The idea
Instead of one agent with 40 tools and a giant prompt, split responsibilities: a triage agent talks to the user and hands off to a billing agent, a technical agent or a sales agent, each with its own instructions and a small tool set.
A handoff is usually implemented as a tool: transfer_to_billing(reason). Your code switches the active agent and passes along the conversation (or a summary of it). The new agent continues from there.
Keep the number of agents small and the boundaries clear. Every handoff is a chance to lose context or bounce the user around.
Specialised agents are easier to prompt, test and secure than one agent that does everything.
Picture it
Words to know
- Triage agentThe front agent that routes to specialists.
Try it without code
Map your specialists
- For a support bot, list 3 specialist agents.
- Write each one’s tools and one rule only it follows.
Use it in code
Handoffs with the OpenAI Agents SDK.
# pip install openai-agents
from agents import Agent, Runner
billing = Agent(name="Billing", instructions="Handle refunds and invoices. Never promise refunds over 100 EUR.")
tech = Agent(name="Tech support", instructions="Debug login and API errors step by step.")
triage = Agent(name="Triage",
instructions="Work out what the user needs and hand off to the right agent.",
handoffs=[billing, tech])
result = Runner.run_sync(triage, "I was charged twice for March.")
print(result.last_agent.name, "->", result.final_output)Build triage + 2 specialists and test 10 messages. Count wrong handoffs.
Common mistakes
- Specialists that overlap, causing ping-pong.
- Losing key facts during handoff.