# Handoffs, Agents as Tools and Fixed Sequences — OpenAI Agent Builder Workflows

Source: https://www.skillbyai.com/en/openai-agent-builder/m-compose

> Three ways to combine agents.

## Who stays in control?

With a **handoff**, control passes to the specialist, which continues the conversation. With **agents as tools**, a manager agent calls a specialist like a function, receives its result and stays in control, which suits gathering several results into one answer. With a **fixed sequence** (canvas edges or code), you decide the order deterministically: extract, then check, then write. Prefer fixed sequences when the order is always the same; use model-driven routing only where the choice genuinely depends on the content.

## Choosing a composition style

Match the pattern to the control you need.

```text
pattern            control after the step     good for
fixed sequence     your graph                 always-same pipelines (extract -> validate -> reply)
handoff            the specialist agent       conversations that belong to one domain
agent as tool      the calling (manager) agent combining several specialists' results
```

## Default to deterministic

Every model-driven routing decision is a place where the flow can go wrong; use it only where content decides the path.

**Quiz:** When an agent is used as a tool, who stays in control?

- [x] The calling (manager) agent
- [ ] The specialist permanently
- [ ] The end user
- [ ] Nobody

*Answer:* The calling (manager) agent. Agents-as-tools return results to the caller.