Twilio launches open-source Agent-2-Human protocol for AI agents

Twilio launches Agent-2-Human protocol

Cloud communications company Twilio has launched an open-source protocol designed to help AI agents communicate with people. The new Agent-2-Human (A2H) protocol, announced Thursday, provides a framework for agents to hand off tasks to humans and manage those interactions across different channels.

It aims to abstract away the complexity of reaching a person via SMS, WhatsApp, voice calls, or push notifications. “The agent focuses on what it needs from the human, not how to reach them,” said Rikki Singh, Twilio’s VP of product and engineering for emerging technology.

Why agents need a human escalation path

Singh argues that human judgment and trust will remain essential, even as AI agents become more autonomous. The core problem A2H tries to solve is creating a reliable, built-in escalation route that doesn’t burden developers.

“It shouldn’t rest on the developer or the consumer, right? It should rest on the tool,” Singh told The New Stack. The protocol uses a gateway to handle the messaging logistics, allowing the agent to simply state its need.

This approach also creates a signed audit trail for every interaction. This provides clear evidence, for instance, when a human approves a transaction an agent has requested.

How the A2H protocol works

The protocol is built around five core, extensible “intents” that define the purpose of an agent’s communication. These intents are based on Twilio’s experience in business-to-consumer interactions.

• Inform: For sending one-way notifications.
• Collect: For gathering structured information, like a shipping address.
• Authorize: For approving transactions, with authentication support.
• Escalate: To hand off a task completely to a human.
• Result: For reporting task completion back to the agent.

For developers using the popular Model Context Protocol (MCP), A2H functions as another available tool. Agents can use standard tool-calling patterns, such as `humans_inform()`, to trigger these communications.

What’s included and what’s next

The initial open-source release includes the protocol specification, the intent framework, authentication support, and the delivery channel abstraction. Twilio plans to expand the ecosystem with several key additions soon.

Future integrations will include more agent frameworks like LangGraph and CrewAI. The company also plans to add primitives for creating and canceling standing approvals, which would allow for predefined permissions for certain agent actions.

The launch addresses a gap in the growing landscape of agent-focused protocols, which already includes options for agent-to-agent (A2A) and agent-to-commerce (ACP, AP2) communication, but lacked a standardized human handoff system.