Definition
Plain language
When an AI assistant doesn't just suggest code but actually edits files, runs commands, and works on a task autonomously.
As stated in the literature
A class of LLM-based software-engineering systems that interleave reasoning with tool use over a codebase, typically running multi-turn agent loops with shell, filesystem, and test-execution access.
Also called: agentic coding agents
Why it matters: This is the shift from 'autocomplete' to 'junior teammate' that's reshaping how software actually gets written.
For example, an agentic coding assistant might read an open bug report, edit the relevant file, run the test suite, and open a pull request without further input.
Heard on the show
“There's also the framing of "long-horizon agentic coding.”Episode 003 — How to Pick the Best of Sixteen Coding Agent Rollouts