Mike Knoop
Last reviewed
Sources
No citations yet
Review status
Needs citations
Revision
v4 · 2,657 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
Sources
No citations yet
Review status
Needs citations
Revision
v4 · 2,657 words
Add missing citations, update stale details, or suggest a clearer explanation.
Mike Knoop is an American entrepreneur and artificial intelligence researcher best known as a co-founder of the workflow-automation company Zapier (founded 2011) and, since 2024, as the co-founder of the ARC Prize Foundation, a non-profit competition and benchmark organization that he established together with French AI researcher François Chollet to accelerate progress toward agi.[1][2][3] In January 2025 the two also launched Ndea, an AI research lab focused on deep-learning-guided program synthesis as a path toward general intelligence.[4][5]
Knoop spent the first decade of his career building product and engineering at Zapier, where his roles included head of product and, later, general manager of Zapier AI. He stepped back from his day-to-day operating role in 2022 to work full-time as an individual contributor on artificial-intelligence research within Zapier after being "shaken" by the chain-of-thought reasoning literature.[6][7] He remains a Zapier co-founder and board member while leading the ARC Prize Foundation and Ndea.[1][2]
Knoop is a public advocate for Chollet's "skill-acquisition efficiency" definition of intelligence and has argued that scaling current frontier large language models alone will not be sufficient to reach artificial general intelligence, a thesis that underpins the ARC-AGI benchmark series and the ARC Prize competition.[6][8][9]
| Field | Details |
|---|---|
| Born | United States (raised in Wildwood, Missouri, west of St. Louis)[10] |
| Education | B.S. Mechanical Engineering, University of Missouri[11][12] |
| Known for | Co-founder of Zapier (2011); co-founder of the ARC Prize Foundation (2024); co-founder of Ndea (2025)[1][2][4] |
| Current roles | Co-founder & board member, Zapier; co-founder & board member, ARC Prize Foundation; co-founder & CEO, Ndea[1][2][4] |
| Residence | San Francisco Bay Area[1] |
| Twitter / X | @mikeknoop[13] |
Mike Knoop grew up in Wildwood, Missouri, a suburban community west of St. Louis. He has described himself as a quiet child who, while playing sports, spent much of his free time exploring his family's computer.[10] In middle school he taught himself HTML, CSS and PHP while customizing his AOL Instant Messenger profile, and around the same time learned Z80 assembly to write games and utilities for the TI-83+ graphing calculator distributed through his school's math program.[10] He has also written that he took up performing stage magic during this period as a way to push himself to be more outgoing socially.[10]
Knoop attended the University of Missouri in Columbia (commonly known as "Mizzou"), where he earned a Bachelor of Science in Mechanical Engineering. His undergraduate research focused on finite-element modeling and optimization.[11][12]
Zapier began as a side project in Columbia, Missouri in 2011. Co-founders Wade Foster and Bryan Helmig, colleagues at the lender Veterans United Home Loans, began work on the tool in September 2011, and Knoop joined them later that year.[11][14] All three founders were University of Missouri alumni.[11][14]
The product, a service for connecting otherwise-disconnected web applications through user-defined "Zaps", launched in public beta in 2012. Zapier was accepted into the Y Combinator Summer 2012 (S12) batch, and in October 2012 raised a $1.2 million seed round led by Bessemer Venture Partners with participation from Draper Fisher Jurvetson and angel investors.[14][15] The company was profitable by 2014 and operated as a remote-first, largely bootstrapped business for most of the following decade.[14]
At Zapier, Knoop ran all of the company's product engineering until mid-2022. He led the development and launch of multiple core Zapier products and lines of business, including the original Zaps automation product, Transfer, Tables, Interfaces, AI Actions, and Central, Zapier's AI agent / "AI bots" product. He also led the company's partnership with OpenAI, including the ChatGPT plugin and GPT Store launches.[6][7]
In a January 2021 secondary-market transaction, Sequoia Capital and Steadfast Financial purchased shares from existing Zapier shareholders at a $5 billion valuation, making Zapier one of the most valuable bootstrapped companies in Silicon Valley.[14][16] The company has remained privately held; by 2024 it reported approximately $310 million in annual revenue, growing toward roughly $400 million in 2025 according to publicly reported figures.[17]
Knoop has said that the January 2022 chain-of-thought reasoning paper from Google researchers, one of the foundational works behind modern chain of thought prompting, was a turning point in his thinking about AI. He described it as having "shaken" him because it suggested that language models could be coaxed into multi-step reasoning behavior in ways that he had not previously expected, and prompted him to revisit the question of whether existing systems were on a credible path toward general intelligence.[6][7][8]
In the summer of 2022, Knoop and Helmig both stepped down from their executive roles to become individual contributors focused on AI research and development inside Zapier. Knoop has framed this transition as a personal effort to understand "are we on the path for AGI or not?", taking the unusual step for a co-founder of a multi-billion-dollar company of returning to a hands-on technical role.[7][8]
While inside Zapier, Knoop wrote and spoke publicly about how large language models could be integrated into automation products, and Zapier's AI Actions and Central products were central case studies for early "AI agent" deployments in workflow automation. Knoop also led the company's partnership with OpenAI through both the ChatGPT plugin program and the launch of the GPT Store, positioning Zapier as one of the most widely-used third-party action providers across emerging AI assistants.[6][7] He has continued to serve as a Zapier co-founder and board member as his time has shifted toward the ARC Prize Foundation and Ndea.[1][2]
The ARC Prize Foundation was established in 2024 as a non-profit organization with the stated mission of guiding researchers, industry, and regulators toward artificial general intelligence through enduring benchmarks. Knoop co-founded the foundation with François Chollet, the creator of the ARC-AGI benchmark (originally introduced in Chollet's 2019 paper On the Measure of Intelligence) and of the open-source deep-learning library Keras. Greg Kamradt serves as president of the foundation; Knoop and Chollet sit on the board.[2][3][18]
ARC Prize 2024, the first competition run under the new foundation, was announced jointly by Knoop and Chollet on June 11, 2024, the same day they appeared together on Dwarkesh Patel's podcast to discuss the launch.[8][19] The competition was run on Kaggle and on arcprize.org, with a total prize pool of approximately $1 million and a $600,000 Grand Prize reserved for any team that could reach 85% on the private evaluation set of ARC-AGI-1 within the contest's tight compute budget.[19][20]
The contest closed on November 10, 2024, with 1,430 teams submitting 17,789 entries. No team reached the 85% Grand Prize threshold. The first-place "high-score" winners were "the ARChitects," who scored 53.5% on the private evaluation set; the team known as MindsAI achieved the overall top score of 55.5% but was ineligible for prize money because it did not open-source its solution. During the 2024 contest the state of the art on ARC-AGI-1 rose from roughly 33% to 55.5%.[19][20] In commentary published with the technical report, Knoop stated that he believed "AGI progress is no longer stalled" but that "we still need new ideas".[19]
A separate, widely-discussed result followed shortly after the contest. On December 20, 2024, the ARC Prize Foundation published verified results for OpenAI o3, an OpenAI reasoning model still in pre-release. On the ARC-AGI Semi-Private evaluation set, a "high-efficiency" o3 configuration scored 75.7% at a reported cost of roughly $2,680 per evaluation, while a "low-efficiency" configuration using roughly 172× more compute scored 87.5% at approximately $456,000 in inference cost. On the Public evaluation set the equivalent figures were 82.8% and 91.5%. The foundation noted that the o3 system tested had been trained on 75% of the ARC-AGI public training set and that the low-compute 87.5% score, while exceeding the 85% Grand Prize threshold on a different evaluation set, did not satisfy the formal ARC Prize Grand Prize rules because of its compute budget.[21] Chollet, writing on the foundation's blog, characterized o3's results as "a surprising and important step-function increase in AI capabilities".[21]
A second-generation benchmark, ARC-AGI-2, was released by the foundation on March 24, 2025, accompanied by the launch of ARC Prize 2025 on Kaggle. The accompanying technical paper was authored by Chollet, Knoop, Greg Kamradt, Bryan Landers, and Henry Pinkard.[22][23] ARC-AGI-2 preserves the visual, grid-based task format of the original benchmark while introducing tasks designed to be substantially harder for AI systems while remaining tractable for human solvers, and it normalizes the distribution of task difficulty.[22][23]
ARC Prize 2025 attracted 1,455 teams and 15,154 entries; the top score on the ARC-AGI-2 private evaluation set was approximately 24%, well short of an 85% Grand Prize threshold. The foundation distributed more than $725,000 in prizes over the course of the 2025 competition.[18][23]
In July 2025 the foundation released the first public preview of ARC-AGI-3, posting an initial set of three interactive game-style environments. Announcing the preview, Knoop wrote that "version 3 is a big upgrade over v1 and v2 which are designed to challenge pure deep learning and static reasoning. In contrast, v3 challenges interactive reasoning (eg. agents)."[24] The full ARC-AGI-3 benchmark, comprising hundreds of original turn-based environments with thousands of game-style levels and no pre-stated rules or goals, was launched publicly on March 25, 2026 at an event held at Y Combinator's San Francisco headquarters, featuring a fireside conversation between Chollet and OpenAI chief executive Sam Altman on measuring intelligence on the path to AGI.[25]
At launch, ARC-AGI-3 was reported to be solved at 100% by human participants and at approximately 0.5% by frontier AI systems, a gap the foundation describes as evidence that the benchmark remains unsaturated for general agentic intelligence.[25] ARC Prize 2026, run alongside the ARC-AGI-3 launch, offered more than $2 million in total prizes split across an ARC-AGI-3 competition focused on agent-based solutions and a remaining ARC-AGI-2 Grand Prize for the best open-source solution.[25]
On January 14, 2025, Knoop and Chollet jointly announced Ndea (stylized as Ndea), a new private AI research lab focused on what the founders describe as "deep-learning-guided program synthesis." On X, Chollet wrote that he was "joining forces with @mikeknoop to start Ndea (@ndeainc), a new AI lab" and that the company's bet was "a different path to build AI capable of true invention, adaptation, and innovation."[4][26]
The lab's stated mission is "to operationalize AGI to realize unprecedented scientific progress for the benefit of all current and future generations". Ndea's name is derived from the Greek concepts of ennoia (intuitive understanding) and dianoia (logical reasoning), reflecting the founders' goal of combining pattern recognition through deep learning with explicit symbolic reasoning through program synthesis.[4][5]
Knoop is Ndea's co-founder and chief executive; Chollet is co-founder and chief scientist. The lab has publicly described itself as recruiting a globally distributed team of researchers and engineers and as positioning itself as a long-horizon research effort rather than a near-term product company.[4][5][26]
Knoop is one of the more public advocates of François Chollet's framing of intelligence as "skill-acquisition efficiency," the idea that a system's intelligence should be measured by how efficiently it can master genuinely novel tasks from limited experience, rather than by raw performance on any single, well-specified task. He has argued repeatedly that benchmarks which saturate to human-level performance tell us comparatively little about progress toward agi, whereas the ARC-AGI series is designed to remain difficult for AI systems whose pattern-matching capabilities continue to grow.[6][8][9]
He has been critical of what he sees as growing secrecy among leading AI labs and an emphasis on closed-source frontier research. Knoop has cited the open scientific lineage behind innovations such as the Transformer architecture as evidence that AGI-relevant breakthroughs historically depend on open collaboration; one of the core conditions of the ARC Prize is that winning solutions must be open-sourced and reproducible.[6][8] Knoop has also defended the ARC Prize's deliberately tight compute and no-internet constraints, arguing that an honest measure of intelligence must include some notion of efficiency rather than an unbounded brute-force budget.[6]
Knoop has spoken at length on his views in long-form interviews on the Dwarkesh Podcast (joint appearance with Chollet, June 2024), the Sequoia "Training Data" podcast, the Cognitive Revolution, Arize AI's podcast, and others, in addition to writing essays on his personal site mikeknoop.com on topics including the implications of OpenAI o1 and DeepSeek-R1, AI policy, and product strategy.[1][6][8][9][27] He is also an angel investor in a number of AI and developer-tools companies, including Cognition, Limitless, Julius, Cartesia, and Braintrust.[1]
In interviews Knoop has emphasized that he does not view AGI as arriving as a sudden, discontinuous "superintelligence" event. Instead, he describes his expectation as a gradual staircase of capabilities, in which each new generation of systems extends what was previously possible and allows society to assess effects empirically before more capable successors arrive.[6] He has argued that this framing has policy implications: it suggests that regulation can be incremental and grounded in observed behavior of deployed systems rather than in speculation about hypothetical future capabilities, and that it allows researchers, regulators, and the public to update their views together as benchmarks like ARC-AGI continue to be tested.[6][8]
He has also been an outspoken proponent of the view, shared with Chollet, that progress on ARC-AGI is more informative than progress on language- or math-benchmark leaderboards because ARC-AGI tasks resist memorization. In Knoop's framing, this is what makes the benchmark a useful target for the open-source research community and a meaningful constraint on claims about AGI.[6][8][9]