Devin Desktop — Windsurf Is Gone, Cascade Dies July 1

On June 2, 2026, every Windsurf installation silently became Devin Desktop via an over-the-air update. Settings ported automatically. No migration wizard, no opt-in — you opened your editor and the name on the tin had changed. On July 1, nine days from now, Cascade gets removed entirely. If you have CI scripts, automation rules, or .windsurfrules configs that invoke Cascade by name, they break. Interactive in-editor use migrates automatically; named automation does not. Audit your configs before the deadline.

TL;DR

• What: Windsurf silently rebranded to Devin Desktop on June 2; Cascade reaches end-of-life July 1, 2026
• Replacement: Devin Local (Rust rewrite) replaces Cascade — claimed 30% more token-efficient, supports subagents
• Architecture shift: Default launch surface is now Agent Command Center (Kanban board), not the editor canvas
• Action: Audit any automation referencing Cascade by name before July 1; interactive use migrates automatically

What Happened

Cognition shipped a rebrand that goes deeper than the name. Three things changed at once.

First, the default launch surface flipped. When you open Devin Desktop, you land on an Agent Command Center — a Kanban board with columns for Running, Waiting for Review, and Done. The full IDE is still there: VS Code-compatible editor, extensions, keybindings, LSPs, all backwards-compatible. But it is now one click away from the default view, not the default view itself. That is a deliberate product statement: you manage agents, and the editor is where agents do their work.

Second, Devin Local replaces Cascade as the local agent. Cognition rewrote it from scratch in Rust and claims it is up to 30% more token-efficient than Cascade. That figure is self-reported — Cognition states “in our testing” without publishing methodology or independent benchmarks, so verify it against your own workloads before counting on it. The meaningful architectural difference is subagent support. Cascade operated as a single-context-window agent taking one turn at a time. Devin Local can spawn subagents, which matters for complex multi-step tasks where parallel execution reduces wall-clock time.

Third, Devin Desktop ships with Agent Client Protocol (ACP) support at launch. ACP is the open-source protocol (Apache 2.0) created by Zed Industries in August 2025 and since adopted by JetBrains, Google, GitHub, and over 25 agents. This means Codex, Claude Agent, OpenCode, and custom in-house agents can run as first-class citizens inside Devin Desktop. You are not locked into Cognition’s agent — you bring your own.

Cascade EOL is July 1, 2026. Any CI pipeline, shell script, or automation config that explicitly references Cascade by name will stop working. Interactive editor sessions migrate automatically, but named invocations do not. Search your repos for cascade references now.

Why This Matters

Most tool rebrands are marketing exercises — new logo, same product, move on. This one is architecturally different because it repositions the IDE as a secondary surface behind an agent management layer.

The Agent Command Center default is not a cosmetic choice. It signals that Cognition views the future of development as managing multiple autonomous agents working in parallel, not as a human typing code with AI suggestions. Compare this to Cursor, which keeps the editor as the primary surface and treats its AI features as augmentations to the typing experience. Both bets could be right for different workflows, but they are genuinely different product philosophies — not feature parity with different marketing.

ACP support is what makes this bet credible rather than aspirational. Without it, “agent operating system” would just mean “our agent plus our editor.” With ACP, Devin Desktop becomes a multi-agent host. If you already use Claude Agent for reasoning-heavy tasks and Codex for straightforward implementations, you can now orchestrate both from the same interface with shared context via Spaces (Cognition’s feature for bundling agent sessions, PRs, files, and context around a single task). No other IDE shipping today offers this kind of multi-agent orchestration as a first-class feature.

ACP compatibility means you are not locked into Cognition’s agent ecosystem. If Devin Local underperforms on your codebase, swap in Claude Agent or OpenCode without switching editors. That optionality is the actual value of the protocol — not the protocol itself.

Pricing tells a subtler story. Pro stays at $20/month, Max at $200/month, Teams at $80/month plus $40/seat. The old Windsurf had a $15/month price advantage over Cursor Pro — that disappeared in the March 2026 pricing restructure and the rebrand did not bring it back. At pricing parity with Cursor, the differentiation case rests entirely on the Agent Command Center, Spaces, and ACP multi-agent support. If you do not need multi-agent orchestration, Cursor is the simpler, more proven tool at the same price. If you do, Devin Desktop is currently the only IDE offering it.

The Take

I have watched every major AI coding tool pitch itself as “the last IDE you’ll ever need.” Cognition’s move is different: they are not pitching a better editor, they are pitching an agent operating system that happens to contain an editor. The Agent Command Center-first default is a deliberate statement about where development is heading — you manage agents, the editor is a side panel. That is a real product bet, not a rebrand.

What makes me skeptical is the governance context. The Windsurf founders who built the IDE — CEO Varun Mohan, co-founder Douglas Chen, and key engineers — are now at Google DeepMind, hired in a deal reportedly worth $2.4 billion. Cognition picked up the remaining entity for roughly $250 million and is shipping on a codebase the original architects left behind. The product is solid right now. The Rust rewrite of the local agent is technically sound. ACP integration is genuinely forward-looking. But the 12-month roadmap is a trust exercise in a team that inherited a codebase rather than building it.

For teams evaluating this today, here is the decision framework: if you need multi-agent orchestration and want a single IDE surface to manage it, Devin Desktop is the only shipping option. If you are a solo developer or small team that primarily uses one AI model for inline assistance, Cursor remains simpler and equally capable at the same $20/month price point. If you are already deep in Windsurf’s ecosystem, the migration is automatic for interactive use — just audit your automation before July 1.

The multi-agent host thesis is the right long-term bet. Whether Cognition is the right team to execute it without the founders who built the foundation — that is the question you are actually answering when you commit to this tool.