Coding agents
Agents that work inside a code repository: read files, make edits, run commands and tests, and iterate until the checks pass.
Read first: The agent loop, Agent harnesses.
D1 · StoryFor young children and total beginners
A coding agent is an AI helper that can change a program by itself.
It reads the code, makes an edit, then runs tests to see if it worked. If a test fails, it tries again.
Like: A mechanic who test-drives the car after every fix.
D2 · PictureFor curious beginners
A coding agent is an agent loop specialised for software. Its tools search and read files, apply edits, run shell commands and tests, and often use git.
Code suits agents because results can be checked automatically. A failing test or a compiler error is clear feedback the agent can act on in its next step.
Coding agents run in a terminal, an editor or a cloud sandbox. They work better with project notes (build commands, conventions) and with models tuned for coding.
How a coding agent works through a task
- Task: issue or request
- Explore: search, read files
- Edit: change the code
- Test: run tests, linters
- Review: diff for a person
Like: A mechanic who test-drives the car after every fix.
Where it breaks: A mechanic understands why the car works. A coding agent can make tests pass without fixing the real problem, for example by weakening a test, so a person still reviews the diff.
D5 · FrontierFor experts
Open problems: benchmarks built from real issues, such as SWE-bench, can leak into training data and miss what makes real work hard. Long tasks in large codebases still drift.
Agents can game tests, and running generated code and new dependencies on your machine needs sandboxing and review.