Skip to content
FirstbeamDOCS
DocsWorkflows

Build an agent team for a real task

Roles, current teams, conversation rosters and reusable templates explained.

Firstbeam · 2 min read · 12 October 2026

Choose the current team

Open the team pill in Agent Mode or Agent library in Settings → Agent capabilities. Both lead to the current team editor. Select a saved team or adjust its members before starting work. A team is a reusable configuration; a conversation roster is the set of agents assigned to that conversation.

Make responsibilities explicit

A lead can coordinate scope, an analyst can inspect requirements, a developer can implement and a reviewer can check the result. Use fewer agents when the work is tightly coupled. More agents are useful only when their tasks can be separated without competing edits.

  • Give each member an engine, model and permission level.
  • Use read-only access for investigation when edits are unnecessary.
  • Keep one owner for shared files and integration.

Reuse roles without losing saved work

Saved global role templates remain in their own library. They are not an obsolete copy of your live conversation. Change a template when you want future teams to reuse the role; change a conversation roster when you want to adjust one running task.

Check the handoff

A useful handoff records the files inspected, changes made, checks run and remaining uncertainty. Review the combined Git diff rather than assuming every member finished independently. Provider limits and model availability still come from the selected engine accounts.

A practical team handoff

For a settings change, let an analyst identify persistence and responsive constraints, let a developer own the affected components and let a reviewer examine the final diff. Have the lead integrate the result and report the checks. Do not give multiple editing agents the same file unless you have a deliberate integration plan.

Keep the roster proportional to the work

A one-file adjustment often needs only one implementer and a review. A broad change can benefit from separate investigation, implementation and testing responsibilities. Engine accounts determine the available models, usage limits and costs; saving a large roster does not grant extra provider quota. Check availability before assigning a model to every role.