AI glossary · Beginner · Agents
Coding agent
A coding agent is an AI program in which an LLM takes a programming task requested in plain language, plans it on its own, and carries it through to the end by reading, editing and running code.
About this term
How it works
When a user states a goal such as 'fix the login bug', the LLM first plans what needs to be done. It then calls tools directly, such as reading files, editing code and running terminal commands. It checks the outcome by running tests or reading error messages, and if problems remain, it revises the plan and repeats the process. When it judges the goal has been met, it summarizes the changes and reports them to the person.
Why it matters
Earlier code autocompletion only suggested a line or two, but a coding agent can take on work that spans multiple files at once. Developers can hand off repetitive edits or test writing and focus more on design and review. This is why coding is cited as the field where AI agents first took hold in real work. Development tools and cloud services are also releasing features in forms that coding agents can easily use.
Things to know
Coding agents can also hallucinate, using functions that don't exist or producing plausible but wrong code. Because they run commands directly, it is important to limit their permissions so they don't do things that are hard to undo, such as deleting files. There is also a risk of prompt injection, where instructions hidden in code or documents trick the agent. In the end, a person must review the output and verify it with tests.
Example
Developers use a coding agent in an editor or terminal by assigning a task such as 'add tests for this feature', then reviewing the changes the agent made before applying them. This site also covered the news 'AWS releases aws-ai-ml, SageMaker inference optimization skills for coding agents', which illustrates this trend well: a cloud company is separately packaging working knowledge for coding agents, not people, to read and use.