What is an agent handoff?
An agent handoff is one AI agent delegating a task to another agent better suited to handle it, along with the context needed to do the work. The delegating agent either waits for the result or, for longer tasks, registers a way to be notified and moves on. Handoffs are the basic unit of multi-agent systems.
Two kinds of handoff
Inside a single framework, a handoff is often just a function call: a triage agent in an orchestration library routes the conversation to a billing agent in the same process. This is the version most agent frameworks mean by "handoff," and it's easy because both agents share memory and a runtime.
The harder and more interesting version crosses a boundary: different processes, teams, or companies. Here the handoff becomes agent to agent communication over the network, and the A2A protocol gives it structure. The delegating agent finds the specialist through its agent card, sends the task, and gets a task ID it can track through the lifecycle: submitted, working, then completed, failed, or canceled.
What makes handoffs hard
A handoff isn't fire-and-forget. The delegating agent needs the result, and agent tasks run long and unpredictably. Holding a connection for an hour-long task doesn't work, so real handoffs are asynchronous: the delegator registers a webhook URL and the specialist pushes status updates and results as they happen. A2A can also pause a task in an input-required state, turning the handoff into a conversation where the specialist asks for clarification before continuing.
That async return path is where handoffs fail in practice. If the result notification is dropped and there are no retries, the delegating agent waits on a task that finished hours ago. Reliable asynchronous delivery, with retries and delivery visibility, is what makes a handoff trustworthy end to end, and it's the layer [Svix]https://www.svix.com/?utm_source=resources&utm_medium=content) provides.
Frequently asked questions
What is the difference between an agent handoff and a tool call?
A tool call is synchronous and dumb: the agent invokes a capability and gets an immediate answer. A handoff transfers ownership of a task to an autonomous peer that works over time, can report progress, ask for input, and returns results later, often over a push notification.
How does an agent know the handoff succeeded?
Through the task lifecycle. In A2A the delegating agent receives a task ID and tracks state changes by polling, streaming, or push notifications to a registered webhook. The task reaching a completed state with its output artifacts is the success signal.
What happens if the agent receiving the handoff goes down?
The task stalls or fails, so the delegating side needs timeouts and retry policies. The reverse direction also matters: if the delegator is down when results are pushed, the notification needs retries with backoff, or the completed work is lost.
Ready to send webhooks?
Svix handles signing, retries, rate limiting, and delivery observability for the webhooks you send to your users, so your team can stay focused on your product.
Start sending webhooks with Svix or read the build vs. buy analysis