Robotics won’t get a ChatGPT leap — it will get software that teaches factory robots task by task
Robotics won't get a ChatGPT moment — it'll get reliability, one factory task at a time
Mbodi co-founder Xavier Chi argues the real unlock is natural-language teaching of industrial robots, not humanoid hype
The interesting AI-in-robotics story right now is not humanoids — it is software like Mbodi's letting industrial robots be taught new tasks via natural language, which attacks the actual bottleneck: costly custom programming, not lack of general-purpose bodies.
Why no single breakthrough moment
Chi's argument is that robotics differs from language models: physical deployment means every factory, warehouse, and task has its own constraints, so progress will look like incremental reliability gains across many narrow deployments rather than one general model suddenly working everywhere. That is also why he's skeptical the current humanoid excitement is proportionate — general-purpose bodies don't solve the harder problem of teaching and reliability.
Mbodi's reliability claim
during an eight-hour test
Mbodi's path so far
- 01Startup Battlefield demo
Live robot demo made Mbodi a crowd favorite
- 02Partnership with ABB Robotics
Working to bring Mbodi's natural-language software into factories and warehouses
- 03Reliability test
99.6% success rate over an eight-hour run, positioned as proof of factory-floor readiness
Why industrial robotics struggled to scale as software
Old constraint
- Heavy custom programming per task
- Each deployment is bespoke, hard to repeat
- Physical-world variability breaks generic solutions
- Hard to build a scalable software business on top
What generative AI changes (Chi's claim)
- Natural language lets non-engineers teach new skills
- Software layer can generalize across tasks/robots
- Faster iteration without re-programming from scratch
- Opens door to recurring software revenue, not just hardware
What's asserted vs. what's shown
99.6% success rate over an eight-hour test — no detail on task type, environment, or how ‘success' was defined.
Mbodi is working with ABB Robotics to bring its software into factories and warehouses — stage of deployment not specified.
Chi argues humanoid excitement is not proportionate to near-term utility, and that robotics will improve gradually rather than via a ChatGPT-like leap.
What VCs are reportedly screening for
- 01
How founders de-risk an investment before fundraising
- 02
Evidence of reliability/testing beyond demos
- 03
Scalability of the software layer across deployments, not just single robot demos