The excitement around humanoid robots in the industrial arena is understandable. Spaces were built for people, and a human-like form can move through those spaces and attempt a wide range of jobs. But industrial automation shouldn't begin with how closely we can replicate a person; it should begin with the task.

If the job is moving 60-pound cases across a flat warehouse floor, a wheeled robot is hard to beat. If the job is sorting goods, building store-ready pallets, or coordinating thousands of machines in a shared environment, the critical questions are different: Can the system perform safely? Reliably? At scale?

Advances in AI do not mean every robot needs to become a general-purpose machine. The real opportunity is to make the intelligence generalizable: software that helps machines perceive their environment, learn from experience, and make better decisions within a controlled scope. That approach preserves what industrial systems require most—the physics and proven safety controls that matter when software directs physical machines.

That same intelligence can then orchestrate a fleet of machines built for different tasks, planning around constraints, adapting as conditions change, and treating the system (not the individual robot) as the unit of performance. Each machine can focus on the job it was designed to do and do it exceptionally well.

Biology is a useful source of inspiration; it shouldn't become a design constraint. A machine doesn't need a human silhouette to create value. In many industrial settings, designing around the task produces a system that is faster, safer and more practical.

Physical AI will advance as capable software is paired with machines purpose-built for specific, real-world work. The future of automation is not one form factor for every job. It is the right machine for the task, and the intelligence to orchestrate them together.

James Kuffner, chief technology officer at Symbotic