AMI Labs' CEO rejects the AGI label — and the superintelligence hype around it
Alexandre LeBrun, who runs Yann LeCun's world-model startup, says the industry's favorite buzzwords miss the point of what AI actually needs.
What matters
- Alexandre LeBrun, CEO of AMI Labs, publicly dismisses the terms 'AGI' and 'superintelligence' for his company's work.
- AMI Labs is building world models — AI systems designed to understand the physical world, not just predict text tokens.
- The startup raised a $1.03 billion seed round at a $3.5 billion pre-money valuation, announced March 2026.
- LeBrun's stance aligns with co-founder Yann LeCun's long-held critique of LLM-only approaches to intelligence.
- Rival World Labs (Fei-Fei Li) is reportedly raising at a $5 billion valuation, intensifying competition in the world-model space.
What happened
Alexandre LeBrun, CEO and co-founder of AMI Labs (Advanced Machine Intelligence), has publicly dismissed the terms "AGI" and "superintelligence" as labels for what his company is building, according to a TechCrunch report published July 16, 2026. The pushback is notable because AMI Labs is one of the most closely watched AI startups in the world, having launched in late 2025 with Turing Award laureate Yann LeCun as co-founder and secured a $1.03 billion seed round at a $3.5 billion pre-money valuation announced in March 2026.
AMI Labs is building "world models" — foundational AI systems designed to understand the real physical world, not merely process text or generate tokens. On its website, the startup describes its mission as building "intelligent systems that understand the real world." LeBrun's rejection of the AGI label aligns with a argument he laid out in a LinkedIn post announcing his role: that generative architectures trained by self-supervised learning do not constitute genuine intelligence, and that token prediction, while powerful, "works best for discrete and low-dimensional tasks like information retrieval, summarization, coding, and mathematics" — but falls short in environments like factories, hospitals, and robotics.
LeBrun is a serial AI entrepreneur with more than two decades of experience. He founded VirtuOz (a conversational AI company acquired by Nuance Communications in 2012–2013), co-founded Wit.ai (acquired by Facebook in 2015), and co-founded Nabla, a healthcare AI company. He also served as head of engineering at Facebook AI Research (FAIR) and was engineering lead on the Facebook M virtual assistant. He is an alumnus of École Polytechnique (class of 1994).
Why it matters
The AI industry's vocabulary has become a battleground. Companies like OpenAI, Google DeepMind, and Anthropic routinely invoke "AGI" and "superintelligence" to describe their long-term goals, framing the race as one toward human-level or beyond-human machine intelligence. LeBrun's public rejection of that framing — from the helm of a billion-dollar startup co-founded by one of the most influential AI scientists alive — signals a meaningful counter-narrative.
The disagreement is not merely semantic. It reflects a deep technical divide. LeCun has long argued that large language models, which predict the next token in a sequence, are insufficient for achieving true machine intelligence. World models, by contrast, aim to give AI systems an understanding of physical reality — spatial reasoning, causality, and the ability to plan and act in open-ended environments. LeBrun's comments suggest AMI Labs will position itself explicitly against the LLM-centric, AGI-chasing orthodoxy.
The competitive landscape is also heating up. World Labs, a rival world-model startup founded by AI pioneer Fei-Fei Li, became a unicorn shortly after emerging from stealth and launched its first product, Marble, which generates physically sound 3D worlds. World Labs is reportedly in talks to raise fresh funding at a $5 billion valuation. AMI Labs, with its $3.5 billion post-money valuation and LeCun's scientific reputation, is positioned as a leading contender in the same space — but one that is deliberately refusing to sell the superintelligence story.
What to watch
- Whether AMI Labs elaborates on its preferred terminology and technical roadmap in upcoming talks or publications.
- The startup's first product or research output — AMI Labs has not yet shipped a public product, and its timeline for doing so remains unclear.
- How investors and the broader AI community respond to the explicit rejection of AGI framing, especially as competitors like World Labs move toward commercialization.
- Whether LeCun himself publicly echoes or expands on LeBrun's stance in upcoming appearances.
What to do next
Developers
Follow AMI Labs' website and LeBrun's public posts for any API, SDK, or research release announcements, as no developer-facing product is available yet.
AMI Labs is in an early research phase with no public product, but its world-model approach could eventually offer novel tooling distinct from LLM-based APIs.
Founders
Reconsider whether 'AGI' framing helps or hurts your startup's narrative, especially if your technology involves embodied AI, robotics, or physical-world understanding.
LeBrun's deliberate rejection of AGI hype from a well-funded position suggests there is investor and talent appetite for more grounded AI narratives.
PMs
Evaluate whether world-model architectures could address use cases where token-prediction LLMs underperform — such as spatial reasoning, robotics, or physical simulation.
LeBrun explicitly argues that token prediction works best for discrete tasks and falls short in open physical environments, signaling a product gap.
Investors
Compare the world-model competitive landscape — AMI Labs vs. World Labs — and track valuation trajectories, team depth, and product timelines before making allocation decisions.
AMI Labs raised at $3.5 billion post-money while World Labs is reportedly targeting $5 billion, making this one of the most capital-intensive early-stage AI sub-sectors.
Operators
Monitor world-model progress for applications in manufacturing, healthcare, and robotics where current LLMs have demonstrated clear limitations.
LeBrun named factories, hospitals, and robots as environments where token-prediction approaches are insufficient, pointing to where world models may eventually deliver operational value.
Testing notes
Caveats
- AMI Labs has not released a public product, API, or research paper as of the reporting date. There is nothing to test yet. The startup's website describes its mission but does not offer developer access.