Several specialist agents must work together. Which Agent Bricks option fits?
Why A is the answer
Multiagent Supervisor
A Multiagent Supervisor coordinates several specialist agents and decides which one handles each part of a request.
In detail
- Multiagent Supervisor coordinates several agents.
- It decides which specialist handles each part of the work.
- Databricks now calls this Supervisor Agent and labels it legacy. It points new low-code agents to Genie Agents, and multi-agent code to the Agent Bricks CLI.
- Give each specialist a clear description, so the supervisor can route to it.
Use case
An illustrative scenario, not a real incident.
A bank has one agent for policies and another for account data. Multiagent Supervisor routes each question, so a question spanning both gets a joined answer.
See it in different ways
flowchart TD A["User"] --> B["Supervisor agent"] B --> C["Knowledge agent"] B --> D["Genie Space: structured data"]
mindmap
root((Agent Bricks))
Knowledge Assistant
Questions on documents
Information Extraction
Fields from text
Multiagent Supervisor
Coordinates agentssequenceDiagram participant P0 as User participant P1 as Supervisor participant P2 as Policy agent participant P3 as Data agent P0->>P1: Question P1->>P2: Policy part P1->>P3: Data part P2-->>P1: Result P3-->>P1: Result P1-->>P0: Joined answer
Key point: A supervisor coordinates specialist agents.
Why the other options are wrong
- B. Information Extraction, run once for each agent in the group
- Information Extraction pulls fields, and does not coordinate agents.
- C. A single prompt that holds the instructions for every agent
- One prompt merges the specialists instead of letting them work together.
- D. A Vector Search index that stores each agent as a document
- An index stores content, and cannot coordinate anything.