Provision Labs launches in San Francisco to build tooling for physical AI
The AMW Read
A new tooling player joins the physical AI map with a specific data, eval and deployment focus, but no funding, customers or products are disclosed, so the impact stays within the sub-segment.
Provision Labs launches in San Francisco to build tooling for physical AI
Anjali Sardana, the founder of Pronto, has launched Provision Labs, a startup based in San Francisco. The company plans to build data, evaluation and deployment tools for AI systems that perform physical tasks. The announcement does not mention funding, investors, customers, named products or a launch date. For now, it is a statement of company formation and intended scope.
The company is not entering the hardware or model layer of robotics. It is entering the tooling layer underneath it. Physical AI differs from software-only AI in one important way: training data has to be collected from real or simulated environments, and evaluation has to measure behavior in the physical world, not scores on a static benchmark. Deployment also means getting models onto machines that operate under real-world constraints. Provision Labs is betting that these steps can become a standalone business, much as data labeling, evals and serving tools became separate categories around language models. The open question is whether companies building robots and embodied models will buy this tooling from outside vendors or keep it in-house, where it can serve as a competitive advantage.
For investors, Provision Labs is an early test of a horizontal tooling market in physical AI. Its value depends on how many separate teams are building physical AI systems and whether their needs are similar enough for one vendor to serve. Builders should watch which part of the stack Provision Labs ships first (data, evaluation or deployment) and whether early customers are robot makers, embodied-model labs or industrial operators. That choice will show where the company sees the most pressing bottleneck.