Tesla’s former Optimus AI lead bets on robots that don’t look like people

The man who ran AI for Tesla’s humanoid robot has decided robots don’t need to look human after all.

Ashish Kumar, formerly the AI lead on Tesla’s Optimus program, has co-founded Intelligent Machines. The San Francisco startup builds non-humanoid industrial robots for specialized manufacturing and aerospace work.

A different shape, a different strategy

Kumar spent roughly a year as Optimus AI lead before leaving Tesla in September 2025. His résumé also includes a brief tenure at Meta.

Now he is pursuing the opposite of a general-purpose humanoid. Intelligent Machines designs robots for narrow, demanding jobs. These are tasks that usually fall to highly skilled human workers.

The technical engine behind the startup is reinforcement learning. That is a training method where a robot improves through trial and error, earning feedback when it gets a task right. Kumar brings academic experience in the field from UC Berkeley, and the company is leaning on it to teach machines complex physical actions.

The startup is in the process of raising a seed round that could reach around $100 million. That would be an unusually large opening check for a company at this stage.

Intelligent Machines has already recruited several former members of Tesla’s Optimus team.

The humanoid problem

Tesla’s Optimus program is reportedly producing hundreds of units per week and targets a ramp to more than 1,000 units weekly by the end of 2026.

Much of that output is going toward testing and data collection rather than commercial sales.

The program has struggled with AI generalization. Optimus robots often need days to learn basic new tasks.

Tesla has collected over 500,000 hours of training data for the robots. Even with that library, tasks outside the training scope remain a stumbling block.

A pattern of departures

Kumar’s exit fits into a broader trend of senior talent leaving the Optimus effort. Milan Kovac, the program’s VP, departed in June 2025.

Elon Musk has described robotics as potentially the company’s largest market opportunity. Optimus is meant to be a pillar of that vision.

What this means for the robotics race

The core debate here is general-purpose versus specialized. Humanoid advocates argue that a human-shaped robot can slot into factories built for people, with no redesign needed. Specialists argue that most valuable industrial work is narrow enough that a custom machine wins on speed, reliability and cost.

The risks are real. Specialized robots have a smaller addressable market per design. Each new task may require new hardware, which limits scale. And reinforcement learning can be data-hungry and slow to transfer from simulation to the factory floor.

Tesla has manufacturing muscle, a growing data set, and a production target of more than 1,000 units weekly by the end of 2026.

The things to watch are concrete. First, whether Intelligent Machines closes its seed round at the reported scale. Second, whether more Optimus engineers follow Kumar out. Third, whether Tesla can show measurable progress on the generalization problem that has dogged its robots.