Founder & CEO / Proception
Jay Li
He is teaching machines to touch. Twenty-two joints, a skin that feels, and a workshop full of fingers in Palo Alto.
Jay Li. Ten years of hardware, one obsession: the human hand.
He is teaching machines to touch. Twenty-two joints, a skin that feels, and a workshop full of fingers in Palo Alto.
Jay Li. Ten years of hardware, one obsession: the human hand.
Jay Li runs Proception out of Palo Alto, and the company sells one thing: a robotic hand good enough to make humanoid robots useful. Not a gripper. Not a claw. A hand - twenty-two degrees of freedom, multiple joints per finger, cables that pull like tendons, and a skin that registers contact the way your fingertips do. In June 2026 the first batch left the building, headed for researchers and robotics companies wrestling with the same problem Li has been circling for years: getting machines to manipulate the messy, contact-rich real world.
The product is called ProHand. Its stated north star is threading a needle - the sort of task that sounds trivial until you try to make a motor do it. Fine motor control, tactile feedback, an adaptive grip that adjusts a hundred times a second. Li's bet is that you cannot bolt intelligence onto bad hardware. You build hardware close enough to a human hand that a robot can learn from human hands directly.
That bet has a companion product, and it is the clever part. Before Proception built a robot, it built a glove.
You need both hardware and data, and those need to come hand-in-hand to get dextrous manipulation to work.- Jay Li, on why Proception builds a glove and a hand together
ProGlove takes the same skin-like sensors that cover ProHand and puts them on a person. Wear it, reach into a drawer, tie a knot, fold a shirt - and the glove records how a human hand actually solves the problem, no robot required. It is a data-collection machine disguised as clothing. The insight is that the scarce resource in robotics is not compute, it is contact: millions of examples of hands touching things. Capture that from people, and you skip the slow, expensive dance of teaching a robot from scratch.
Twenty-two degrees of freedom, multiple joints per finger. Tendon-driven: motors pull cables so the hand stays light and compact. Built impact-tolerant on purpose.
Integrated skin-like sensors detect contact and support grip control during manipulation - the difference between holding an egg and crushing it.
A wearable that turns the sensor skin into a capture rig. A person interacts with real objects; the glove logs human manipulation with no robot in the loop.
Roughly how ProHand's 22 degrees of freedom stack against the shapes robots usually settle for.
*Human hand DoF is an approximate reference point, not a Proception spec.
It was a resilience test, or pressure test.- Jay Li, on a year spent building while being sued
The lawsuit was not a footnote. Tesla filed in June 2025, alleging Li had taken confidential files about robotic hand actuation and claiming Proception's hands bore "striking similarities" to its internal designs. Li had resigned and started his company six days later, which made for an easy narrative and a hard year.
It resolved in June 2026. The suit was dismissed after a settlement; Tesla did not comment. On roughly the same day, Proception announced its $11M seed round and began shipping. A year of litigation, closed with a raise and a product out the door. Li's read on the ordeal is characteristically dry - a pressure test the company passed.
I would not be surprised if Tesla eventually comes to Proception for help with its own hand problem. I think it will happen.- Jay Li
Humanoid robots have gotten good at walking, balancing, and looking impressive on stage. What they still cannot reliably do is the ordinary work of hands: buttoning, threading, sorting, the ten thousand small manipulations a person does without thinking. That gap is the whole game, and it is where Jay Li has planted his flag.
His method is unfashionably physical. Where much of AI chases bigger models, Proception chases better contact - hardware that mimics the human hand and gloves that harvest human demonstrations at scale. Hardware and data, hand-in-hand, as Li puts it. Whether that becomes the standard interface for every humanoid or a specialist tool for the labs pushing manipulation forward, the first hands are already on benches, and the founder who built them thinks the giants may come knocking.