Industrial Next Wants Your Robots to Stop Being Reprogrammed
Two engineers who ran Tesla's autonomous factory left to sell that edge to everyone else. Their pitch is blunt: teach the robot by showing it, and retool a production line in minutes instead of weeks.
Walk into most factories and you will find robots that are brilliant at exactly one thing. Bolt this panel. Weld that seam. Change the part they handle, and the machine goes quiet while engineers rewrite it - a project that can run for weeks and cost more than the robot itself. Industrial Next, a San Francisco company from Y Combinator's Winter 2022 batch, was built to make that pause disappear.
The company's founders are not automation tourists. Allen Pan led Tesla's autonomous factory efforts at Fremont and spent more than a decade across GM, Momenta and the Chinese autonomy firm Neolix. His co-founder, Lukas Pankau, was Tesla's lead electrical architect for the Model X, 3 and Y, then spent three years at Waymo defining sensor architecture. They lived inside the fastest-moving factory in the car industry, and then left to sell its underlying advantage to companies that will never have Tesla's budget.
The problem hiding in plain sight
Automation was supposed to buy factories flexibility. In practice it bought them speed at the cost of rigidity. A conventional work cell is a stack of hand-written scripts tuned to one product, one orientation, one set of tolerances. It is fast and it is precise, and it is also brittle: nudge the part, introduce a new variant, and the whole thing needs a specialist to reprogram it. For a manufacturer juggling product refreshes, that reprogramming tax is the reason the line stands still.
Industrial Next's argument is that the tax is optional. Instead of scripting every motion, the company teaches a work cell the way you would teach a new hire - by demonstration and simulation. The robot watches, the model generalizes, and the cell adapts to variation instead of breaking on it.
- Hand-written scripts per part
- Weeks of specialist reprogramming
- Breaks when the part changes
- Line sits idle during retooling
- Foundation models, no scripts
- Taught by demonstration in days
- Adapts to product variation
- Retool in minutes, not weeks
What they actually build
The core product is a robotic work cell wrapped around a set of proprietary foundation models. Where a chatbot's model maps text to text, Industrial Next's maps sight to action: it connects visual perception, the context of the task, and the robot's motion in one loop. The company designs the seeing and the doing together, building its own smart cameras alongside the software so the two are tuned to each other rather than bolted together from off-the-shelf parts.
On top of that sits a training layer. New tasks are taught through demonstration and simulation, which is what lets a deployment happen in days. The ambition extends past a single station to what the company calls flexible - or "Matrix" - production: an autonomous mobile robotic assembly platform that can reconfigure a line on the fly.
The numbers they put on the board
Performance figures are company-reported. Treat them as claims to verify on your own line, not independent benchmarks.
Who's buying, and why it matters
The early customers are exactly where you'd expect a Tesla-alumni team to land: automakers, automotive Tier 1 suppliers, and large electronics contract manufacturers. Publicly, the company points to Compal Electronics and a production deployment at Apple's largest contract manufacturer - the kind of high-mix, high-volume environment where the cost of a rigid line shows up fastest.
That customer profile is also the strategy. Contract manufacturers change products constantly and run on thin margins, so the value of retooling in minutes is not abstract to them. It is the difference between a line that earns and a line that waits.
Where it sits in the market
Industrial Next is squeezed between two worlds. On one side are the machine-vision and industrial-robotics incumbents - Cognex, Keyence, Fanuc, ABB, KUKA - whose hardware is everywhere but whose intelligence is largely scripted. On the other is a wave of AI-robotics challengers like Covariant, Path Robotics, Bright Machines and Intrinsic, all circling the same insight that models, not code, should run the cell.
The company's wager on how to win is vertical integration: build the camera and the model together, tuned as one system. It is a harder path than selling software alone, and a slower one. But it is also the thing that is difficult to copy, and the reason a small team from Fremont thinks it can go up against giants.
The money and the milestones
Industrial Next has raised roughly $22 million, with Khosla Ventures leading and Y Combinator and Lenovo's capital group participating along the way. A $12 million Series A landed in 2022, and in April 2025 the company added another $10 million, again led by Khosla.
- 2021Founded in San FranciscoPan and Pankau start Industrial Next to democratize autonomous manufacturing.
- 2022YC W22 and $12M Series AJoins Y Combinator's winter batch and raises its first major round.
- 2023Taught by demonstrationShips cells trained by demonstration and simulation, cutting deployment to days.
- 2024Flexible "Matrix" productionExtends from smart cameras to a mobile-robot assembly platform.
- 2025$10M more from KhoslaRaises another Series A round to scale AI-defined factories.
The bet
The interesting thing about Industrial Next is how narrow its claim is. It is not promising a humanoid that walks your factory or a general robot brain. It is promising that the specific, expensive ritual of reprogramming a line can be replaced by teaching one. That is a small door, but a lot of value is stuck behind it - and the people trying to open it have already done it once, at the one factory everyone else is trying to copy.
Explore Industrial Next
- Website industrialnext.ai
- LinkedIn linkedin.com/company/industrial-next
- Y Combinator ycombinator.com/companies/industrial-next
- Careers workatastartup.com/companies/25400
- Founder Allen Pan on LinkedIn