Agent Handoff

What is an agent handoff?

An agent handoff transfers active responsibility for a task or conversation from one agent to another. The receiving agent gets the context, state, constraints, and expected next outcome needed to continue without restarting the work.

A handoff is complete only when the receiver accepts responsibility and can act.

How a handoff works

Imagine a general support agent discovering that a request requires account-security expertise. It can hand the conversation to a security agent with:

• the user's verified identity status;
• a concise problem summary;
• actions already attempted;
• relevant tool results;
• unresolved questions;
• permission and escalation limits.

The system should record which agent currently owns the task and whether the transfer succeeded.

Handoff versus delegation

Delegation assigns a bounded subtask while the delegating agent retains responsibility for the larger goal. A handoff transfers active ownership of the task or interaction.

For example, asking a research sub-agent to collect sources is delegation. Moving the full customer conversation to a specialist who now owns resolution is a handoff.

The patterns can combine: an agent may delegate background work before deciding whether to hand off the whole task.

Handoff versus routing

Routing selects a destination based on request type or capability. A handoff is the operational transfer, including context and ownership.

A router may choose the security agent, while the handoff mechanism packages state, obtains acceptance, and updates the conversation owner.

Failure modes

Poor handoffs lose constraints, duplicate completed work, expose unnecessary data, or leave both agents assuming the other is responsible. Large raw transcripts may overwhelm the receiver, while aggressive summaries can omit critical evidence.

Reliable systems use structured handoff contracts, minimum necessary context, source links, explicit acceptance, and audit records. Sensitive transfers must preserve permission boundaries.

Why it matters

Handoffs let specialized agents continue work without forcing one agent to handle every domain. They are central to multi-agent and human-agent collaboration.

The design challenge is continuity. A good handoff preserves verified progress and ownership while avoiding irrelevant context. See [Single-Agent vs Multi-Agent Systems](/single-agent-vs-multi-agent-systems/) for how handoffs fit into collaborative architectures.