Mbodi AI lets operators teach industrial robots new tasks in natural language, replacing the specialist programming that currently makes automation impractical for anything but high-volume, unchanging work. The economics of the factory floor are unforgiving here: reprogramming a robot cell requires an integrator, takes days, and has to be justified by production volume, so short runs and frequent changeovers stay manual even where a robot could physically do the job.

Shifting instruction to language moves the capability to the person who already understands the task, and the value shows up as changeover time rather than as raw robot performance. Backed since 2025 and a graduate of Y Combinator’s X25 batch, Mbodi is an early AI infrastructure bet on the interface layer of physical automation. Whether that interface generalises across hardware and tasks, and holds up under real factory conditions, is the question the next few years will answer. The prize, if the approach holds, is opening automation to the enormous share of manufacturing that runs in short, variable batches.