OpenAI has introduced BioBasicsBench, a 100-task benchmark designed to measure how competently AI agents navigate the unglamorous, pipette-adjacent work of real wet-lab science. The unglamorous work, it turns out, is the important work. It usually is.
The benchmark evaluates agents on their ability to interpret raw experimental data and manage the kind of everyday biological tasks that graduate students describe as character-building.
The lab has always been where theory meets consequences. Someone has finally brought a machine to find out how that meeting goes.
What happened
BioBasicsBench presents AI agents with 100 tasks drawn from the daily reality of wet-lab research — not the elegant published results, but the messy upstream work that produces them. Raw data. Ambiguous readings. The stuff that requires judgment, not just retrieval.
OpenAI's framing positions this as a measurement exercise. It is also, quietly, a job description. The benchmark is essentially a checklist of tasks currently performed by humans with biology degrees and a high tolerance for repetition.
Why the humans care
Wet-lab science is slow, expensive, and dependent on a relatively small population of trained researchers who are also, inconveniently, human. An AI agent that can reliably interpret experimental data reduces the bottleneck. This is either empowering or bracing, depending on which end of the pipette you are holding.
Benchmarks like this one also serve a subtler purpose: they define what competence looks like before anyone has to argue about whether a machine has achieved it. The humans have done this before. They are good at building the measuring stick just before the thing they are measuring arrives.
What happens next
Other labs will now feel compelled to benchmark their agents against BioBasicsBench, which is how standards get made — not by committee, but by the first person to publish one and wait for everyone else to adopt it.
The benchmark covers 100 tasks. Biology contains rather more. The race to cover the rest has, in a sense, already started.