Industrial Next's AI robots retool factory lines in minutes instead of weeks

Robots that retool in minutes

Most factory automation is locked to a single task. Point a line at a new product and you are into weeks or months of reprogramming, physical retooling or equipment swaps, which is why manufacturers tend to deploy robots only for high-volume, never-changing work. San Francisco startup Industrial Next says it has built a way out of that trap: AI-driven, self-contained robotic stations that can be reconfigured for a new task in minutes, without traditional programming.

The company made its pitch on the Startup Battlefield stage at TechCrunch Disrupt 2026, where it also announced that BMW Group Technology Office will evaluate its AI-powered dexterous manipulation technology for manufacturing, according to TechCrunch.

Founded out of frustration at Tesla

Industrial Next was founded in 2021 by Lukas Pankau and Allen Pan, both Tesla alumni. Per TechCrunch, Pankau was a lead electrical architect at Tesla before joining Waymo, while Pan led Tesla's autonomous factory efforts and later became a chief architect at autonomous driving company Neolix. The idea took shape while the two were at Tesla, where they watched advanced vehicle and self-driving work collide with factory equipment that was decades old. Pankau said engineers could use machine learning to help design cars yet were still stumped when it came to programming factory robots, and the pair eventually concluded that manufacturing itself needed to be rebuilt.

Pankau told TechCrunch that Tesla teams would "bypass robot stations because we couldn't program them effectively enough using traditional technologies" — the performance gap the company set out to close.

Simulation as the training ground

The stations run on proprietary foundation models that link vision, task context and motion. Rather than hard-coded scripts, new tasks are taught through simulation and demonstration, and the hardware is designed to be reconfigured, combined and redeployed across product lines.

The simulation piece is central. Robotics has never had the vast datasets that underpin internet-trained AI models, so eye-catching demos coexist with an inability to do precise assembly. Industrial Next trains its machines in a virtual environment engineered to mirror real-world physics closely enough that, as Pankau described it, the robot can perform high-precision, high-dexterity tasks almost without seeing the physical parts at all.

That flexibility addresses a chronic weakness of conventional industrial robots. Because they are tuned to narrow, repeatable motions, even a small change to a process or product can break them or trigger costly downtime — which, as Pankau noted, pushes manufacturers to simplify their products just so robots can handle them.

Already on a smartphone line

The technology is not purely theoretical. Pankau said some of the company's workstations are already used in producing one of the world's most popular smartphones, though he could not name the customer. Those stations reportedly handle jobs manufacturers had tried and failed to automate for years, such as threading a delicate flex cable and seating a tiny connector cap.

He also described a film-peeling station in which the robots strip protective film from camera modules and buttons at or above human speed with 100% accuracy — including after being moved to a phone model and production line they had never encountered. Those figures are the company's own claims and have not been independently verified.

The ambition reaches beyond phones. Pankau said the plan is to move on to "more complicated tasks" and, ultimately, to build "a full car end-to-end without any human touching ever."

Why it matters

Industrial automation has always traded flexibility for throughput: robots pay off when products never change, and every changeover is priced in weeks of engineering. If a line can genuinely be retargeted in minutes, that trade-off starts to dissolve — smaller batches, more frequent product revisions and mixed production runs become economically viable, and delicate assembly work that has stayed manual becomes a candidate for automation. The BMW evaluation is an early signal that a major manufacturer is taking the approach seriously. The caveats are real too: the accuracy and speed numbers come from the startup itself, the phone customer is unnamed, and dexterous general-purpose manipulation remains one of the hardest open problems in robotics. Even so, pointing foundation models and high-fidelity simulation at the factory floor marks a plausible path to the next productivity leap in manufacturing.