Agent teams

A single agent handles one task well, but some problems are best solved by several agents working side by side. An agent team is a group of agents launched together against a shared objective, each owning a slice of the work and contributing its result to the whole. Syntic Code coordinates the hand-offs so the team behaves like a small, focused crew rather than a set of disconnected sessions.

Teams are the natural fit for work that is wide rather than deep: reviewing every module of a large service, migrating many files at once, or researching several questions in parallel. Because each member runs in its own context, the team can cover far more ground than one agent could hold in a single window.

Composing a team

Start by breaking the objective into independent parts. Assign each part to an agent — often a custom subagent chosen for its specialty. A code-quality team, for example, might pair a security reviewer, a performance auditor, and a test-coverage checker, each pointed at the same branch but looking for different things.

The key to a healthy team is independence. Members that do not depend on one another’s output can run fully in parallel, and Syntic Code will fan them out at once. When you do need coordination, keep it explicit: name which agent produces what, and which consumes it.

Collaboration and aggregation

Members share results through the orchestrating session. As each agent finishes, its findings return to Amara, which can synthesize them — merging three reviews into one ranked report, or combining research threads into a single summary. You watch the whole team through the agent view, stopping any member that strays and letting the rest continue.

For work with ordering constraints — where one step must finish before another begins — reach for dynamic workflows instead, which are built for sequential and branching pipelines rather than flat parallel teams.