An AI agent choosing a treatment and controlling a surgical robot sounds like science fiction. Exin Therapeutics says it tested that idea in a mouse model of epilepsy — with a human surgeon, a supervising veterinarian and an emergency stop still in the loop. This 104-second explainer looks at what the team reported: selecting a therapy from 19 candidate vectors, using scientific papers to plan a target, operating a motorised rig and checking camera images. It also explains the reported differences between Claude Opus and Sonnet, and why the agent's refusal to retry a movement matters. The authors say the mouse recovered. That is not yet evidence that the treatment reduced seizures. This video reports the researchers' account; it does not independently validate a treatment or describe a procedure to reproduce. PRIMARY REPORT Gabriel Ocana-Santero: https://x.com/ocanasantero/status/2115442964574908902 Exin Therapeutics: https://www.exintherapeutics.com/ MADE WITH FRAMEWRIGHT Created by Anton Biletskiy-Volokh using Framewright, my project for repeatable, agent-assisted video production. https://github.com/markoblogo/framewright More of my work: https://abvx.xyz/ The supplied laboratory photos and interface screenshot are credited to the reported Exin experiment. They are animated through camera moves, crops and explanatory overlays; this is not a recording of the operation. English narration uses local Supertonic 3. Visuals use PixiJS and HyperFrames, with a reused original instrumental soundtrack. #AIAgents #Robotics #Neuroscience