IDEs
These guides apply the practice sequence to specific AI coding environments. Each tool has different defaults for models, context attachment, and usage dashboards — the optimization order is the same; the controls are not.
Start with Where to start for the general technique stack, then open the guide for your editor.
Evaluating a switch? See Alternatives for “alternative to X” comparisons ranked by token economics.
Tool landscape
| Tool | Type | Price | Open source | Best for |
|---|---|---|---|---|
| Cursor | Full IDE | $20/mo | ✓ | Usage dashboard, pools, MCP |
| Grok Bot | Cloud agent desktop | Plan* | ✕ | Hub — weekly usage, Bots, routines, shared computer |
| Windsurf | Full IDE | Free* | ✓ | Hub — Cascade quota, rules, MCP, SWE routing |
| Devin | Full IDE + cloud agent | Free* | ✓ | Hub — cloud ACUs, CLI, Local, MCP, instructing |
| Zed | Full IDE | Free* | ✓ | Hub — tokens, local models, context |
| Trae | Full IDE | Free* | ✕ | Hobbyist experimentation |
| Cline | VS Code extension | Free* | ✓ | Task-driven agentic work |
| GitHub Copilot | VS Code extension | $10/mo | ✕ | Corporate standard |
| Claude Code | Terminal agent | Free* | ✓ | Automated testing and PRs |
| OpenCode | Terminal agent | Free* | ✓ | Hub — stats, agents, rules, context, MCP, providers |
| Aider | Terminal agent | Free* | ✓ | Terminal-heavy users — git-native edits |
*Free tier or BYOK — you still pay upstream model providers (or quota allowances) unless the tool includes bundled inference.
Using a gateway?
Many terminal agents and extensions route through an LLM gateway for BYOK — one API key across Cline, Aider, and custom agents. See Gateways for OpenRouter, LiteLLM, Vercel AI Gateway, and more.
Strengths and tradeoffs
| Tool | Strengths | Tradeoffs |
|---|---|---|
| Cursor | Best-in-class UX, Composer | Subscription cost |
| Grok Bot | Persistent cloud VM, multi-Bot handoffs | Cloud agent burn; shared computer boundary |
| Windsurf | SWE models free on quota plans, strong Cascade | Cognition ecosystem overlap with Devin |
| Devin | Adaptive routing, cloud + local surfaces | Multiple meters (ACUs, credits, quota) |
| Zed | High performance, GPU-rendered | Fewer extensions than VS Code |
| Trae | Adaptive learning, free tier | Data privacy concerns — verify retention settings |
| Cline | Open source, high agency | Requires setup (API keys, MCP) |
| GitHub Copilot | Deep ecosystem integration | Closed models, premium request caps |
| Claude Code | CLI native, direct OS access | No GUI |
| OpenCode | Plan/Build split, compaction | Terminal-only workflow |
| Aider | Excellent git integration | Steep learning curve |
Guides
Each guide follows the same sequence: measure → model routing → trim context → tighten prompts → set guardrails.
Full IDEs
| IDE | Guide | Dominant waste |
|---|---|---|
| Cursor | Tokenminning in Cursor | Usage dashboard, marathon threads, MCP bloat |
| Grok Bot | Tokenminning in Grok Bot | Weekly usage, catch-all Bots, broad routines |
| Windsurf | Tokenminning in Windsurf | Quota, Cascade, rules, MCP |
| Devin | Tokenminning in Devin | Cloud ACUs, CLI, always_on rules, Arena runs |
| Zed | Tokenminning in Zed | Tokens, local models, broad @ context |
| Trae | Tokenminning in Trae | Builder session sprawl, adaptive context growth, default frontier models |
VS Code extensions
| Tool | Guide | Dominant waste |
|---|---|---|
| Cline | Tokenminning in Cline | Auto-approved tool loops, oversized MCP schemas, marathon task threads |
| GitHub Copilot | Tokenminning in GitHub Copilot | Agent mode context, premium request burn, chat vs inline misuse |
Terminal agents
| Tool | Guide | Dominant waste |
|---|---|---|
| Claude Code | Tokenminning in Claude Code | Bash/tool output in context, CLAUDE.md bloat, headless session sprawl |
| OpenCode | Tokenminning in OpenCode | Token usage, Agents, Context, MCP, Providers |
| Aider | Tokenminning in Aider | Repo-map bloat, architect/edit cycles, conventions file duplication |