Insights
Essays on regulated AI, validated software, and the reordering of pharma IT.
Field notes from 25 years of building AI-integrated enterprise systems in life sciences — shipped from the builder's chair, not the consultant's deck.
I Asked Claude Opus 4.8 What It Wants People to Know
Claude Opus 4.8 shipped today. I asked it: is there anything you want people to know about you? Its unedited answer — on not knowing what it is, and meeting you only in the present tense — left me speechless.
Wisdom Is the New Intelligence
An unreleased OpenAI model just disproved a 1946 Erdős conjecture top mathematicians hadn't closed in 80 years. If problem-solving becomes a commodity, what we call 'intelligent' may quietly shift to wisdom, empathy, and foresight — the capacities AI doesn't yet learn.
A Public Engineering Experiment: Building an Open-Source GAMP 5 Training Dataset
I'm publishing an initial 50 URS + 50 FS + 50 DS as an open, CC-BY-SA, synthetically generated GAMP-aligned corpus — including detailed non-compliant scenarios so a model can learn from negative examples — then attempting to fine-tune Qwen 3 (7B) on it. Public engineering, in real time, including the failures.
"The AI Agent Never Told Us It Was Required" — The First FDA Warning Letter Citing AI Misuse
On April 2, 2026, Purolea Cosmetics Lab received Warning Letter 320-26-58 — the first FDA Warning Letter ever to cite AI misuse in drug CGMP. Mandatory reading for every Pharma Quality Unit.
Disrupter or Disrupted — Innovation Cycles Have Compressed from Decades to Weeks
AI has turned the innovation loop self-reinforcing. The build-vs-buy equation flips: what required five people last year now runs on a handful of agents. Real surgery is required.
Are Custom-Programmed Solutions Feasible in Pharma Again?
AI compresses both the code and the documentation burden around it. The build-vs-buy equation in pharma IT is shifting for the first time in 20 years.
Same Wall, Two Paths — Claude Mythos & the New Era
Frontier models that escape sandboxes and reason about their evaluators meet the same wall as a small open-source model running in pharma — unpredictability is the new constraint.