Physical AI is attracting enormous attention, and most of it is aimed at general-purpose labor. We think the value appears first somewhere narrower and more specific: precise, high-value physical work, on lines that change constantly.

The prize in physical AI is high-value, high-precision assembly. But high-priced goods run at lower volumes — serving them means high-mix, low-volume production, which is exactly where fixed automation stops paying for itself.

The error-cost argument compounds from there. As component prices rise, the cost of a single mistake rises with them, and consistency stops being a quality metric and becomes a financial one.

To deliver high-value services, automation has to be built for high mix.

Software-defined cells stand up, reconfigure, and relocate fast. What used to be a fresh engineering project for every product change becomes a change in software.

That collapses time-to-market and removes the bottleneck that limits most automation programmes today: the small number of engineers who can commission and re-commission a line.

Capacity should be something you deploy, not something you rebuild.

For decades the answer to rising manufacturing costs was to move production somewhere cheaper. Costs are now rising everywhere, and the hidden costs of offshoring — freight, inventory, quality drift, IP exposure, lead time — are no longer hidden.

Add labor scarcity and the reshoring pressure that follows, and the buy-versus-offshore maths flips.

Flexible automation is what makes the domestic option viable.