Government agents went into production while everyone was watching DeployCo

verified

Monday, May 11, 2026

The News

PYMNTS reported Monday that in its first week live, an AI agent called Bobbi resolved 82% of citizen queries across three police forces in the United Kingdom without once passing the call to a human . Separately, Estonia is piloting a network of interoperable agents that cross agency lines so that a citizen renewing a passport works with one system that coordinates everything on their behalf . Google Cloud Next '26 announcements included the Gemini Enterprise Agent Platform designed to connect data, people and operational goals across large institutions, and Gemma 4, Google's latest open model that gives agencies a deployable option they can customize on their own terms . Source: PYMNTS.

The Read

An 82% deflection rate at three UK police forces is the kind of number that, if it holds for a quarter, ends a debate that has run for two years. Government call centers are the hardest-possible enterprise deployment — multilingual, legally sensitive, politically scrutinized, with users who cannot opt out — and they are now the first place agents are clearing 80%+ deflection in production. The implication for private-sector deployments is uncomfortable: your customer service team's "AI isn't ready for our context" argument just got measurably weaker, because almost nothing in your context is harder than a 999 call.

Do This Week

At your next risk-and-controls meeting, ask the person who owns your customer-facing chat AI to name a single deflection-rate benchmark from a comparably regulated peer. If the answer is "we don't have one," put Bobbi on the agenda for the following meeting. The question isn't whether to deploy — it's whether you can articulate why your context is harder than a police non-emergency line.

For Legal ops & GRC