RADAR: Measuring How AI and Agents Reach Government Services

One of the central ideas of the Agentic State is a new interface between citizens and government, made of AI agents that governments do not run. Yet since the publication of the vision paper, most of the attention has gone to a more comfortable version of that idea: the state building agents of its own. Much less attention has gone to whether public services can be operated by the agents citizens already bring with them. That condition had never been measured at scale.

Tiago C. Peixoto, co-author of The Agentic State vision paper, joined forces with Luke Jordan and Manuel Ramos-Maqueda at the World Bank to change that, with contributions from The Agentic State team and many expert colleagues.

RADAR (Readiness for AI Discovery and Agentic Reach) measures how AI and agents can reach government services. Across 166 countries, it runs live model queries and autonomous agent attempts against real government services, scoring whether a service can be located, whether the information returned is country-specific and traceable to an official source, and whether an agent can get far enough to file an actual request. The results are somewhat counterintuitive: several governments that rank well on standard digital-government indices fare less well here, and many do better than those same rankings would suggest.

The paper closes on what it calls sovereign legibility, the deliberate work of making authoritative government content readable and actionable by AI systems. Many of the recommendations are low cost and no regret, and the last section sets out where the research goes next.

Read the preprint: RADAR: Readiness for AI Discovery and Agentic Reach.

We’re curious to hear your thoughts — write to us.

Download the preprint