What is an AI coding agent? Definition and examples

An AI coding agent is an LLM-powered tool that writes, edits, and runs code on request — taking a task description and producing the code changes that satisfy it.

The difference from earlier AI-coding tools is in how much of the work the agent does autonomously. An autocomplete tool finishes the line you started typing. An agent reads the task, opens the files it needs, decides what to change, applies the edit, often runs the tests, and reports back. The human stays in the loop for the high-level decisions; the agent does most of the keystrokes.

What an agent needs from the human

The bottleneck is no longer the typing — it is the context handed to the agent. A vague task ("make the cart better") produces vague changes. A task with the screen attached, the source URL named, the broken behavior described, and the expected behavior stated produces changes that match the intent.

That is why visual feedback formats matter for agent workflows. A markdown document with a cropped screenshot, the source URL, and dictated explanation gives the agent enough context to act without a second round of clarification — the same shape that worked for humans is what works for agents.