Claude Code
Anthropic's terminal-first AI coding agent — best-in-class for long autonomous tasks and tool use.
Best for
- Long autonomous tasks (large refactors, test suite generation, codebase analysis)
- Engineers who live in the terminal and SSH into remote machines
- Workflows that need hooks, custom subagents, MCP servers, or skills
Weak at
- GUI lovers — there is no first-party visual diff/UI
- Discoverability — its capabilities are deep but require reading docs
Pricing
| Tier | Price | Notes |
|---|---|---|
| Claude Pro | $20 / month | Includes Claude Code with usage caps |
| Claude Max | $100 / month | Higher caps; recommended for daily heavy use |
| API | Pay-as-you-go / usage | Use your own Anthropic API key |
First-hand notes
TODO replace with your first-hand testing notes before publishing. Suggested angles: how it handles a 4-hour-equivalent task, what hooks and subagents actually unlock, the difference between Sonnet and Opus models on real work, friction points (e.g. permission prompts, context management), and what made you switch (or not switch) from Cursor.
What it is
Claude Code is Anthropic’s CLI agent. You run it in your terminal in the directory you want to work on, give it a goal, and it edits files, runs commands, and checks its work — pausing for permission on destructive operations.
What it does well
It is currently the best at long, autonomous, tool-using tasks among the major coding agents. The hooks system, subagents (parallel specialized agents), skill packs, and MCP server support give power users a deep configuration surface.
For tasks that take 30+ minutes to complete, it stays on-task far better than Composer-style in-editor agents.
Where it falls short
There is no first-party GUI. You are reading and reviewing diffs in the terminal. If that is not your preference, Cursor or Windsurf are closer to your existing habits.
Discoverability is the other weak spot — features like skills, hooks, and subagents are powerful but easy to miss.
Verdict
The sharpest tool for serious autonomous work — most capable per dollar at the Max tier. The terminal-first model is a feature, not a bug, but expect a learning curve.
Alternatives to Claude Code
Cursor
Free + $20VS Code fork with the best Tab completion in the business and a competent Composer agent.
GitHub Copilot
Free + $10The original AI coding assistant — broad IDE support, predictable pricing, and the deepest GitHub integration.
Aider
Free + VariableOpen-source CLI coding agent — bring your own API key and get a power-user terminal workflow.
Windsurf
Free + $15VS Code fork with the Cascade agent and an unusually generous free tier.