Cognition AI, Inc.
Developer of the autonomous AI software engineer “Devin” — valuation reaches $48B following its September 2026 Series E round
Cognition AI, Inc. (“Cognition”) is an applied AI lab founded in San Francisco in 2023 that develops Devin, an autonomous AI software engineer, as its flagship product. In our assessment, Cognition’s founding-team composition — drawn heavily from competitive-programming and elite AI-research circles — is a standout case of technical credibility underwriting early product velocity.
Sources disagree on Cognition’s official founding month. Wikipedia cites August 2023, while Crunchbase-derived databases and the majority of press coverage (Frederick AI, Clay, and others) cite November 2023. We adopt November 2023, the date supported by the greater weight of sources, while flagging that the company has not issued a formal, verifiable incorporation date and further confirmation would be required.
Raised in Baton Rouge, Louisiana. Wu won three consecutive International Olympiad in Informatics (IOI) gold medals from 2012 to 2014, placing first with a perfect score in his final competition. He attended Harvard and represented the university at the International Collegiate Programming Contest (ICPC) before leaving to build companies. He worked in performance engineering at Addepar and, in 2017, co-founded the AI-powered professional-networking platform Lunchclub as CTO alongside Vlad Novakovski. In March 2024, Wu unveiled Devin as “the world’s first AI software engineer,” drawing significant industry attention.
Serves as Cognition’s co-founder and Chief Technology Officer, understood to oversee Devin’s core agent architecture and model-training pipeline (per company disclosure). Multiple sources identify Hao as a top-tier competitive-programming alumnus, consistent with the founding team’s broader profile. Detailed biographical and career information is only partially disclosed through official channels, and we flag this as an area warranting further verification.
Co-founder and Chief Product Officer, understood to lead product strategy and roadmap across both Devin and the acquired Windsurf asset. Yan, along with Wu and Hao, assembled the founding team through overlapping competitive-programming and AI-research networks.
Per company statements and media interviews, Cognition’s engineering organization includes multiple IOI gold medalists and alumni of DeepMind, Waymo, and other leading AI research institutions. In a 2025 Lenny’s Newsletter interview, CEO Scott Wu stated that a roughly 15-person engineering team works with about five Devin instances per engineer, and that a substantial share of the company’s own pull requests are generated directly by Devin (self-reported; not independently verified).
In our assessment, the founding team’s competitive-programming and top-tier AI-research pedigree correlates directly with Devin’s technical trajectory — from a 13.9% SWE-bench resolution rate at its March 2024 launch to a reported ~90% by 2026 — and this is one of the clearer evidentiary threads supporting the company’s benchmark leadership relative to peers. We note that these benchmark figures rest on company disclosure and industry commentary, and we have not identified independent third-party audit confirmation.
Cognition’s core product is Devin, an autonomous AI software engineer that takes natural-language requirements and self-sufficiently performs codebase analysis, task planning, and code writing, testing, debugging, and deployment within a sandboxed environment. In July 2025, Cognition acquired the remaining assets of the IDE-based AI coding tool Windsurf (formerly Codeium), giving it a combined product portfolio spanning both agent-first (Devin) and IDE-first (Windsurf) workflows.
Cognition’s revenue model is built around three principal tracks.
Usage-based and seat-based subscriptions aimed at individual developers and small teams. Devin integrates into existing collaboration tools such as Slack, Linear, and GitHub, functioning as a remote engineer. Per company disclosure, ARR grew from $1M in September 2024 to $73M by June 2025, driven primarily by bottom-up developer adoption.
At the time of its July 2025 acquisition, Windsurf reportedly carried run-rate revenue of approximately $82M, more than 350 enterprise customers, and hundreds of thousands of daily active users (self-reported). Windsurf’s top-down enterprise sales motion reaches a customer segment distinct from Devin’s.
Named customers reportedly include Citi, Goldman Sachs, Mercedes-Benz, Dell, Santander, Elevance Health, the U.S. Army, and the U.S. Navy (per company disclosure in the May 2026 Series D announcement). Large-organization, multi-seat deployments underpin a recurring-contract structure.
Model-Agnostic Positioning: Cognition positions itself as an “independent agent lab” that selects and combines its own trained models (e.g., SWE-1.6) with third-party models depending on the task, rather than tying its product to a single foundation-model provider. This differentiates it from model-tethered competitors such as GitHub Copilot (Microsoft/OpenAI) or Claude Code (Anthropic). In our view, this structure offers some buffer against pricing or policy shifts from any single model vendor, though it does not eliminate compute-cost exposure.
Capital Efficiency: In a September 2025 blog post, Cognition stated that cumulative net burn since founding remained under $20M (self-reported; not independently verified). The revenue multiple implied by the September 2026 Series E — roughly 53x ARR ($900M against a $48B valuation) — is broadly consistent with the multiple implied at the prior round.
In roughly two and a half years since its first external round in March 2024, Cognition has scaled from a $21M seed-stage raise to a $48B valuation. We view the roughly 85% valuation step-up between the May 2026 Series D ($26B) and the September 2026 Series E ($48B) — occurring in under four months — as a notable illustration of the intensity of venture capital flowing into the AI coding-agent category.
Series labeling for the 2024–2025 rounds is inconsistent across sources. The April 2024 $175M raise is described as a “Series A extension” in some sources (Sacra) and as a “Series B” in others (Tracxn, CBInsights). The March 2025 round to a $4B valuation is similarly labeled “Series B” in some sources and “Series C” in others, and its exact raise amount is undisclosed. We have organized the timeline below by date, amount, and valuation rather than by disputed series labels.
Founders Fund led the company’s first institutional round. That same month, Cognition unveiled Devin as “the world’s first AI software engineer,” posting a 13.9% SWE-bench resolution rate and drawing broad market attention to the AI coding-agent category.
Founders Fund led a substantial follow-on raise, pushing Cognition to unicorn status ($1B+ valuation) roughly six months after founding. 8VC, Khosla Ventures, and Pear VC participated as early investors.
8VC led a follow-on round that lifted Cognition’s valuation to $4B. The exact capital raised in this round is not confirmed in public filings or disclosures.
Background: In May 2025, OpenAI agreed to acquire Windsurf (formerly Codeium) for approximately $3B. That deal collapsed on July 11, 2025, reportedly over concerns tied to Microsoft’s access to Windsurf’s intellectual property. The same day, Google DeepMind struck a roughly $2.4B non-exclusive technology licensing and talent deal — a so-called “reverse acquihire” — bringing on Windsurf CEO Varun Mohan, co-founder Douglas Chen, and other core research staff.
Cognition’s Acquisition: Three days later, on July 14, 2025, Cognition announced a definitive agreement to acquire Windsurf’s remaining intellectual property, product, brand, and roughly 210 remaining employees. At the time of acquisition, Windsurf reportedly carried run-rate revenue of approximately $82M, more than 350 enterprise customers, and hundreds of thousands of daily active users. Cognition committed to equity participation for all remaining employees and accelerated vesting.
Founders Fund led the round, with existing backers Lux Capital and 8VC re-upping and new investors Bain Capital Ventures and D1 Capital joining. The company stated that combined ARR more than doubled following the Windsurf integration.
Led by Lux Capital, General Catalyst, and 8VC. The company disclosed run-rate revenue of $492M at the time of the round and reported more than tenfold growth in enterprise usage since the start of 2026.
New investors Andreessen Horowitz and Accel co-led the round, joined by existing backers Founders Fund, General Catalyst, and Avenir. A broad syndicate — including Benchmark, Bessemer, Kleiner Perkins, Greylock, Lightspeed, T. Rowe Price, and NVIDIA — also participated. The valuation rose roughly 85% versus the Series D in under four months.
• Capital Raised: $2,000,000,000+
• Post-Money Valuation: $48,000,000,000 (up ~85% from the $26B Series D)
• ARR Growth: $492M (May 2026) → ~$900M (Sep 2026, per company), an ~83% increase
• Implied Multiple: ~53x ARR — broadly consistent with the prior round (per company disclosure)
• Company Guidance: ARR exceeding $1.5B by year-end 2026, $4–5B by 2027 (internal company forecast; unconfirmed)
In our assessment, Cognition’s competitive position rests on (1) a founding team drawn from an unusually deep pool of elite technical talent, (2) a combined agent-plus-IDE product portfolio following the Windsurf acquisition, (3) a model-agnostic architecture, and (4) an expanding set of large enterprise and government reference customers. We note, however, that the AI coding-agent market remains crowded with well-capitalized competitors — Cursor (Anysphere), GitHub Copilot (Microsoft), Claude Code (Anthropic), and OpenAI Codex among them — and the category is shaping up as a multi-winner market rather than a winner-take-all one.
A high concentration of International Olympiad in Informatics (IOI) medalists and DeepMind/Waymo alumni across the founding team and engineering organization is credited with lifting Devin’s benchmark performance from 13.9% SWE-bench at its March 2024 launch to a reported ~90% (SWE-bench) and ~80% (SWE-bench Pro) by 2026 (per company and industry sources; not independently audited). We view this as a technical basis for the company’s first-mover advantage in the category.
While most competitors concentrate on either the IDE category (Cursor) or the agent category (Claude Code), Cognition’s Windsurf acquisition gives it a presence in both. Offering a developer-controlled “IDE workflow” alongside a fully delegated “agent workflow” simultaneously is, in our view, structurally supportive of deeper customer lock-in across segments.
Running proprietary and third-party foundation models in parallel gives Cognition leverage against pricing or API policy shifts from any single model vendor, relative to competitors tethered to one provider. This flexibility, however, comes with an ongoing and substantial compute-investment burden, which warrants continued monitoring of the company’s margin profile.
The company states it has secured customers including Citi, Goldman Sachs, Mercedes-Benz, Dell, Santander, and the U.S. Army and Navy (self-reported; not independently verified). Such references can support follow-on enterprise sales credibility, though regulated and public-sector customer segments typically carry longer sales cycles — a factor worth weighing against the headline logo list.
Market Interpretation: Coverage of the September 2026 Series E — including from TechCrunch — has framed the round as a signal that the AI coding market is “far from winner-take-all,” a reading reinforced by continued large venture inflows to competitors such as Cursor (Anysphere) around the same period. We regard this as simultaneously a positive signal for category growth and a source of uncertainty around the durability of current valuation premiums.
- Implied valuation multiple of roughly 53x ARR ($900M) sits at the premium end even within the broader AI sector
- Core metrics — revenue, growth rate, net burn — rest almost entirely on voluntary company disclosure; no independent audit or SEC filings exist as a private company
- Crowded, well-capitalized competitive set (Cursor, GitHub Copilot, Claude Code, OpenAI Codex) creates ongoing pricing pressure and customer-churn risk
- Two large rounds closed within roughly four months of each other may itself signal elevated compute-cost pressure or a degree of market overheating
- The 2027 ARR target of $4–5B is an internal company forecast with no external verification mechanism
- ARR growth from $492M to $900M in roughly four months — an ~83% increase, among the fastest-growing profiles in the category
- Participation from strategic and crossover investors such as NVIDIA and T. Rowe Price is consistent with pre-positioning ahead of a potential IPO
- Windsurf integration gives the company a dual growth engine spanning bottom-up developer adoption and top-down enterprise sales
- Reference customers in regulated industries and the public sector (Citi, Goldman Sachs, the U.S. Army and Navy) could serve as a beachhead for further large-scale contracts
- A model-agnostic architecture provides flexibility to optimize the cost structure as open-source and lower-cost models proliferate