For anyone tracking alternatives to today's scale-everything AI race, Ndea is worth watching. It is a fully remote team of roughly fifteen researchers, already well funded going into Y Combinator's Winter 2026 program, betting that program synthesis, not larger language models, is the more promising route to real general intelligence and faster scientific discovery.
Ndea is a San Francisco AI research lab betting that artificial general intelligence needs deep-learning-guided program synthesis, not larger language models. It was started by Francois Chollet, the researcher behind Keras and the ARC-AGI benchmark, together with Mike Knoop, formerly Zapier's president, both of whom also lead the nonprofit ARC Prize Foundation.
Founded: 2025 · HQ: San Francisco, CA, USA (fully remote team) · Team: 10-20 · CEO: Mike Knoop · Funding: $43.5M raised pre-YC (investors: Y Combinator, Coatue Management, Factorial Capital, Quiet Capital)
About Ndea
Ndea was founded in January 2025 in San Francisco by Francois Chollet, creator of the Keras deep learning framework and the ARC-AGI benchmark, and Mike Knoop, co-founder and former president of Zapier. Chollet left Google after roughly a decade on the Keras and TensorFlow teams to start the lab; Knoop stepped back from day-to-day Zapier duties to serve as Ndea's CEO. The name blends the Greek ennoia (intuitive understanding) and dianoia (logical reasoning), reflecting the founding bet that deep learning and program synthesis need to be fused, not chosen between. Ndea is structured as a fully remote company, with researchers and engineers distributed across Europe and North America rather than concentrated in one office. The lab's technical thesis, which Chollet calls symbolic learning, searches for compact symbolic programs that explain observed data rather than interpolating within a fixed neural embedding space, using deep learning to guide that search so it stays computationally tractable. The stated goal is AI that can pose new problems and adapt to genuinely unfamiliar ones, not just execute patterns it has already seen in training data, an ability Ndea frames as the missing ingredient for compressing decades of scientific progress into a few years. Chollet's ARC-AGI benchmark, the reasoning test he built before founding Ndea and continues to develop through the nonprofit ARC Prize Foundation, is the lab's primary yardstick for that progress; a third version, ARC-AGI-3, is due out in 2026. Ndea joined Y Combinator's Winter 2026 cohort with $43.5 million already raised, reportedly the largest pre-YC check of that batch. Disclosed backers include Y Combinator, Coatue Management, Factorial Capital, and Quiet Capital; the round structure and any valuation have not been made public. As of mid-2026 the team numbers around fifteen people, and the lab has not shipped a commercial product, an API, or a public model checkpoint; its output so far is research direction, hiring, and a podcast and YouTube channel where Chollet and Knoop discuss the work. Both founders are also co-founders and board members of the ARC Prize Foundation, the separate nonprofit that runs the ARC-AGI competition and publishes its leaderboard. Ndea's positioning is explicitly a bet against the rest of the frontier field: where OpenAI, Anthropic, and Google DeepMind continue to scale transformer-based language models, Ndea argues that path is running into diminishing returns and is betting on program synthesis instead. DeepMind is the closest overlap on method, through neurosymbolic systems like AlphaGeometry and AlphaProof, but Ndea is alone among the group in having no shipped product and no revenue, trading commercial traction for a research thesis that, if it works, its founders argue would be a bigger leap than another round of scale.
Mission
Building AGI that can innovate, to accelerate scientific progress for current and future generations.
Products
- Ndea Podcast (Research podcast): https://open.spotify.com/show/67zmiUKRMLQApaMnxpsHKY
Links
Frequently Asked Questions
How much funding has Ndea raised?
Ndea raised about $43.5 million before it even joined Y Combinator's Winter 2026 batch, reportedly the batch's largest pre-program raise. Reported backers are Y Combinator alongside Coatue Management, Factorial Capital, and Quiet Capital. The company has not disclosed a valuation or broken out individual round sizes.
What products does Ndea make in 2026?
As of 2026, Ndea has not shipped a commercial product, developer API, or public model checkpoint. Its public output is research direction and hiring, plus a podcast and YouTube channel where founders Francois Chollet and Mike Knoop discuss the lab's program-synthesis approach to AGI. The lab's benchmark of record, ARC-AGI, is published separately through the nonprofit ARC Prize Foundation.
How does Ndea handle data privacy and compliance?
Ndea has not published a privacy policy, terms of service, or any compliance certification, consistent with a research lab that has no live product or customer data to govern yet. There is no public SOC 2, ISO 27001, or similar attestation as of 2026. This is worth revisiting once the lab ships a product.
Who are Ndea's main competitors?
Ndea's closest peers are the frontier labs it explicitly argues against: OpenAI and Anthropic keep scaling transformer-based language models, and Ndea bets that path is hitting diminishing returns. Google DeepMind is the nearest methodological overlap, through neurosymbolic work like AlphaGeometry and AlphaProof, though DeepMind ships products and Ndea currently does not. Ndea's edge, if the thesis holds, is a fundamentally different route to general intelligence rather than a faster version of the same one.
How do you start using Ndea's products?
There is no consumer app or developer API to sign up for yet. The practical entry points are the Ndea podcast and YouTube channel, the ARC-AGI benchmark maintained through the ARC Prize Foundation, and the lab's careers page, which lists openings for researchers and engineers, particularly in program synthesis.