A new job showed up at the bench. Nobody posted it.

We keep meeting the same person in different labs. A working scientist, deep in the science, who now spends part of the week training a model on their own analytical judgment. No one wrote the job description. They call themselves AI Creators.

Watch how they work and the pattern is consistent.

The model starts beside them, reading every sample and proposing a call. The scientist reviews the same samples in parallel and compares. Where the model's call matches theirs, they let it clear that kind of result on its own. Where it doesn't, they teach it: a handful of examples, a threshold they had never had to state, an edge case neither of them has seen before.

The scientist's review narrows. Weeks later they are not reviewing every sample, only the exceptions the model sends back.

This is the cadence of governance, and it is how automation gets built in a wet lab. Where expert judgment carries the risk, trust and accuracy are built by the same act: a scientist looking at a call and deciding whether it holds. Nothing is handed over until they have watched the model get it right.

The training runs both ways, and that is the part nobody expects. Which calls come back tells the scientist what to teach next, and whether it belongs to this one batch or to every future batch. The scientist trains the model, and the model shows the scientist where their own standard was never written down.
One AI Creator, a few months in production on the platform, said it plainly last month: "The model is as good as I am now. Probably better."

What do you need to see before you stop reviewing every result yourself?

#AnalyticalChemistry #DrugDevelopment #ExpertIntelligence #AIinTheLab #LabAutomation