Anthropic cuts off Windsurf's Claude access overnight — AI Engineer World's Fair organizer Shawn Wang explains what it means for the OpenAI acquisition rumors

Summary

Shawn Wang, organizer of the AI Engineer World's Fair, spoke from the conference floor on June 4th. He addressed the Anthropic-Windsurf dispute, the shape of the agentic coding market, GitHub's competitive position, and the state of developer hiring.

Anthropic revokes Windsurf's Claude access

Anthropric cut off first-party Claude access to Windsurf overnight without warning. Users can still bring their own API keys, but the default integrated connection is gone. Wang views this as a significant breach of trust. Google has not taken equivalent action despite similar competitive pressures. Anthropic appears to have classified Windsurf as a competitor product, and Wang sees the move as increasing the odds that the rumored OpenAI acquisition of Windsurf for roughly $3 billion actually closes. Neither Windsurf nor OpenAI has confirmed the price. Windsurf was scheduled to speak at the conference that evening, directly before Greg Brockman's session.

Windsurf's IDE market share is roughly 5 percent of Cursor's, far smaller than the valuation gap suggests. Windsurf's parent company, formerly Codeium, spent four years building GPU serverless inference for enterprises, so the OpenAI acquisition price bundles that infrastructure alongside the IDE.

Coding agents are converging

Both Cursor and Windsurf built large user bases around Claude 3.5 and 3.7, which developers regard as the best coding models. Cursor now faces uncertainty about whether Anthropic might cut access to them as well.

Wang rejects the framing that synchronous IDEs (Cursor, Windsurf) and asynchronous agents (Devin, Factory AI) represent separate permanent markets. They started in different layers of the development loop but are converging. OpenAI's Codex is explicitly designed to merge both paradigms, mirroring how GPT-5 aims to combine reasoning and instant-response capabilities. The technical challenge is that one environment runs on the user's machine and the other runs in the cloud. The standard stack—sandbox, browsing environment, branching, codebase fine-tuning—is now table stakes across Codex, Jules, and Cognition. Execution and customer fit separate them.

GitHub's moat

Microsoft's 2018 GitHub acquisition ranks among the smartest enterprise developer-tools deals ever made, and the competitive moat is stronger than the market recognizes. Code is the most expensive data for a company to hand over. Once it lives with GitHub, coding agents naturally follow. Wang estimates GitHub Copilot revenue at roughly $500 million, enough to be a standalone public company if separated from Microsoft. GitHub's public repo corpus is smaller than YouTube's video library, and Microsoft restricts training on private repos. Code differs from video because it is verifiable: it compiles and runs, creating an RL loop for synthetic training data that video cannot replicate.

Hiring reality

Both pessimistic and optimistic views on developer hiring are correct simultaneously. Half the engineers at Wang's conference worry about their jobs. The other reality is that companies cannot find qualified people. Junior developers fresh out of college face a difficult position unless they are genuinely skilled, because AI amplifies the gap between strong and weak engineers rather than narrowing it. The shift from writing code to managing agents that write code is real but poorly defined. Supervising an agent when it breaks still requires deep technical knowledge. Wang hired an intern this week who shipped a working product before his first day, which he frames as the new template for entry-level engineers.

Conference themes

Voice interfaces are finally mature enough for real products. Security has become a developer priority because actual money now runs through their systems. MCP drew enough interest to fill the main ballroom with the full Anthropic steering committee presenting. A dedicated robotics track launched this year for the first time, with Tesla Optimus, Physical Intelligence, and Waymo all represented. The robotics ecosystem remains vertically integrated with no meaningful horizontal layer. One exception is Cloud Chef, a kitchen robot that runs demonstration learning from a single chef and claims to transfer across any robot arm substrate, pricing its labor at $12 per hour.