A pattern appears whenever generative computational methods enter a creative technical field.
Generation becomes cheap. Selection becomes the real work.
Medicinal chemists have lived with this dynamic for years.
Historically, designing molecules was a deeply generative task.
Chemists sketched structures. They explored modifications. They proposed the next compounds in a series from experience and intuition.
Then computational tools arrived.
Enumeration engines could generate thousands of analogues. Docking tools produced ranked lists. Property predictors filtered molecules before they were ever made.
Suddenly the limiting factor wasn’t ideas. It was judgement.
Chemists increasingly found themselves scanning lists of machine-generated molecules, filtering out the unrealistic ones, identifying the few worth making, and steering the direction of the series.
The job quietly shifted from drawing molecules to deciding which molecules mattered.
The conversation around AI coding tools sounds very familiar.
Developers describe starting with generated code and then shaping it.
They review. They adjust. They correct assumptions. They guide the model toward the intended design.
The workflow looks strikingly similar to computational chemistry pipelines.
The machine proposes.
The expert decides.
The core expertise didn’t disappear. But its location moved.
When generation becomes cheap, expertise migrates upstream into:
In medicinal chemistry this judgement determines which molecules are synthesised.
In software it determines which systems are actually built.
In both cases the bottleneck becomes taste, strategy, and direction.
For many medicinal chemists, this pattern will feel familiar.
The arrival of generative tools doesn’t remove the need for expertise.
It changes where that expertise is applied.
When generation becomes abundant, the real skill becomes navigating the space of possibilities.
And that is a problem medicinal chemists have been solving for a long time.
