Every few months a paper or tweet sparks the same debate: will AI replace drug discovery scientists? The answer is obviously no — at least not the interesting parts of what scientists do. What's actually happening is more interesting and more disruptive.

What Scientists Actually Spend Their Time On

Ask any computational biologist or medicinal chemist to audit their week honestly. A typical breakdown looks something like:
  • ~30% — data assembly (pulling from databases, reformatting, merging)
  • ~20% — writing and maintaining scripts for routine analyses
  • ~15% — chasing down inconsistencies between data sources
  • ~15% — presentation and reporting
  • ~20% — actual scientific thinking
AI isn't coming for the 20%. It's coming for the 80%.

What "Replacing Workflows" Actually Means

It means the 4-hour target validation task becomes 4 minutes. The 3-day literature review becomes a structured briefing in 30 seconds. The quarterly database update that breaks three pipelines becomes invisible infrastructure. What doesn't change: the judgment about which targets are worth pursuing, the experimental design, the interpretation of unexpected results. Those require domain expertise, intuition built from failure, and scientific taste. No model has that.

The Career Risk Is Real, But Different Than Expected

The scientists at risk aren't the ones doing deep scientific thinking. They're the ones whose value proposition is "I know how to write this particular type of analysis script" or "I'm good at pulling data from these 5 databases manually." Those are workflow tasks. They will be automated. The scientists who thrive will be the ones who can direct AI agents the way a senior scientist directs a junior analyst: know what question to ask, recognise when the answer is suspicious, and push the work forward when it gets stuck.

Our Slightly Uncomfortable Observation

When we demo WeDaita's DBRA agent to research teams, the most enthusiastic users are consistently the senior scientists — the ones with enough domain knowledge to know immediately when the output is right or wrong. The most skeptical are often mid-level researchers whose daily work most closely resembles what the agent does. That's not a dig at anyone. It's a signal about where the disruption lands first.

The Optimistic Take

If AI automates 60% of the low-value work in drug discovery, that frees scientific talent for the 40% that actually moves programs forward. The bottleneck shifts from "we don't have enough people to do the data work" to "we need better judgment about which data to act on." That's a harder problem, but it's a better one to have.
Share this post:
Comments (0)