Microsoft's real-time voice agents, and why barge-in is table stakes

On 28 April 2026, CX Today reported Microsoft's general availability of real-time voice agents in Copilot Studio, launching first in Dynamics 365 Contact Center in North America. Microsoft's own language is specific: agents optimized for low-latency, interruptible, speech-to-speech conversations with real-time reasoning, plus a path to hand context to a human.

Why interruption changed the demo

A caller who cannot cut in will talk over the agent, then repeat themselves to a person. That double work is how Voice AI quietly raises average handle time. Interruptible speech is the difference between a conversation and a recording that happens to answer.

Microsoft also flagged the rollout bounds: North America first, with more languages, regions, and channels later, including a roadmap toward Teams Phone. If you buy for a multi-country floor, ask for the region list in writing. Do not assume a North America GA date is a global one.

A five-minute interruption test

• Call the number and say the account ID while the greeting is still talking.
• Ask a second question before the first answer ends.
• Give a payment or ID, then ask whether recording paused.
• Demand a human. Confirm the person can see the transcript and the last intent.

Avaga AI employees sit on the same line as your team. The test above is how we judge the receptionist, sales, support, booking, and lead seats: can a person take over without asking the caller to start again?

You do not need Dynamics 365 to steal the lesson. You need a workspace where the AI employee and the human share one thread. If the handoff drops the story, the latency win does not matter.