The term is broadly applied to anything from a simple code-generation chatbot to a fully autonomous agent that can complete multi-day engineering tasks. In practice, the most useful coding agents sit in the middle: they can operate autonomously on a well-defined task for tens of minutes to several hours before requiring human input.
Well-known coding agents include Claude Code (Anthropic), GitHub Copilot Workspace, Devin (Cognition AI), and OpenAI's Codex-based agents. Each makes different tradeoffs on autonomy, cost, and the breadth of tools available to the agent at runtime.
For teams adopting coding agents at scale, the key decisions are: which tasks to assign to agents versus humans, how to review agent output efficiently, how to set cost limits, and how multiple agents coordinate when working on the same codebase simultaneously.