July 2025, Revisited: The Windsurf Saga and the Weekend Vendor Risk Got Real

This post is part of our Three-Year Retrospective series: thirty-six posts, one per month, looking back at what actually mattered in software engineering. This one covers July 2025.

If you want a single story that captures how strange the AI tooling market was in 2025, it's the second weekend of July. On Friday, July 11, OpenAI's roughly $3 billion deal to acquire Windsurf — the AI-native IDE formerly known as Codeium — officially fell apart. Hours later, Google announced it was paying about $2.4 billion to hire Windsurf's CEO Varun Mohan, co-founder Douglas Chen, and a group of senior researchers into DeepMind, taking a non-exclusive license to Windsurf's technology but no equity and no control. By Monday, Cognition — the company behind the Devin coding agent — had acquired everything that remained: the IDE, the IP, the enterprise customer base, and the roughly 250 employees the Google deal left behind. A $3 billion acquisition turned into a three-way split in roughly 72 hours.

How a deal that size falls apart

The OpenAI acquisition had been reported since the spring, with a signed letter of intent. What killed it, by most accounts, was Microsoft: OpenAI's partnership terms gave Microsoft rights to IP that OpenAI acquired, and Windsurf's technology landing inside Microsoft — owner of GitHub Copilot, Windsurf's direct competitor — was a problem nobody could negotiate away. When the exclusivity window expired, Windsurf was free to talk to anyone, and Google moved within hours.

The Google arrangement deserved more attention than it got amid the drama. It wasn't an acquisition. It was a "license-and-hire" — pay for the people and a non-exclusive license, leave the company standing. Regulators couldn't review a merger that never happened. It was the most visible example yet of a structure that would recur across the industry, and it left an awkward question hanging: what happens to the product, the customers, and the employees who don't get hired?

The week the customers noticed

That question is why July 2025 mattered to working engineers and not just deal-watchers. Windsurf had real enterprise customers — hundreds of them — who had standardized on an AI IDE, trained their teams on it, built workflows around its agentic features, and then watched its founders walk out the door on a Friday news cycle. For a weekend, nobody could tell them who owned their vendor's roadmap.

Cognition's acquisition stabilized things quickly, and to their credit the product kept shipping. But the lesson had landed. We spent the rest of 2025 fielding a version of the same question from engineering leaders: how exposed are we if this happens to our tool? It was the right question. The AI developer-tools market was consolidating in real time, and the assumption that your IDE vendor would exist in recognizable form in eighteen months was no longer safe — not because the products were bad, but because the strategic value of the teams building them exceeded the value of the businesses themselves.

What we started telling teams

The practical guidance we settled on that summer still holds. First, prefer tools whose core artifacts are portable: if your prompts, rules files, MCP configurations, and workflows can move between IDEs, an acquisition is an inconvenience rather than a crisis. Second, treat AI tooling selection as a process with a review cadence, not a one-time decision — the market was moving too fast for three-year commitments. Third, distinguish between betting on a model and betting on a product; in 2025 those were different risk profiles, and the Windsurf customers who understood that recovered fastest.

Looking back from June 2026

Windsurf under Cognition turned out fine — the IDE survived, integrated with Devin's agentic capabilities, and kept its enterprise base. The license-and-hire structure became a recognized pattern with its own name and its own regulatory scrutiny. And the vendor-risk reflex that July 2025 instilled proved useful again and again as the market kept consolidating into 2026: teams that built portability into their AI workflows absorbed each shock as a migration task, not an emergency. The saga reads today less like an aberration and more like the moment everyone learned the rules of the new market.

If your organization is making AI tooling decisions with one eye on vendor risk, Choosing Your AI IDE works through exactly this evaluation — capability, portability, and continuity risk — and Agentic Coding with Windsurf gets teams productive on the tool that survived the most dramatic weekend in dev-tools history.