The Signal · Column 043

Biology is becoming engineering, and hiring follows

Andrei van Noordt · 31 August 2026 · 2 min read

Vijay Pande spent years deploying a roughly four billion dollar biotech practice at a16z. Last year he walked away from it to start something deliberately smaller. His reasoning, laid out this week, is that biology is finally shifting from a discovery science to an engineering one, that clinical trials remain brutally expensive whatever the models promise, and that open shared datasets rather than walled off proprietary ones are what will let AI actually change medicine. We are not doing thirty bets a year, he said of the new firm.

Most people will read that as a venture story about fund size. I read it as a statement about labour. A discovery science pays for rare intuition and tolerates a high failure rate, because nobody can say in advance which experiment will land. An engineering discipline pays for something else entirely: the ability to make a result reproducible, to specify it, to hand it to a system and have it come out the same way next Tuesday. Those are different people. They interview differently and they cost differently.

When a field turns from discovery into engineering, the scarce person stops being the one who finds the answer and becomes the one who can make it happen twice.

The open data argument sharpens it further. If the datasets everyone builds on are shared, then the data stops being the moat and the team reading it becomes the entire advantage. That is an uncomfortable kind of clarity for anyone building a hiring plan. You cannot buy a defensible position with a proprietary corpus and average people. You need the small number of hybrids who hold real domain depth on one side and genuine fluency with the models on the other, and there are not many of them in any market I recruit in.

For hiring leaders, the practical read is simple. If your field is crossing from discovery into engineering, stop sizing the team by how many experiments you want to run and start sizing it by how many results you need to make repeatable. That is usually a smaller headcount at a much higher bar, and the bar is the part people get wrong.