Why We Stopped Recommending Cursor for Daily Driver Work — And What We Use Instead

TL;DR: Cursor Composer became unreliable for sessions >90 minutes — context loss, RAM spikes to 8GB+, and forced model upgrades broke workflows. We migrated daily driving to Zed + Continue.dev; Cursor now only for <30 min type-heavy edits. Full comparison →

The Context

Three-dev team, 6–8 hours daily AI-assisted coding across TypeScript/React (frontend) and Go/PostgreSQL (backend). Cursor was default from Jan 2025. By May 2026, every dev had hit: Composer forgetting running dev servers after 60–90 min, RAM creeping to 8–12 GB on M3 Max, and “model update” popups mid-refactor that reset agent context. The breaking point: a 4-hour Stripe webhook migration where Cursor lost the stripe listen tunnel and local Supabase instance 3× — 45 min total recovery time.

What We Tested

Tool	Use Case	Verdict	Why
Cursor Composer	3-hr auth refactor (14 files)	❌	Lost stripe listen tunnel + local Supabase 3×; 45 min total recovery
Zed + Continue.dev	Same refactor, next week	✅	Native performance (no Electron), Continue.dev agent persisted terminal/DB context entire session
Cursor Composer	Quick single-file edits (<30 min)	✅	Fast, LSP-aware, no context loss on short tasks
VS Code + Cline	Greenfield feature (new API endpoint)	✅	Open-source, BYOK, terminal-native agent loops; no vendor lock-in
Zed	Quick edits	✅	Instant startup, modal editing, but Continue.dev panel adds UI overhead for <15 min tasks

The Pivot Point

June 3, 2026: Mid-refactor, Cursor pushed a “model update” notification that forced a Composer restart. Lost the running tunnel, DB connection, and 14-file context. Same task in Zed + Continue.dev — agent kept the tunnel PID, DB connection, and full file context. Finished in one sitting. Team convention locked in: zed + Continue.dev for refactors/migrations/debugging (>90 min); VS Code + Cline for greenfield autonomy; cursor only for <30 min type-heavy edits.

What We Use Now

Zed (free) + Continue.dev (free, self-hostable Ollama models) for sessions >90 min:

• Continue.dev config: ~/.continue/config.json with ollama provider, qwen2.5-coder:32b for planning, deepseek-coder:33b for execution
• Project instructions in .continue/instructions.md: “Maintain terminal state. Never assume server restarted.”
• Zed settings: vim_mode: true, terminal: {dock: "bottom"}, Continue.dev panel pinned right

VS Code + Cline (free, BYOK) for greenfield features:

• Cline settings: apiProvider: "openrouter", model: "anthropic/claude-3.5-sonnet", allowCommands: ["npm run dev", "npm test", "psql"]
• Team convention: cline for new feature scaffolding, test writing, docs

When You’d Choose Differently

• Cursor Composer still wins for TypeScript-heavy work where LSP type-checking catches errors mid-edit — Zed’s TypeScript LSP is solid but Continue.dev doesn’t integrate with Zed’s language server diagnostics.
• Windsurf ($15/mo) if you want IDE-integrated persistent context without leaving a VS Code fork — Cascade agents handle long sessions well.
• Aider (free, terminal-only) if you live in tmux and want git-native BYOK workflow.

Tool Crucible Rating

Dimension	Rating (1–5)	Notes
Overall	3	Cursor still capable for short tasks; reliability regression on long sessions is a blocker for daily driving
Ease of Use	2	Forced updates, context loss, RAM bloat create daily friction
Value	2	$20/mo Pro for degraded experience; free alternatives (Zed, Cline) outperform on reliability
Support	3	Responsive on GitHub issues but root causes (Electron architecture) structural

---

This is part of our AI Coding Tool Evaluation series. See full comparison: Cursor Alternatives 2026: Zed, Cline, Windsurf, Aider Tested