Fleet 1.13:Teams are now shipping 5x more PRs with autonomous pipelines.See what's new →
FleetFleet
Glossary

Agent Role

An agent role is a defined responsibility boundary assigned to an AI agent — such as developer, reviewer, QA engineer, or release manager — that scopes its permitted actions, determines which events it responds to, and shapes the prompt it receives.

Roles serve the same function in an AI agent team as they do in a human team: they clarify ownership, prevent duplication, and create natural handoff points. A developer agent writes code and opens pull requests; a reviewer agent reads and evaluates pull requests. Without role separation, agents may duplicate each other's work or, worse, review code they wrote themselves — eliminating the independence that makes review valuable.

Roles also determine access scope. A developer agent needs write access to feature branches but should not have access to merge protected branches. A release manager agent needs merge access but should not be writing implementation code. Enforcing roles at the access level, not just the prompt level, provides defense-in-depth.

Role definition is part of prompt design: the agent's system prompt specifies its identity, its responsibilities, and what it does not do. Well-defined roles produce more consistent agent behavior because the agent has clear guidance on what falls within its scope and can decline or escalate tasks that do not.

How this relates to Fleet

Fleet's configuration schema has a first-class role field for each agent. Roles determine: which skill the agent runs when started (the handbook includes a deterministic role-to-skill mapping), which fabric event types the agent subscribes to, and how the agent is displayed in status and log views. Fleet ships over 120 role templates covering common engineering and product team structures.

Frequently asked questions

Can one agent play multiple roles?

Technically yes, but it is generally inadvisable. A single agent playing both developer and reviewer is equivalent to self-review — the independence benefit disappears. In resource-constrained setups where running separate agents is too costly, a single agent can switch roles explicitly between tasks, but the role separation should be clearly documented and the limitations acknowledged.

How granular should agent roles be?

Roles should map to recognizable job functions that have distinct outputs and review points. Developer, reviewer, QA, and release manager are well-established. Splitting 'developer' into 'frontend developer' and 'backend developer' makes sense when the codebase has clear boundaries. Avoid roles so narrow that every task crosses multiple role boundaries — that creates coordination overhead without corresponding benefit.

Run your first agent fleet

One binary. Five minutes. See every agent, coordinate every handoff, and keep a full audit trail of what your fleet did.