Agentic Coding
An AI that writes software by working in a loop — editing files, running tests, fixing failures — not just suggesting snippets.
Think of it like
The jump from a spellchecker that flags words to an editor who rewrites the chapter, checks it reads well, and revises.
Example
Given “add dark mode,” the agent finds the relevant files, makes the edits across them, runs the build, and fixes what breaks.
How it actually works
Agentic coding puts a model inside a real dev loop with tools: read the repo, edit, run tests, read the failures, try again. Tight feedback from the compiler and tests is what makes it work — the environment tells the model when it’s wrong, so it can self-correct. It shines on well-scoped, testable changes and struggles on ambiguous, cross-cutting design where there’s no quick signal.
For product teams
The shift from autocomplete to a teammate who can take a ticket end to end.
For engineers
A coding agent loop combining file edits, test execution, and error feedback for iterative self-correction.
Related
- Copilot — The autocomplete-era predecessor it grew out of.
- SWE Agent — A named class of these systems.
- Code Interpreter — Running code to check the work.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome