Start training for the lab job AI is changing

Ask what AI does to a scientist's job and the honest answer is that the job is not one thing to automate or protect. It is a bundle of separate tasks, and each task in that bundle moves at its own pace. A week in a life sciences lab holds a search through the literature, a protocol drafted before the first reagent is weighed out, a plate of wells pipetted by hand, a dose-response curve fitted once the reader has finished, and a decision about which of several replicates belongs in the write-up. Some of that bundle already runs on software, some of it goes faster with help and stays a person's call, and a task or two did not exist five years ago at all. That is why the effect of AI on a lab depends on which of those tasks you are looking at, on how quickly the lab actually adopts what is available, and on how fast the tasks making up the job today get swapped for the ones that will make it up next. Two labs running the same kind of science can end up with very different years.

The tasks move at different speeds

Counting a plate's low-signal wells, checking that a catalogue number resolves to a live product page, and formatting a results table are the kind of task that hands over cleanly, because the correct output is fixed and checking it is fast. Fitting a curve, drafting a protocol, and reviewing an experimental design for a weak control go faster with a second pass from software, but the person doing the work still decides whether the fit makes sense against everything else they know about the system, or whether the flagged control actually matters for this assay. A newer kind of task has appeared beside both of those: deciding which of Elnora's skills should be trusted with which piece of work, and checking that what it produced is grounded in a real source and not a plausible guess. None of those groups is going away, and none of them is standing still either.

Why adopting now is worth doing

A lab that waits to see how this settles falls behind the labs already moving tasks across that boundary and learning where the mistakes are while the stakes are still small. The tasks that make up a scientist's job in ten years will not arrive all at once as a single new job description. They arrive one task at a time, and the ones that show up first are the ones a lab is already prepared to hand over or already knows how to check. Starting now gives a lab a say in which tasks move first, ahead of being handed a version of the job decided somewhere else, on someone else's timeline.

Keep the tasks that build judgement

The same shift raises a question worth answering on purpose, while adoption is still moving fast enough to get ahead of it. If nobody studies computer science or writes a line of code on their own, the skill of coding without help disappears from a generation, not because anyone decided that, but because nobody kept practising it. The same holds for the base science underneath a lab: if training stops asking a student to run the assay by hand, to sit with a plate that looks wrong before any software says so, the field keeps the results and loses the instinct behind them. Updating a curriculum on purpose is how a university or a lab decides which tasks stay a taught skill and which ones become something a person only directs, and holding both at once is a reasonable, achievable thing to ask of the next few years of training.

What a lab keeps by building its own methods

A lab that wants Elnora to run its own method has to write that method out in enough detail that another agent could follow it, then author it as a skill the rest of the group can call by name. Writing a method to that level of detail is a different discipline than asking a general question and reading back whatever comes, and it asks nearly the same thing of the person writing it that running the method by hand used to ask: know it well enough to explain every step. The same pattern shows up on the automation side of the lab. Elnora's lab automation skill writes a liquid-handling protocol against the actual deck it will run on, so a task that used to mean pipetting by hand becomes specifying the steps precisely and checking that the machine carried them out the way the assay needed. Both are examples of the same trade: the manual execution moves to software, and the understanding that used to sit inside the execution has to move somewhere else that still gets taught and still gets used, which is what a curriculum updated on purpose is for.

None of this argues for slowing down. It argues for choosing on purpose, adopting the tasks that are ready to move and keeping the training that lets a scientist tell a good result from a plausible one, in the lab and in the classroom where the next generation is still learning to look. A lab that starts that work now gets to keep the speed and the skill together, and decides which tasks change on its own terms.