.png)
.jpg)
August 5, 2025
3 Takeaways for Government Leaders Right Now
When we opened our Public Benefit Innovation Fund (PBIF) call for proposals this summer, we knew it was a pivotal moment for government. New federal requirements were colliding with a growing enthusiasm among states for AI experimentation—creating both pressure and possibility, and a window to turn ideas into action.
What we didn’t know was how ready the field would be. The response was immediate and overwhelming, presenting some difficulty in narrowing to just a few selections: more than 400 pilot proposals from across 45 states, all focused on helping public benefits work better for the people they serve.
Taken together, these pilot submissions offered a rare view into how benefits leaders are thinking about AI right now. Here’s what we saw:
1. This moment is ripe for government-driven experimentation with AI in public benefits.
Across these pilot submissions, one thing was clear: public benefit agencies are working in an environment ripe for experimentation. The passage of H.R. 1 has increased scrutiny on states, particularly around work verification and error rates. With federal funding tied to performance, agencies are looking for tools that can reduce mistakes without adding burdens for caseworkers or beneficiaries.
Given that landscape, it’s not surprising that interest in AI has surged. Many agencies are looking for new tools to help them manage tighter requirements, rising caseloads, and limited staff capacity.
Layered onto this is a quieter shift. Federal downsizing has moved experienced technical and policy talent into states, nonprofits, start-ups — bringing that expertise closer to the front lines of benefits delivery.
Together, these forces are creating a moment uniquely suited for cross-sector collaboration.
Alongside 18 state, local, and tribal government agencies that applied to the PBIF Summer Open Call, PBIF’s open call drew applications from over 120 startups, 92 non-profits, 43 established companies, and 19 academic institutions. Around 60% of applications were submitted with letters of support from government or community implementation partners, indicating growing cross-industry momentum.
Notably, of the government agencies that applied for PBIF grants, eight came from counties or municipalities, indicating the emergence of local governments as laboratories for new technologies.
2. AI can help government do more to address deeply consequential backend challenges.
In this open call, most focused on tools that help people move through the benefit approval process, such as chatbots, document uploaders, eligibility screeners, and application preparation tools. In fact, about 72% of proposals submitted by both government and private sector included some sort of beneficiary and caseworker front-end AI tools.
It’s easy to see why they’re popular: these products can make it easier for people to understand what they qualify for, cut down on application time and errors, and build confidence that AI can support, rather than complicate, service delivery.
While these tools are doing important work at the front door, there’s still untapped potential in the back-end data agencies already hold. Many agencies already have rich data in-house — from tax records to enrollment histories — that could do even more to improve workflows, reduce administrative burdens, predict risks and trends, and ultimately get benefits to eligible people faster.
For example, AI could enhance ex parte renewals or to help identify people who are exempt from work requirements, reducing unnecessary verification steps.
The next opportunity is to use back-end data alongside these tools. That way, AI doesn’t just remove barriers — it also helps build more proactive systems.
3. Getting AI right means getting systems right.
PBIF proposals show real promise to help caseworkers and the people they serve. But whether that promise holds will come down to how these tools are put to work in real systems. In reviewing these ideas, we saw that some of the biggest open questions sit not in what gets built, but in how AI is woven into day-to-day service delivery.
Many proposals referenced a “human-in-the-loop” approach, but the details often stopped there. Determining where in the loop to insert a human and which human should be inserted can make or break an AI implementation.
Adding humans at the wrong stage or without the right triage can increase caseworker workload. If the right person isn’t looped in, they may not be able to catch when an AI tool is wrong. Relying on beneficiaries to catch inaccuracies risks errors not getting flagged at all.
We also saw open questions about how AI tools will fit into the work itself. Caseworkers already juggle multiple systems, screens, and logins. Without thoughtful workflow design, new AI tools risk becoming unused add-ons.
Whether agencies are layering on a single AI tool or rethinking entire workflows, they will need time and support for monitoring system errors, training staff, and iterating on what they learn. Involving caseworkers in testing and research can help, but many impacts only become clear once the AI tools are tested in a production-like environment.
What’s Next
First, a big thank you to all the organizations and teams who submitted pilot proposals and engaged in this process. We’re emerging from this PBIF call for proposals reaffirmed that there are both big questions and big possibilities ahead for how AI shows up in public systems.
PBIF is one way Center for Civic Futures is leaning into that work—by backing teams that are testing concrete AI solutions in real service settings and helping government leaders sort out what’s useful, what’s responsible, and what genuinely improves how people access benefits.
In the coming weeks, we’ll announce the first cohort of PBIF awardees and share more about what they’ll be building with their government and community partners. As that work moves forward, we’ll keep surfacing what we learn and inviting others into the conversation, so we can chart the path forward together.
---
Amy Ashida is the Program Manager for the Public Benefit Innovation Fund at the Center for Civic Futures. Specialized in public sector UX and service design, Ashida served as the Director of the Technology Transformation Service's Public Benefits Studio and a Principal Consultant for 18F, a federal digital services agency, providing UX research, coaching, and product support to numerous agencies including Centers for Medicare and Medicaid Services, the states of Vermont, Washington, and Alaska, and the National Science Foundation.
