Sebastian Apelgren

To get to a world where every disease is cured and there's a medicine for everything, we need far more experiments than people can run. That means many dark labs: lights off, only robots and agents working, around the clock.

Europe's draft rulebook for AI in drug manufacturing says LLMs "should not be used in critical GMP applications." The industry pushed back, and the EU is reworking the text now. My bet: the bar won't become "trust the model." It'll become "show me where every number came from."

The FDA's first warning letter about AI in drug manufacturing has my favorite line of the year. The firm skipped a required step and told the FDA it didn't know, because the AI agent "never told you it was required." AI can write your SOPs. It can't take the inspection for you.

The software running regulated labs is 20 to 40 years old. None of it was built for AI. Soon every one of them will have to answer one question: which AI do you use, and how is it controlled? Whoever becomes the safe answer keeps that spot for a decade. We intend to be it.

Anyone can build an AI that writes a lab report. The hard part is the inspector who points at one number two years later and asks where it came from. Every regulated lab will need one approved place where AI can touch the record. That's what we're building at Ordo Labs.