Windsurf Is Now Devin Desktop: What to Do Right Now

Windsurf silently became Devin Desktop on June 2, 2026. If you restarted the editor on or after that date, you are already running it — Cognition shipped the transition as a standard over-the-air update, and its FAQ states that all of your Windsurf settings are ported to Devin Desktop automatically. For most users the migration is genuinely zero-effort.

There is one thing you cannot ignore, and it is the reason this is a "do this week" post rather than a news recap: Cascade, the local agent that powered Windsurf, is end-of-life on July 1, 2026. Cognition's documentation says plainly that "Cascade remains available through July 1st." Any CI pipeline, automation script, or workflow rule that explicitly invokes Cascade must be repointed to Devin Local before that hard deadline.

This guide leads with the actionable checklist, then explains what actually changed under the rebrand — the new Agent Command Center, the open Agent Client Protocol bet, and the 2025 acquisition arc that ended with Cognition owning the Windsurf brand. Every dated claim below traces to a primary source listed at the points it appears.

On June 2, 2026, Cognition published Windsurf Is Now Devin Desktop on the devin.ai blog and pushed the change as an over-the-air update. There was no migration wizard and no reinstall. Users who restarted the editor on or after June 2 received Devin Desktop automatically, with their existing account, plan, extensions, and keybindings intact.

The temptation is to file this as a logo swap. It is not. The default experience now opens on the Agent Command Centerrather than the editor canvas — a shift from "code-first IDE that can call an agent" to "agent-management hub that contains a full IDE." The editor is still there, one click away. But the framing has changed, and the framing is the point.

Most of these take minutes. The point of running them now, rather than on June 30, is that the only item with a hard deadline — Cascade-dependent automation — is also the one most likely to fail silently in a build pipeline where nobody is watching the editor.

From June 2, Devin Local is the default local agent inside Devin Desktop. Cascade is not deleted overnight — Cognition's FAQ keeps it available through July 1, 2026 — but it is on a fixed exit ramp. After that date, the migration window closes.

Devin Local is a full Rust rewrite of Cascade. Cognition states it achieves up to 30% greater token efficiency and adds support for subagents — parallel task handling that Cascade lacked. We treat the 30% figure as vendor-stated: no independent benchmark had confirmed it as of this writing, so weigh it accordingly before basing a budget on it. The directional claim — fewer tokens, parallel subagents — is what to plan around.

Powering the paid experience is SWE-1.6, Cognition's current proprietary coding model, released April 7, 2026 and included free on paid Devin plans alongside frontier models from Anthropic, OpenAI, and Google. Cognition reports it runs at up to 950 tokens per second on the fast tier (Cerebras) and 200 tokens per second on the free tier (Fireworks). On benchmarks, Cognition states SWE-1.6 improves on its predecessor SWE-1.5 by more than 10% on SWE-Bench Pro — we use the company's own "more than 10%" wording rather than the specific point scores circulating in community testing, which have not been confirmed by a standalone Cognition benchmark post.

For a team weighing the switch, the practical lens is token spend per task, not headline tokens-per-second. A more token-efficient local agent compounds across thousands of agent runs — which is exactly the kind of math worth modeling against your own usage. Our breakdown of per-task agent cost walks through how to do that estimate honestly rather than trusting a single efficiency percentage.

The Agent Command Center is the headline new surface. It presents every agent session — local Devin Local runs and cloud Devin sessions side by side — as a Kanban board sorted by status: in progress, blocked, and ready for review. Instead of babysitting one agent in one editor tab, you dispatch and monitor many from a single screen.

A companion feature, Spaces, groups sessions, pull requests, files, and context into shareable workspaces so multiple agents can read the same project state. In Cognition's launch materials, early adopters frame the value as consolidation — one place to dispatch and watch agents rather than a scatter of terminals and tabs.

The most strategically interesting part of the rebrand is the quietest. Devin Desktop ships as a multi-agent IDE built on the Agent Client Protocol (ACP) — an open-source protocol (Apache 2.0, JSON-RPC 2.0 over stdin/stdout) that standardizes communication between code editors and coding agents. The clean analogy: ACP is to coding agents what the Language Server Protocol was to language tooling.

At launch, Devin Desktop supports a roster of first-class agents inside the editor: Codex (OpenAI), Claude Agent (Anthropic), OpenCode, Junie (JetBrains), and Gemini CLI (Google). By June 2026 the protocol had been adopted by JetBrains, Google, GitHub, and 25+ agents. The meaning is hard to overstate: Cognition is not betting that Devin's model wins. It is betting that the IDE becomes a neutral, multi-agent orchestration layer — and that owning the best host wins even when someone else owns the best model.

For comparison-minded readers, this is a different answer to the same problem Cursor tackles with its agent architecture. We covered Cursor 3's agent and Composer model separately; the contrast between a single-vendor agent window and an open multi-agent host is the central design fork in 2026 IDEs.

The rebrand makes sense only against the chaotic 2025 that produced it. Codeium rebranded as Windsurf on April 4, 2025, pivoting from an autocomplete plugin to a full AI-native IDE; the Windsurf Editor had attracted a reported one million-plus developers by that point. Then three deals reshaped the company in a single 72-hour stretch in July 2025.

OpenAI had made a reported $3 billion offer to acquire Windsurf, but the deal's exclusivity period expired on July 11, 2025 when OpenAI and Microsoft could not agree on IP access — per TechCrunch, "OpenAI didn't want its largest backer to get Windsurf's AI coding technology as well." Within 72 hours, Google DeepMind signed a reported $2.4 billion reverse-acquihire: a nonexclusive technology license plus compensation to hire CEO Varun Mohan, co-founder Douglas Chen, and key researchers, taking no equity stake. Days later, on July 14, 2025, Cognition — the maker of the Devin coding agent — acquired Windsurf's IP, product, trademark, and the roughly 210 remaining employees. The price was reported at approximately $250 million by third-party outlets; Cognition has not disclosed the terms.

Cognition framed its acquisition in mission terms. The deal was negotiated at extraordinary speed — its president has described going "from first call after 5pm on Friday to a signed definitive agreement this morning." The company has since compounded: a reported $1 billion-plus round at a $26 billion valuation closed May 27, 2026, with Cognition citing a roughly $492 million revenue run rate. Windsurf's reported $82 million ARR and 350-plus enterprise customers at acquisition give a sense of what Cognition absorbed — and the install base that now seeds Devin Desktop. We place this saga inside the broader upheaval in our H1 2026 AI coding retrospective.

Most published comparisons predate the rebrand or skip Devin Desktop's agent-management angle entirely. The table below positions the post-rebrand Devin Desktop against the tools it actually competes with on a procurement shortlist. Pricing and feature cells reflect vendor pricing pages and primary sources as of June 2026; verify current numbers before committing budget.

The honest read for a procurement decision: these tools are converging on shared protocols (ACP, MCP) faster than they are diverging on capability. Market-share splits often quoted for this category — a Copilot plurality, Cursor in the low-to-mid twenties — are analyst and survey estimates rather than disclosed figures, so treat them as directional. For a fuller cross-tool view, our Q2 2026 agentic coding tool matrix and our AI coding adoption statistics give the wider numbers behind these positions.

Here is our read, labeled as analysis rather than reported fact. The Windsurf saga is the clearest case yet that standalone AI IDEs are being absorbed into the platforms whose models they depend on. An editor that wraps someone else's frontier model has a structural problem: the model vendor can ship its own editor at any time. Codeium answered that by becoming Windsurf — a full IDE. Cognition is answering it differently, by making the IDE a neutral host that runs every agent, including its competitors'.

That is why ACP, not the new logo, is the story. If the IDE is where agents are coordinated rather than where one model is privileged, the defensible position is owning the coordination layer — the Command Center, Spaces, the shared-context primitives — not owning the smartest model. Cognition appears to be pricing this bet aggressively: a company that grew its run rate roughly thirteenfold in a year and now states that the large majority of its own code is written by its AI is signaling that it believes the orchestration layer is where the value accrues.

Looking forward, we expect the next twelve months to reward whoever makes multi-agent coordination boring and reliable. The teams that win will not be the ones chasing the highest benchmark on a single model; they will be the ones whose host lets a developer dispatch five agents, see which is blocked, and ship the one that is ready — without leaving the window. Devin Desktop is an early, explicit bet on that future. Whether it pays off depends on execution and on whether ACP becomes a genuine standard or fragments. For the month-earlier baseline this post updates, see the May 2026 AI IDE landscape — Windsurf was still a standalone brand in it.