What is a coding agent?
A coding agent is an AI system that works on software the way a developer does: it reads the codebase, plans a change, edits files, runs builds and tests, reads the failures, and iterates until the task is done. Unlike autocomplete, it operates in a loop with tools — the model decides the next action, executes it, and learns from the result.
How coding agents work
The loop is perceive → act → check: read relevant code, make an edit or run a command, observe the output, adjust. The quality ceiling comes from two things — the model's reasoning, and the context it can hold. The second is the underrated one: an agent that spends its context window re-reading files has little room left to think.
Where they struggle
Large codebases. A human developer carries a mental map built over months; an agent starts cold every session and rebuilds its map by reading — slowly, expensively, and partially. This is why agent infrastructure matters as much as the model: tools that serve structured knowledge about the repository (who calls this, what breaks if it changes) replace thousands of lines of re-reading with answers.
They also inherit every security concern of autonomous systems — a coding agent that browses documentation or reads issue text is exposed to prompt injection, and one that holds real API keys can leak them.