Building The Government We Need For What's Next

In Atlanta last month, Center for Civic Futures brought together chief AI officers, technology leaders, and policy practitioners from 34 states, territories, and tribal nations at the State AI Officers Summit.

We gathered under the theme, "The government we need for what's next," a prompt to look beyond individual tools and pilots toward the institutions this era will require. An opening panel on what the AI era demands of government set the frame, and the group spent the rest of the two days working through those demands in practice, in candid sessions on procurement, workforce, governance, and service delivery and future ambitions.

The Summit also brought together leaders from frontier AI labs, AI researchers, and government officials from Singapore and Estonia, creating an opportunity to compare how governments around the world are approaching many of the same questions.

This community has built deep relationships with each other over time and together they found a safe space to do collective work and imagine. Here are the common themes from our discussions:

‍

The chatbot era is already giving way to something bigger

Most people still picture AI as a chat interface. You type something in, get a response, and then decide what to do with it.But many of the leaders in Atlanta were already looking beyond that model.

Agentic AI systems can pursue goals, use tools, and take actions across organizational systems. Governments are beginning to experiment with these capabilities now, but the broader govtech field has not fully reckoned with what a shift to agents could mean. As in-depth projects like The Agentic State have highlighted, unlike traditional chat tools, these systems have deep reach into organizations, can be unpredictable, and are surprisingly easy to manipulate without the proper guardrails in place, posing significant governance, privacy, and cybersecurity concerns. 

We are still in the early days of understanding how this tech will show up in government service delivery. A small, but growing number of states are actively exploring the impact of agentic systems on service delivery, governance, and oversight. Others are still taking a more cautious approach, continuing to explore AI primarily through chatbots and productivity tools. 

While variations in technical capacity, risk tolerance and program maturity are to be expected, it is critical that states leaders have the support they need to build a baseline understanding of these capabilities, which will in turn make them better buyers (they are already being inundated with agentic pitches). 

No state should have to figure out agents alone. The experiments happening now will produce hard-won lessons about what works and what breaks, and the field needs a trusted place to capture and share them, so later adopters start from experience rather than from scratch. Center for Civic Futures intends to continue supporting states to share these insights. 

States have more leverage than they think

A “procurement escape room” breakout session run by the Open Contracting Partnership, participants were asked to review a deliberately terrible AI contract, and to identify what was broken. It had no data ownership provisions, no performance benchmarks, automatic multi-year renewals, and left usage analytics to the vendor. Several participants recognized it immediately because they lived it. 

The “aha” from this session is that many leaders assume they have little leverage when negotiating with AI vendors. In practice, most of the problematic provisions participants identified could be avoided with the right frameworks in mind, even in resource and time-strapped environments. States can learn more in OCP’s Hacks for Better AI Procurement guide.

The harder challenge is that government procurement timelines move much slower than the AI market itself. Contracts that were signed two years ago often don’t require vendors to disclose any AI features that have been added since. The product a state is running today may be meaningfully different from what was evaluated and purchased.

Many problematic contract provisions are negotiable, and the states that know where to push and what questions to ask are in a stronger position than they may realize.

‍

We need more people who are fluent in both AI and government

For many years, changemakers in government have remarked that technology isn’t the hard part, it’s bringing people along for the journey. Across our two days together, one need kept surfacing: people who can translate between programs and the technology meant to serve them.

As AI capabilities continue to advance, the government's ability to benefit from them will depend on whether institutions can absorb the technology and adapt their ways of working. That, in turn, requires people who are fluent in AI while also understanding what its capabilities mean for the caseworker, the benefits applicant, and the agency director. Government staff, with their expertise on complex processes, rules, and insight into the experience of users and workers will continue to be critical moving forward.

And while AI Advisors felt some needed to be fluent in AI, many felt baseline literacy was critical for all workers. Workers needed to understand not only how to use AI but what information was critical for working with AI well: state employees using AI tools without enough grounding to know when outputs are wrong or technology teams optimizing for efficiency without enough knowledge of the programs they’re supposed to improve. Several jurisdictions described training every public officer to a baseline AI literacy standard. 

While technical literacy will continue to be important as AI becomes a more routine part of government work, its professional judgment and discernment may matter even more. Public servants will need to know when to trust, what to question, and where human expertise must remain central to decision-making.
‍

We need a shared definition of governance

“Governance” came up in most sessions, and we discovered that it meant something different to everyone. The challenge of AI governance in government has become at least four separate problems: 

• Deployment & procurement — how we decide which technology to build or buy, when to deploy it, and how we manage it once it's live.

• Security — how we protect these systems against increasingly sophisticated bad actors and threats that could disrupt government services and erode public trust.

• Data privacy — how we keep residents' sensitive and identifiable data safe and private.‍

• Responsible use — how we ensure AI-assisted decisions are fair, explainable, and accountable to the residents they affect.

State AI teams are putting most of their resources toward the IT and operational concerns, because it generates the most immediate friction. But that focus can leave less time and capacity for broader conversations about resident protections, long-term accountability, and public oversight.

As every state grapples with these issues and the capacity they have to engage with them effectively, there’s a real need for more cross-state learning, sharing, and support. This allows the best ideas to surface and prevents unnecessary duplication across the field. Building that exchange is the work Center for Civic Futures has taken on.

The states that are moving fastest are the ones leaning on peer networks and shared learning rather than going it alone. The field needs more common infrastructure, including shared benchmarks, model policies, and transparent reporting on what is working and what is not.

The challenge moving forward may be less about creating a single governance framework and more about creating shared language for the different problems governance is trying to solve.

Looking ahead

As we reflect on the past two years of growing and supporting this community and look ahead to what’s next, it feels like the right time to move beyond disconnected pilots to real institutional evolution as this tech continues to advance. 

AI is expanding what government can realistically deliver. Realizing that potential will depend less on the technology itself and more on the government's ability to adapt alongside it.

The states making the fastest progress are closely collaborating and sharing notes with peers. They are borrowing ideas, testing approaches together, and building on one another's work. That spirit of shared learning may ultimately become one of government's greatest advantages as AI continues to reshape the work ahead.

If there are resources, frameworks, or practical tools that would help your team navigate these challenges, we'd love to hear from you. Send your ideas to info@centerforcivicfutures.org.

‍