Skip to content
BENIBIX Learn

Parallel Agents

AI🤖 AI AgentsAgent Orchestration

Also known as Parallelization (AI Workflow), Sectioning and Voting, Agent Fan-Out

Many AI agents work at the same time on one job. Then their results are joined.

Think of it like

It is like a class that splits a big poster into parts. Each child draws one part at the same time.

Example

Emma needs a review of a long contract. She splits it into five parts, five agents read one part each at once, and a last step joins their notes.

Why it matters

Many jobs split cleanly into parts. Doing them at once saves time, and voting can make answers more reliable.

Definition

Parallel agents are several model calls or agents that work at the same time, and then their results are joined. There are two main ways to use them. In sectioning, a job is cut into parts and each agent does one part. In voting, several agents do the same job, and the most common answer wins. Both finish sooner than one agent working alone, but they cost more tokens.
The left panel is parallel agents. One job splits three ways, and the answers merge into one.

Common mistake

Splitting work whose parts depend on each other. Each agent then guesses what the others found.

Key takeaways

  1. 1
    Sectioning. Each agent does a different part.
  2. 2
    Voting. Many agents do the same job, and the answers are compared.
  3. 3
    Join at the end. One step merges what came back.

Compare