People are already using AI as an unofficial interface to government, and Google’s new ATLAS report, based on nearly 15 million Gemini interactions, gives us a sense of the scale:
-
The report notices that people are increasingly using AI for “high-friction” bureaucratic tasks
-
The category “Government Services and Civic Obligations” is over-represented in AI conversations by a factor of almost twenty (!) compared to the actual time Americans report spending on these activities in the American Time Use Survey (ATUS)
-
Analysis of the “Government and Civic Tasks” domain reveals that 40.1% of these interactions involve obtaining licenses or paying taxes, fines, and fees
-
The report notes that nearly half (specifically around 50%) of these government-related consultations occur outside standard 9-to-5 business hours, allowing users to bypass the temporal barriers associated with accessing these services
Most discussion of the The Agentic State has focused on governments deploying their own assistants. That is useful, but the more consequential shift is the emergence of agents that citizens bring with them.
That makes agentic operability both the biggest opportunity and the hardest challenge for AI in the public sector: enabling third-party agents to discover, understand and transact with government services under a verifiable mandate.
Demand for this kind of intermediation is already large and growing, visible now in the traffic logs of one of the world’s biggest AI products. The supply side, the part governments actually control, has barely begun.