Yan LeCun's startup bets on AI that learns from reality

Yan LeCun's new Paris startup is betting that AI must learn from physical reality if it is ever to reason and plan in the real world. The venture, AMI Labs, stands in sharp contrast to the industry rush toward larger language models and Meta's own push into superintelligence.

LeCun is one of the most influential figures in modern AI, and his credentials give the gamble unusual weight. He shared the 2018 ACM A.M. Turing Award with Yoshua Bengio and Geoffrey Hinton for work that helped make deep neural networks central to computing. He joined Facebook in December 2013 to create FAIR, Facebook AI Research, and led it until 2018.

At FAIR's 10-year anniversary in 2023, Meta highlighted LeCun's view that AI was still a research problem. He pointed to ordinary human abilities, like clearing a table or doing dishes, that a child can learn in one shot but current AI still cannot. That skepticism is the backbone of AMI Labs, which LeCun has framed as a contrarian bet against large language models.

In January 2026, AMI Labs was described as pursuing world models, systems designed to learn from the physical world rather than simply predicting the next word. In March, LeCun said the company aims to build systems capable of reasoning and planning in complex real-world settings, and he argued that models built only to predict the next word or pixel are unlikely to produce broadly capable intelligent agents.

AMI's first customers are likely to be companies with complicated physical operations. LeCun said he is targeting manufacturers, automakers, aerospace companies, biomedical firms and pharmaceutical groups. He also said he has talked with Meta about potential deployment in Ray-Ban Meta smart glasses, tying the startup back to the hardware arena where Meta has been pressing its AI ambitions.

Meta has said it sold millions of Ray-Ban Meta glasses in 2026 and has cast AI glasses as its defining new hardware category. The company also says its EMG Neural Band can translate muscle signals into device actions, while its Orion glasses program is meant to show what the next generation of AI and augmented reality computing could look like.

That corporate backdrop makes LeCun's move more than a personal side project. Meta reorganized its AI work under Superintelligence Labs in 2025, led by Alexandr Wang, as the company accelerated toward systems it says can surpass today's models. LeCun's startup is pushing in the opposite direction, betting that the next decade of AI will belong to systems that can anchor themselves in the world before they try to conquer it.