A while back I wrote about why production AI needs a session layer, not just a stream. This is the follow-up: a live demo showing what those patterns actually look like in a working app, published on the Ably blog.

The demo is a holiday planning app built on Ably AI Transport, and it walks through the things a raw SSE pipe can’t give you:

  • Session continuity: refresh the page mid-stream and catch up exactly where you left off, without re-running inference
  • Barge-in: interrupt the agent mid-response with a real cancel signal, rather than an ambiguous dropped connection
  • Human handover: a supervisor joins hours later, on a different device, and picks up pending approvals from channel history
  • Multi-agent coordination: specialist agents publish their progress directly, instead of bottlenecking through an orchestrator

You can watch it here:

The full write-up is on the Ably blog: AI agent streaming in action: barge-in, human handover, and session continuity.