Cognition AI closes $400M at $10.2B valuation — Scott Wu on Devin's 8-15x engineering speedups and the Windsurf integration

Summary

Cognition AI has closed a $400 million round at a $10.2 billion valuation, led by Founders Fund. Scott Wu, Cognition's CEO, says the timing was directly tied to the Windsurf acquisition: Founders Fund signaled support for the deal in real time, and the raise closed within roughly 72 hours of that conversation.

Devin's commercial traction

Wu describes Devin's core enterprise use case as absorbing engineering toil — migrations, replatforms, version upgrades, testing, and documentation — rather than creative or greenfield development. With enterprise customers, he says Cognition typically sees 8 to 15x speedups: work that would take an engineer eight hours gets done in one. July and August were Cognition's biggest months ever by a wide margin, he adds, pushing back on Census data that showed enterprise AI adoption dipping from 14% to 12%.

The hiring-freeze narrative doesn't match what Cognition is seeing on the ground. Wu says one Latin American bank, a major customer, 10x'd its contract within eight to nine months of onboarding and is growing headcount rather than cutting it. The pattern he observes more broadly is insourcing: companies are pulling work back from contractors and system integrators because Devin makes it faster and cheaper to do it internally.

Windsurf integration

The Windsurf acquisition gives Cognition both an IDE and an agent in a single portfolio, covering the two dominant developer product categories. Wu says both products were already growing independently; combined, growth has accelerated further in the six weeks since the deal closed. Windsurf also brought a go-to-market team with existing enterprise relationships and, notably, a head start on FedRAMP certification — something Cognition is now pursuing as well.

Stack positioning

Wu is explicit that Cognition intends to stay narrow: the company's lane runs from base model to finished product experience, and he has no current plans to move into foundation model training or chip-level infrastructure. He describes Cognition as "0.1% of the way there" and says meaningful success requires focus, not expansion. He doesn't rule out further acquisitions but says there are no active plans.

Defensibility and inference costs

On the question of whether AI application-layer businesses can hold margin as inference costs fall, Wu makes two arguments. The cost-value case for customers is straightforward — if AI makes every engineer, lawyer, or accountant meaningfully faster at a fraction of the labor cost, the ROI math is obvious. The defensibility question is harder and company-specific, but Wu points to two sources of durability: deep specialization in a particular use case, and what he calls a personalization effect — the compounding advantage a tool gains as it learns a customer's codebase, trade-offs, and preferences. Devin's knowledge system, which stores and applies that institutional context, is his clearest example.