Why Robotics Is a Separate Revolution
Why progress in language models is not enough to establish the timing of useful, reliable, general-purpose physical automation.
Model & Argument · Initial edition, November 2026
Software intelligence and embodied intelligence are related but different transitions. A model that can reason about a task may still be unable to complete it safely with motors, sensors, tools and physical objects.
Robots operate in the world of friction, uncertain surfaces, changing environments, damage and wear. Successful deployment requires reliable perception, planning, control, manipulation, power management and maintenance.
Why robotics could be enormous
A robot that can perform a useful range of ordinary physical jobs at acceptable cost would change the economics of small enterprises, farms, buildings, transport and production. It could make local automation feasible without the specialised engineering historically needed to build a separate machine for every task.
Its significance would not be limited to replacing particular occupations. It could change where work can be done and who can afford to operate productive equipment.
Why timing is difficult
Robotic demonstrations are not the same as affordable continuous operation. Reliability, exception handling, maintenance, energy use and the range of tasks that can be performed matter at least as much as headline demonstrations.
Progress may arrive unevenly. Some applications could become economical long before a genuinely versatile household robot exists.
Why local AI matters
A machine that cannot function when a distant cloud service is unavailable is not fully autonomous. Safety-critical control must remain local; for useful general-purpose independence, much perception, planning and handling of routine exceptions may also need locally available intelligence.
Cloud services could remain valuable for unusual problems, training and updates. The open question is how much everyday work can proceed without them.
This joins Robots Need Local Intelligence with The Local Mode of Production.
We need benchmarks grounded in real task completion, all-in cost, downtime and maintenance—not just model capability curves.