Speaking as someone who spends most of their time building automation platforms and teaching people to do it (enough that I built a whole class around it)... but who also spent six years of my life splitting stem cells every 36 hours... I am split on this one. On one hand, @mkoeris is right that the primitives are learnable and that novel combinations are not some sacred skill. On the other, @hsalis is 💯 that the large majority of experiments are far from autonomous. Most methods that are truly automatable, I can still do way faster and probably better with a multichannel pipette, I just have better things to do with my time. Automating bespoke science is the hard part, but that is where the most exciting progress is to be made. I'd encourage all to challenge themselves to think beyond automating the mundane and ask what science simply could not be done in the absence of a robot: real time responsive feedback, or changing many parameters at once based on the data as it is generated. The kind of experiment where no amount of extra hands would ever get you there. We are only scratching the surface there. (Currently grading midterm proposals for my automation class at the moment. One aim has to justify a genuine *need* for automation, and any proposal that justifies it with throughput, combinations, scale, or time is a big yawn. Think bigger, people 🦾)