Agent Team
Run a team of AI specialists in a single workflow step. Pick a preset, describe the goal — the team handles coordination, delegation, and synthesis.
What is an Agent Team?
An Agent Team step replaces the need to wire together multiple AI steps manually. You describe the outcome you want, choose a preset that matches your use case, and the team coordinates internally: specialists gather information, an analyst synthesises it, and the step returns a single structured result.
Agent Teams work on a single task at a time, not a batch of items. For bulk processing across a list, see Batching.
Presets
Each preset is a pre-configured team pattern. Choose the one that fits your task.
Research & Summarize
Specialists gather information from multiple angles. A coordinator synthesises everything into a concise, structured summary. Best for company research, topic briefings, and due diligence.
Analyze & Recommend
Multiple analysts evaluate an opportunity independently, then a coordinator weighs the findings and produces a ranked recommendation. Best for investment decisions, vendor selection, and scoring.
Generate, then Review
A specialist drafts the output, then a panel of reviewers critiques and refines it. Best for content production, code generation, and anything that benefits from a second opinion.
Compare Options
Specialists argue for competing options. A coordinator evaluates the arguments and picks the strongest. Best for A/B decisions, strategy trade-offs, and architecture choices.
Investigate in Parallel
Independent specialists each explore a different angle of the same problem simultaneously, then combine their findings. Best for competitive analysis, risk assessment, and broad research.
Review & Improve
A panel identifies weaknesses in an existing draft or plan. An editor applies the improvements. Best for document polish, proposal refinement, and iterative improvement loops.
Example: Research & Summarize
This step researches a company from its URL and produces a one-page briefing:
{
"id": "research-company",
"type": "agentOrchestrator",
"name": "Agent Team",
"orchestratorConfig": {
"pattern": "supervisor",
"workers": [
{
"id": "researcher",
"name": "Researcher",
"systemPrompt": "Research {{input.company_url}} — team size, funding, products, market"
},
{
"id": "analyst",
"name": "Analyst",
"systemPrompt": "Analyse the research. Identify growth signals and risks."
}
]
},
"metadata": {
"agentTeamPreset": "research-and-summarize",
"agentTeamMode": "simple",
"agentTeamUxVersion": 1
},
"next": { "stepId": "milestone" }
}The metadata block controls how the step opens in the builder. Setting agentTeamMode: "simple" keeps the preset UI visible. Without it the step opens in advanced mode.
Advanced Configuration
The preset UI covers most cases. When you need full control, switch to advanced mode in the builder or set agentTeamMode: "advanced" in the step metadata. Advanced mode exposes:
- •Pattern — coordinator-led (supervisor), parallel, or specialist review (debate)
- •Custom worker roles — define any number of specialists with custom prompts
- •Model selection — choose different models per worker
- •Effort level — controls how long workers deliberate before producing output
Existing steps built with raw orchestratorConfig open in advanced mode without any changes needed.
Execution View
Each run shows a summary card with team size, coordination pattern, total steps, credits used, and duration. Expand any specialist's output to see exactly what they produced, then read the team conclusion synthesised by the coordinator.
The step label in the execution timeline reads Agent Team regardless of the underlying pattern, so legacy workflows render correctly without any migration.
Next Steps
- Agents as Step — Single-agent steps for lighter tasks
- AI Steps — Standard AI steps with prompts and structured output
- Batching — Fan-out processing for bulk lists
- Human-in-the-Loop — Add approval gates after an Agent Team produces a draft