What is Mode 4 in customer service?

The infrastructure is already here

Two AI agents can already transact on behalf of their principals without a human touching the interaction. AI can use a computer the way a person does. Model context protocols let it use tools and data. Google's A2A lets agents from different companies talk to each other directly, and AP2 lets them make payments with cryptographic proof of permission.

What is not in place is a shared expectation of what each party must declare, and what happens when one side arrives without the authority to finish the job.

What goes wrong without structure

Agents over-commit, refuse without a reason, loop indefinitely, or leak data they were not authorised to disclose. The costs are real: token spend, resolution time, customer trust and, increasingly, regulatory exposure.

What it looks like when it works

A customer's AI contacts a retailer's AI about a damaged delivery. Both present a Service Handshake declaration. The customer's side declares the goal (a replacement or refund), the constraints and the data it will share. The retailer's side declares its authority (up to £150 replacement value), its data commitments and its fallback. The declarations align, the replacement is dispatched and confirmation is sent: 4.2 seconds, no human involvement, a full audit trail on both sides.

When the declarations don't align, the standard escalates cleanly. The brand's agent declines, says why, and triggers its fallback: a named person commits to resolve it within a set time, with the order, evidence and the customer's hard rule attached.

How to prepare for Mode 4

Publish your declarations before you ship the agent: what your AI can commit to, what authority it will accept from an incoming agent, what data it can share and where the fallback goes. In the State of Dual CX audit, none of 100 UK brands exposed any machine-readable declaration an incoming agent could act on.

Common questions