule actually requires (and doesn't).
Prove me wrong in an afternoon
An open invitation: run the check against a study you already know the answers to. Designed so it can visibly fail.