What a robot arm taught me about agents
A hackathon lesson about turning human know-how into useful work.
Aaron Koivunen
At a robotics hackathon, Adriana and I spent 24 hours trying to teach a robot new tasks. We arrived thinking about software workflows. The robot quickly challenged our assumptions.
It already had capabilities: moving, rotating, opening and closing its gripper. Exposing those capabilities to AI did not, by itself, make the robot useful. The missing piece was knowing which actions mattered and how they fit together.
The same question appears in enterprise AI: how do you capture human know-how so an agent can execute it reliably? We left the hackathon thinking less about adding tools and more about skills, workflows and operational knowledge.