What good AI in government actually looks like

Last spring, when employees from Elon Musk’s Department of Government Efficiency arrived at the National Endowment for the Humanities, they used ChatGPT to decide which federally funded grants to eliminate. 

They typed one, simplistic prompt: “Does the following relate at all to D.E.I.? Respond factually in less than 120 characters. Begin with ‘Yes’ or ‘No.’”

A documentary about Jewish women forced into slave labor during the Holocaust? DEI. A 40-volume scholarly series on the history of American music? DEI. 

Within two weeks, DOGE staff had flagged nearly every active award from the Biden administration and cut the National Endowment for the Humanities budget by half. 

This search-and-destroy approach uses artificial intelligence as an ideological wrecking ball. 

But we can use artificial intelligence differently. Instead of stripping communities of federal funding, we can use AI to help them access the money Congress appropriated, fulfilling congressional intent and strengthening our democratic institutions.

A free platform called GrantWell does exactly that, and shows the way we should be building AI with communities in the public interest. 

What are federal grants?

To understand what’s at stake, it helps to understand what federal grants actually include. Every year, Congress appropriates more than $1 trillion in tax dollars in the form of grants and directs them back to states, cities, towns, and communities across the country to invest in replacing lead pipes before more children are poisoned, repairing bridges before they fail, expanding broadband to rural areas the market has never reached, putting clean school buses on the road, and modernizing an electrical grid that is decades behind.

The challenge of obtaining those grants often results from a lack of staff knowledgeable about how to apply. Massachusetts alone estimates that their communities are eligible for approximately $17.5 billion in federal grants, yet less than 30% of that money is successfully accessed by local governments, according to a state study. The money is technically available. Congress specifically reserved it for this exact purpose of ensuring that local governments can improve services for its constituents. But finding it, understanding it, and applying for it has become so complex that smaller communities are effectively locked out. 

AI for Impact: Community-centered AI in action

I run a program called AI for Impact, where graduate, undergrad, and community college students are paid to take time off—think of it like a semester abroad, only in Boston and in New Jersey—to build AI-enabled products for and with government partners. 

Anjith Prakash and Jai Surya Kode took time out of their graduate studies in AI at Northeastern University to create the first version of GrantWell, working in partnership with Massachusetts’ Federal Funds and Infrastructure Office. 

We built GrantWell by listening to the communities most in need of these grants. Anjith and Jai joined the team from the Massachusetts Federal Funds and Infrastructure Office and traveled across the state, meeting with local officials, tribal representatives, and community organizations. 

They heard the same story from Haverhill to Barnstable: officials spending hours reading 100-page grant announcements only to discover that their town wasn’t even eligible. Deadlines missed. The applications never started. Not because communities didn’t need the money, but because they lacked the capacity to pursue it. Wealthy municipalities had consultants and dedicated grant writers. In smaller towns, one person did five jobs. 

When we started the project, we assumed that people would want to use AI to help with drafting. But the challenges, the team learned, started much earlier: Which grants exist? Am I even eligible? What does this requirement mean? For many smaller communities, answering those questions alone was enough to stop an application before it began.

Community-centered AI means building with people, not just for them. It treats communities as co-creators, drawing on the expertise of families, caregivers, residents, and frontline workers. Those people are closest to the problems these tools aim to solve, whether the product is a grant-writing app or one that helps navigate government benefits. It starts by asking what people actually need, not what AI can do, a shift that often reveals different problems than designers initially assumed and, in our case, led to entirely new features for GrantWell.

What GrantWell does

So we designed GrantWell around their needs. GrantWell now aggregates federal and state funding opportunities into a single searchable database, generates plain-language summaries of dense government documents, and includes a chatbot that answers questions directly from the source text to reduce the risk of confabulation. 

A structured drafting workflow produces a first draft that human grant writers can refine and own. The tool is deliberately modest about what AI should do. It doesn’t submit applications. It gets people to the starting line, supporting them to do the work better for themselves.

In March, the Governor of Massachusetts launched GrantWell as a free AI-powered platform on mass.gov here. To help towns and civic groups trust the tool is free, secure, and will not misuse their information, we’ve integrated it into state infrastructure and made it freely available to all, thanks to this collaboration between a state and one of its leading universities.

From one to many

Building on our collaboration around AI and innovation training, California, Colorado, and Rhode Island are exploring adopting the tool next, with several other states waiting in the wings. In considering GrantWell, leadership from the Colorado Governor’s Office emphasized the value of reducing barriers to entry for communities seeking grant funding to meet local needs, which will help their Federal Funds office better scale its impact across the state.

There is no reason taxpayers in these states should start over, procuring and rebuilding the same capabilities that all states need, wasting limited funding, leading to wildly uneven quality and widening the gap between well-resourced governments and under-resourced ones.

As we expand the GrantWell model to other states, AI for Impact Fellows are not simply replicating the Massachusetts version. We’re building collaboratively, convening states regularly to share what’s working and what isn’t, and going directly to the communities we’re trying to serve with questions like: What tools do you already use? What challenges do you face when applying for grants? What do you need the tool to do? 

AI for Impact is also working with multiple states to share, refine, and deploy a procurement drafting tool, a contract review tool and partnering with cities to roll out the tool we built with the City of Boston to help residents explore and use open municipal data.

AI is increasingly embedded in everyday life, especially in public systems like education, benefits, housing, and healthcare. Yet most AI tools are still designed far from the lived realities of the people who depend on these systems. The result is predictable: systems that fail to reflect real needs, tools that are technically impressive but practically unusable, and communities treated as subjects of technology rather than participants in shaping it. Trust erodes, particularly among groups historically excluded from decision-making.

Federal grants are taxpayer dollars that Congress directed back into our communities for lead pipe removal or clean school buses. We can use AI to help those dollars reach the people they were meant for. When we build AI with those who will use it, we also create better tools, stronger institutions, and more capable, resilient communities.