Coding agent
A software tool that combines an LLM with access to the development environment to execute coding tasks autonomously.
Definition
A coding agent is a program that uses an LLM as its reasoning engine and gives it access to tools: reading/writing files, executing shell commands, Git operations, running tests. Unlike a classic chat assistant that responds with text, a coding agent acts directly in the development environment. OpenCode, Claude Code, Kilo Code, Cursor Agent, and Aider are examples of coding agents.
postcursors perspective
The current generation of coding agents (2025-2026) marks a clear break from autocomplete tools (GitHub Copilot, Supermaven). The latter suggest code as you type; coding agents execute complete tasks. This isn't a change in degree but in paradigm — which is why optimal workflows are fundamentally different.
In practice
"Running a coding agent" means giving it a task in natural language with a success criterion (tests pass, PR compiles), and letting it iterate. The human reviews the result, not the process.
Common misconceptions
- ✗ Calling GitHub Copilot or Supermaven coding agents — they're autocomplete tools, the distinction matters
- ✗ Believing the coding agent replaces code review — it accelerates production, not validation