The archive · AI & Models · Technical decision · 2025–2026
Ndea: Chollet and Knoop bet program synthesis can build AGI, enter YC W26 with $43M
François Chollet and Mike Knoop left Big Tech to bet deep-learning-guided program synthesis can build AGI; Ndea entered YC W26 with a $43M pre-YC raise
Ndea
What the business is
An AI research and science lab (YC W26) aiming to 'develop and operationalize' AGI by fusing deep learning with program synthesis, then turning the resulting AI into a factory for rapid scientific invention and commercialization.
Starting capital：$43M raised before entering YC W26 (largest pre-YC raise in the batch, per Unicorn Screener); no round-by-round breakdown disclosed
How it started
François Chollet, creator of Keras, announced in November 2024 he was leaving Google after nearly a decade; on 2025-01-15 he and Mike Knoop, Zapier co-founder and head of AI, publicly launched Ndea as an AI research and science lab focused on developing and operationalizing AGI, with Knoop stepping back from day-to-day Zapier while staying on its board.
What happened
Ndea's thesis: current deep learning is constrained by its inability to efficiently learn and adapt, so the lab bets on deep-learning-guided program synthesis — deep learning guides search over discrete programs, needing only a few examples to generalize. Chollet called the chance of a breakthrough 'small but real.' By 2026 the lab had joined YC's W26 batch with a $43M pre-YC raise — the largest in the batch — and was hiring remote research positions.
How it ended up
In progress: as of YC W26 Demo Day (2026-03-24) Ndea is still in the research phase — no product or benchmark results disclosed — building what it calls the world's most talent-dense program synthesis team.
Background
Ndea is an AI research and science lab co-founded by François Chollet — creator of Keras, former Google researcher — and Mike Knoop, Zapier co-founder, launched publicly on 2025-01-15 with the stated goal of developing and operationalizing AGI.
The bet is technical: Ndea argues deep learning is fundamentally constrained by its inability to efficiently learn and adapt, and that the path forward is program synthesis — searching discrete programs that explain data, guided by deep learning to overcome the computational cost. Chollet said the team has 'a small but real chance' of creating AI that learns as efficiently as people.
By YC's W26 batch — 196 companies with Demo Day on 2026-03-24 — Ndea had already raised $43M before entering, the largest pre-YC raise in the batch, per Unicorn Screener. The pair are also co-founders of the ARC Prize Foundation, the nonprofit behind the ARC-AGI benchmark Chollet designed.
As of Demo Day the lab is still in the research phase, hiring remote researchers to assemble 'the world's most talent-dense program synthesis team,' with the ambition of building a 'factory for rapid scientific advancement' that invents and commercializes new ideas.
What has to be true
- Chollet designed the ARC-AGI benchmark the field uses to measure progress toward general intelligence, giving the thesis instant credibility.
- The claim is falsifiable and specific — deep learning alone can't adapt; guided program synthesis is the missing piece — which made it fundable as a research bet.
- Knoop brought a track record of building Zapier and a network that let two founders raise $43M before Demo Day.
- Framing the lab as a 'factory for scientific advancement' set a goal beyond model performance, appealing to investors who wanted a different path from the scaling-law consensus.
- Going through YC despite already being funded bought access to the batch's recruiting and deal-flow gravity while keeping the research-first posture.
What can be applied
A research bet scales on credibility plus a falsifiable thesis: Chollet and Knoop raised $43M on a named problem and a named method — guided program synthesis — before shipping anything.
Aftermath
As of YC W26 Demo Day (2026-03-24), Ndea is in the building phase: no product or benchmark results have been disclosed; the lab is hiring remote researchers and describing itself as building a factory for rapid scientific advancement via deep-learning-guided program synthesis. Its $43M pre-YC raise — the largest in the W26 batch — funds a research runway that most seed-stage AGI bets don't have.
Sources
- AI researcher François Chollet founds a new AI lab focused on AGI
- 5 YC Startup Batch 2026 Breakouts Worth Watching Now
spotted an error? The archive wants to know.
Your turn
You just read one. Describe what you are building, and see who is betting on the same thing.
Free account · 3 free questions · no card