Post by Alvaro Cuba

For two years the story in robotics was "the AI isn't good enough yet." That excuse is evaporating. This week Alibaba open-sourced Qwen-Robot — one model family that drops into humanoids, arms, and quadrupeds. Genesis launched Eno, a wheeled robot built for software-first prototyping. What that unlocks: you can stress-test task logic, integration, and failure modes in a Docker sandbox before you buy any hardware. The most expensive mistake in robotics — committing to the wrong form factor and finding out in month four — becomes a cheap, early test instead. I've watched this same shift play out in agent pipelines. Once the intelligence layer commoditizes, the hard part moves to the unglamorous scaffolding — CI, safety checks, observability, the plumbing that keeps a fast loop trustworthy. Robotics is about to inherit the lesson software teams already paid for: when prototyping gets cheap, integration discipline becomes the bottleneck. If you're building multi-robot or multi-agent workflows — what breaks first when iteration outruns your validation stack?