The regulatory clock, the Wrong Pocket problem, and why governing AI will require AI Midnight is different for everyone. An AI regulation can be enacted today, become effective months later, impose one obligation next year…
What an unusually candid conversation about healthcare economics revealed and why AI may change the equation rather than merely rearrange it. About halfway through an evening of healthcare policy and economics in New York, I…
At Health Tech Investor Day, Rachel Kern described a potential investment that fell apart over a question considerably less glamorous than anything in a technology presentation: how was somebody actually going to pay for it?…
What I heard at CTA’s fourth annual Health AI+ Conference On September 9, I spent the day at the Consumer Technology Association’s Health AI+ Conference in Washington, D.C., including joining Jesse Ehrenfeld, Laura Adams, and…
What research ecosystems have to learn after everyone agrees to collaborate On August 21, 2026 I had the privilege of participating in the launch of the Amity Global Research Hub at the Consulate General of…
A document can pass basic validation and still fail under challenge. That is the uncomfortable point the Red Team Analyst prompt is designed to address. A Fresh-Eyes audit can tell you whether a finished work…
The easiest way to misuse AI is to let the same workflow generate the work and then treat its own self-review as validation. That sounds obvious when stated plainly. Humans do not usually treat self-review…
Most AI-generated dossiers fail in one of two ways. The first failure is harmless but not very useful: the model produces a dressed-up LinkedIn summary. Current role. Prior roles. Education. Maybe a few public quotes.…
Most AI-generated company research is dangerous, dangerous in a very specific way: it can be plausible, polished, and generic enough to be mistaken for decision support. Ask an LLM to “research this company,” and it…
Foundations In the prior essay, I argued that metaprompting creates leverage on prompt construction. A good metaprompt helps turn rough, messy, human intent into a structured assignment that a large language model can actually execute.…
