We've been daily-driving Zed across a large TypeScript monorepo and a couple of smaller side projects for months — first out of curiosity about the "fastest editor" claims, then because it earned a permanent spot next to our other tools.
Here's the honest version: Zed is not trying to out-build Cursor's AI. It's trying to be the fastest, lightest place to work, with or without AI doing the driving. That's a different bet, and it mostly pays off.
The speed is not marketing
Cold start lands around 0.4–0.6 seconds for us, against 1.2–1.3 seconds for VS Code-based editors on the same machine. Idle memory sits in the 150–250MB range. On our large monorepo, Zed held steady around 200–400MB of RAM where Cursor settled into 500–800MB with its indexing active.
That's not a benchmark-page number — it's the difference between an editor that never makes you wait and one that occasionally makes you feel it thinking.
The more interesting bet: don't build the AI, host it
Zed's own in-editor agent panel is capable — multi-file edits, MCP tool support, edit predictions that are solid if not Cursor-Composer-deep. But the feature that actually sets Zed apart is the Agent Client Protocol (ACP): an open standard that lets you run Claude Code, Codex CLI, or Gemini CLI as the actual agent driving the editor, with diffs rendered inline instead of scrolling past in a terminal pane.
If you already live in Claude Code or Codex CLI from the terminal, this is the upgrade you didn't know you wanted — same agent, same models, but file edits show up as real inline diffs in an editor UI instead of a wall of terminal text. ACP is agent-agnostic by design, so Zed isn't betting the whole product on one vendor's model staying the best forever.
Native human collaboration, not just AI chat
Zed ships real-time multiplayer editing — shared cursors, voice chat, true co-editing in the same file — as a core feature, not a bolted-on extension. We paired on a bug fix across two machines and it felt closer to Google Docs than screen-sharing. This is a category Cursor doesn't compete in at all; its collaboration story is entirely AI-facing.
Where it still loses to Cursor
Composer handles multi-file agentic edits as a mature, self-contained system, and Cursor's tab completion predicts multi-line blocks with project-specific pattern awareness — both draw on a continuously updated codebase index. Zed's own agent panel isn't trying to match that depth yet, and in practice it doesn't. If you want the single deepest built-in AI experience with zero external setup, Cursor is still ahead.
The extension marketplace is also smaller than VS Code's — mainstream languages (TypeScript, Rust, Python, Go) feel solid, but a niche VS Code extension may not have a Zed equivalent yet.
Pricing
- Personal (Free): the full editor, 2,000 accepted edit predictions/month, unlimited if you bring your own agent or API key.
- Pro ($10/month): unlimited edit predictions, $5 of hosted AI credits included, usage-based billing beyond that at API list price + 10%.
- Business ($30/seat/month): org-wide model policies, data governance, RBAC, unified spend visibility.
Because the editor itself never requires a subscription, the free tier is genuinely usable for daily work if you supply your own agent or API key — not a crippled trial.
Who it's actually for
Zed makes sense if you're on a codebase where editor latency genuinely slows you down, you already run an external agent and want a faster native home for it, you need real human-to-human collaboration, or you want a fully open-source editor you can inspect. It's probably not for you if you want the deepest built-in AI with zero setup — that's still Cursor — or if you depend on a niche VS Code extension Zed's marketplace doesn't cover yet.
Rating: 4.0/5. Full review, pros/cons breakdown, and our head-to-head with Cursor: devtoolsreview.com/reviews/zed-review
Top comments (0)