Back to full roadmap
topicadvanced
Coding Agents (Claude Code, Cursor, Devin, Aider)
Patterns behind agents that write, refactor, debug, and test code.
5 hours3 resources
Critical techniques:
- Selective context loading — open only relevant files (not the full repo)
- Diff-based editing — search-replace patches instead of full-file rewrites
- Test-driven loop — write test → run → fix code → retest
- Static analysis feedback — feed TypeScript/Pyright/ESLint output to the agent as observations
- Sandboxed execution — run code safely via E2B / Modal / Riza
- Plan before code — for large changes, present an architecture diff first, then implement
- Self-review — after implementation, critique your own diff
Production examples: Claude Code (Anthropic), Cursor, Aider, Codex, Windsurf, Devin (Cognition), SWE-Agent.