Zed’s 2026 collaborative editing and CRDT implementation

Friction in current workflows

Zed provides a shared environment that includes the full lifecycle of code development. Current workflows rely on a trifecta of pull requests, Slack, and screen sharing. Max, one of Zed’s co-founders, describes today’s collaboration as a disjointed process. He envisions a workflow where he can be in a coffee shop, coding solo, and then realize he needs help. He wants to leave a comment in the editor to ask a question without leaving his focus. The developer stays in the code. You might realize you need assistance while coding solo and then have to find where that code exists on a different branch. This jumping between tools ruins the flow. You experience pixelated screen shares in Zoom or Slack that hide your editor theme. VS Code Live Share is in maintenance mode. Zed targets these pain points by integrating collaboration directly into the editor. Thorsten, an interviewer, asks how current tooling makes collaboration difficult.

The workflow aims to remove the jump between editing and discussing code.

The math of convergence

Zed relies on Conflict-Free Replicated Data Types to manage real-time multiplayer editing. These data types allow replicas to modify a document independently and converge to the same state once they exchange data. Instead of using absolute offsets that shift when text changes, Zed uses logical locations called anchors. Every insertion receives a unique ID composed of a replica ID and an incrementing sequence number. This identifier allows replicas to resolve locations without a central server. When a user deletes text, the system marks those fragments with tombstones to ensure the document remains consistent. To apply a remote operation, the system scans the local document for the fragment containing the specified offset of the parent insertion. This process continues recursively, as insertions build upon each other in a tree structure. This mechanism ensures all collaborators see the same content.

The editor uses a "Parallel Agents" model. This model treats AI agents as first-class collaborators with the same write access as humans. When an agent edits a file, the change appears immediately in every collaborator’s view. This capability enables one developer to orchestrate multiple agents, such as using Claude Opus for refactoring and GPT-5.5 for testing, within a single buffer. The model supports four patterns: one developer with one assistant, two engineers pairing live, one driver with multiple agents, and a full multiplayer surface where several humans and several agents share a file. The latter pattern is the closest analogue to a war-room incident response. This design avoids the context-rebuilding tax that consumes roughly a third of total cycle time in team workloads. You know the frustration of re-explaining code context to a teammate. Zed’s model keeps agent reasoning within persistent threads, so the context remains available for the next person who joins. This pattern allows for intense, high-speed collaboration that feels immediate. It is a major departure from the single-player model.

This mechanism ensures all collaborators see the same content.

Performance and ecosystem trade-offs

Performance defines the Zed experience. The editor uses Rust and a custom GPU-accelerated framework called GPUI. Because Zed uses a custom GPU-accelerated rendering framework called GPUI, every pixel renders natively, which eliminates the abstraction layers that make Electron-based editors like VS Code feel significantly slower during heavy workloads. This architecture allows the UI to respond with single-digit millisecond latency. The rendering stays smooth even during heavy refactors.

The extension ecosystem is significantly smaller than VS Code. It uses GPUI. Latency stays low. While VS Code maintains over 100,000 plugins, Zed relies on a WebAssembly model. This allows for a growing but smaller ecosystem. Zed achieves 120 frames per second even with multiple collaborators and a streaming agent. In tests, Zed uses 222 MB of RAM while VS Code uses 3,549 MB. Startup time for a clean install is 0.40 seconds for Zed, while VS Code takes 3.00 seconds. The editor handles 100,000-line files 8x faster than VS Code.

Can Zed eventually match the massive extension catalog of VS Code?