Coder does not do everything in one process. A run fans out across specialized agents, each with a clear role.
Agent roles
Runs can use roles such as analyst, backend, frontend, infrastructure, QA, security, docs, and UX. Coder assigns tasks to the roles that fit the work, so a change gets the right kind of attention.
How agents run
- Each agent works in an isolated workspace tied to the run.
- Agents hand results back to the shared run record.
- Logs, ownership, and validation stay visible across agents.
- The queue coordinates ordering so dependent tasks wait for their inputs.
Local, cloud, and hybrid
Agents run locally, in the cloud, or in a hybrid mix. The runner changes; the orchestration, logs, and audit trail do not. See [local, cloud, and hybrid runs](/docs/coder-local-cloud-hybrid-runs).
Ownership and review
Because ownership is recorded per task, you can see which agent produced which change and review it the same way you would review a teammate's pull request.
Related
- Agents execute the [task queue](/docs/coder-task-queues).
- Their output is checked by [validation](/docs/coder-validation).
Was this article helpful?