YC Decoded Explores Robot-Use Agents: Coding Agents Generalize to Robotics
On September 26, Y Combinator released a new episode of its Decoded podcast featuring the founders of Waddle Labs and RoboCurve, discussing the emerging trend of "robot-use agents": general-purpose large models that directly control different robots and write policies, learning new physical tasks with almost no robot-specific training. The discussion stems from MIT professor Philip Isola's recent essay, and YC is actively betting on startups in this direction.
Confirmed
- The episode's guests are the founders of Waddle Labs and RoboCurve, with the theme of agents generalizing from coding to the physical world.
- MIT professor Philip Isola argued in a recent essay that we may be entering the era of robot-use agents: general models can control different kinds of robots and learn new physical tasks with little or no robot-specific training.
- Both Isola and YC view the strong generalization of coding agents beyond software as one of the biggest surprises in AI in recent years, laying the groundwork for agents entering the physical world.
Why it matters
- If general models truly generalize across robot platforms, robotics development could shift from per-robot bespoke training to general agent-driven control, sharply lowering adaptation costs for new tasks and robot models.
- The migration path from coding agents to physical-world agents is where YC and its incubated companies are placing bets, with Waddle Labs and RoboCurve as representative startups worth tracking.
2026-09-26 ~ 2026-09-27 · 6 related posts
Primary sources
- [source] Y Combinator Podcast: Coding Agents Are Generalizing Into Robot-Use Agents — Y Combinator · 2026-09-26
- YC podcast explores robot-use agents as coding agents generalize beyond software — ycombinator · 2026-09-26
- YC adds "robot operator model" to roadmap as coding-agent generalization reaches robotics — ycombinator · 2026-09-27
3 near-duplicate retellings: ycombinator · ycombinator · ycombinator