Why We Switched Our Daily Driver Editor to Zed — 30-Day Honest Review

TL;DR: Zed eliminates Electron bloat — cold start <1s, RAM ~400MB vs Cursor’s 8GB+. Modal editing and built-in terminal are excellent. But Continue.dev agent doesn’t hook into Zed’s LSP diagnostics, extension library is thin, and team onboarding friction is real. Full comparison →

The Context

Three-dev team (2 senior, 1 mid) on M3 Max MacBooks. Daily stack: TypeScript/React/Next.js frontend, Go/PostgreSQL backend, heavy Docker Compose local dev. Migrated from Cursor (Jan–May 2026) after repeated context loss on long sessions. Evaluation period: June 1–30, 2026. All three devs used Zed as primary editor for 30 days.

What We Tested

Tool	Use Case	Verdict	Why
Zed (stable)	Daily driving (all tasks)	✅	Native speed, modal editing, built-in terminal, multiplayer collab — zero Electron overhead
Zed + Continue.dev	Long refactors (>90 min)	⚠️	Agent persists terminal context but can’t see Zed’s LSP errors inline
Zed + Claude Code (terminal)	Greenfield features	✅	cc in Zed terminal = best of both worlds; terminal-native autonomy
Cursor	Type-heavy TypeScript refactors	✅	LSP diagnostics in Composer panel catch type errors mid-edit
VS Code + Copilot	Team onboarding / junior pairing	⚠️	Familiar but slow; Copilot chat latency noticeable vs local models

The Pivot Point

Day 3: Junior dev opened a 2,000-line TypeScript file. Cursor: 8s lag, 2GB RAM spike. Zed: instant, 400MB RAM. Day 12: Senior dev ran 4-hour Supabase schema migration in Zed terminal + Continue.dev — agent maintained supabase db push tunnel, connection pool, and file context entire session. Day 22: Team realized Continue.dev can’t surface Zed’s diagnostic channel — agent writes code that passes LSP but introduces type errors only caught at npm run build. Workaround: run tsc --noEmit in terminal as pre-commit hook.

What We Use Now

Zed (free, stable channel) as daily driver:

• vim_mode: true, terminal: {dock: "bottom", shell: "fish"}, theme: "Catppuccin Mocha"
• Keybindings: ctrl+shift+p → command palette, space → leader (custom), g d → go to definition
• Extensions installed: docker, github, marksman (markdown), rust-analyzer (side projects)

Continue.dev (free, Ollama backend) for AI assistance:

• Models: qwen2.5-coder:32b (planning), deepseek-coder:33b (execution), codellama:13b (fast edits)
• Config: .continue/config.json with contextProviders: ["terminal", "files", "codebase", "docs"]
• Gap: No diagnostic provider — we run pnpm typecheck in terminal before commits

Claude Code (terminal, $20/mo Anthropic) for greenfield autonomy:

• alias cc='claude-code --allowedTools "Bash,Edit,Write,Read,Glob,Grep"'
• Runs in Zed terminal panel — no context switching

When You’d Choose Differently

• VS Code / Cursor if your team relies on the extension ecosystem (ESLint inline, Prettier format-on-save, GitLens, Prisma, Tailwind CSS IntelliSense) — Zed’s extension API is Rust-based, young, and ~50 extensions vs VS Code’s 50,000+.
• Windsurf ($15/mo) if you want AI-native IDE with persistent context AND full VS Code extension compatibility — Cascade agents + VS Code extension host.
• Neovim + Avante.nvim if you’re already vim-native and want modal editing + AI without leaving terminal.

Tool Crucible Rating

Dimension	Rating (1–5)	Notes
Overall	4	Best raw editing experience; AI integration gaps prevent 5
Ease of Use	4	Modal editing learning curve (2 weeks); then faster than VS Code
Value	5	Free, open-source, no telemetry, no subscription
Support	3	Small team, GitHub issues responsive but feature velocity slow

---

This is part of our Editor Evaluation series. See full comparison: Zed vs Cursor vs VS Code vs Windsurf: 30-Day Daily Driver Test