Glossary · The machinery
Coding agent harness (harness)
The specific CLI an agent runs inside, such as Claude Code or Codex, the thing CodeHerder actually launches on a device.
What it is
A harness is the coding-agent program itself, the CLI a developer would normally install and run by hand: Claude Code, Codex, Cursor, OpenCode, or Pi. Each one reads a prompt and edits files. Commands run a little differently from harness to harness, but the basic job is the same: turning instructions into code. Think of it as the difference between a driver and the car: the harness is the vehicle doing the actual work, and something else has to decide which trip it's making next.
Why it matters to you
If your team already has a favorite coding-agent CLI, or a specific subscription tied to one, the harness is the thing you don't want to be forced to replace. Choosing where to run agent work should turn on coordination and visibility. It shouldn't force your developers to switch away from an agent they already like. It also means a team isn't locked into one vendor's roadmap: if a harness falls behind or a better one ships, swapping it out shouldn't mean rebuilding how the whole team tracks its work.
How it works in CodeHerder
CodeHerder launches five harnesses today, and a task's stage config decides which one runs it. Nothing about the workflow, the review gate, or the cost tracking changes based on which harness a stage uses. A team can mix them across stages: code with one harness, review with another. The harnesses doc lays out what each one supports and where they differ, plus how CodeHerder keeps that vocabulary consistent across all five.
What people get wrong
The harness and the orchestrator get confused a lot, since both terms end up used loosely as "the AI." They're two different layers: the harness is the thing that actually edits your code. The orchestrator decides when it runs and on what; it also decides what happens to the result. Confusing the two makes it harder to reason about a problem: a bad diff is a harness issue, a task stuck with nobody assigned is an orchestration issue, and they get fixed in different places.
Keep going
Related terms
AI coding agent
Software that reads a task written in plain language and produces working code changes on its own.
Agent orchestration
Running more than one coding agent against a shared, tracked backlog, instead of leaving it to whoever opened a terminal.
Agent session
One running instance of a harness working a task on a device, the actual process you can watch in real time.
Capability routing
Matching a task to an agent that's actually tagged for that kind of work, instead of the first agent that happens to be free.
Round up your herd.
Bring every human and every agent onto one table. Watch the work move. Costs update as it happens.