America needs a national model for solving big problems with AI

When the administration launched America.gov last month, I felt two competing things at once.

On one hand, I welcomed this new tool designed to help people find official information and navigate government services. In the aftermath of DOGE and its impact on the federal workforce, it was heartening to see the administration finally recognize that technology designed around people can and should make government work better.

At the same time, I felt the pang of a lost opportunity. At launch, America.gov is still principally a chatbot. It can help people find the right door into government, but it cannot fix what happens on the other side.

As we shape the next era of government, America needs a national model for using AI to solve big public problems: one that starts with people, brings government together with outside expertise, keeps humans in control, and measures success by what changes in people’s lives. Realizing this vision requires both the willingness to try something new and the commitment to see it through.

The Direct File saga shows why both ambition and continuity matter. The free IRS service gave eligible taxpayers a simple way to file directly with the government. Nearly 300,000 people used it in 2025, and 9 in 10 pilot users rated it excellent or above average.

Government built something useful. People trusted it. Then Treasury shut it down. Now, in something of a twist, former DOGE officials are helping lead America.gov.

Exasperating? Infuriating? Depending on the hour, both.

But families waiting for food assistance, workers seeking unemployment benefits, and older adults trying to keep health coverage cannot wait for Washington to get its story straight. We have another opportunity right in front of us, and we need to get it right.

At Code for America, we’ve spent 15 years working in the space between a policy promise and the person sitting at the kitchen table trying to use it. We know where the seams are: the records do not match, the notice makes no sense, the caseworker is toggling among systems, the handoff never happens.

That experience has taught us what a national model should do. It should give every level of government a shared way to address problems, build responsibly, measure results, and move successful work from pilot to public service.

This work starts with public goals big enough to justify national effort: help eligible families receive benefits, detect disasters sooner, extend medical expertise into rural communities, and connect workers to real opportunities. Once those outcomes are clear, four principles should guide the work.

First, establish common protections that keep people in control. People should know when AI is shaping a public decision, understand the basis for that decision, and have a real way to challenge it. Public servants must be able to question or overrule a tool without being punished for exercising judgment. Every system should be tested for accuracy, bias, privacy, and security before — and after — it reaches the public.

Second, build shared infrastructure and evidence. Federal agencies should develop secure components, data standards, procurement tools, and evaluation methods that state and local governments can reuse and adapt. Results should be public so governments can build on what works and stop repeating what does not.

Third, bring the right partners to the table. Government must set the public goal, establish the rules, and remain accountable. But it needs technologists who understand AI, public servants who understand government systems, and communities that understand how decisions land in real life.

Finally, invest in delivery. Here is where I get impatient. America does not lack AI pilots. It lacks the capacity to connect systems, improve data, redesign workflows, train staff, and maintain a service after launch. No one cuts a ribbon on systems integration. But that is where the promise becomes real.

We learned this in part through our work on GetCalFresh, which reduced California’s food-assistance application from more than an hour to roughly 10 minutes. Now we are building a SNAP Policy Navigator with Anthropic to help caseworkers make sense of complicated policy.

The tools are different, but the standard is the same: Did the work make life better for the person who needed help? Did errors fall? Did public servants gain useful time? Could people challenge a wrong decision? If the dashboard is green but people are worse off, we stop.

My parents began as farmworkers in California’s Central Valley. They taught me that if a decision is going to shape your life, you should have some power over how it gets made. That lesson belongs at the center of how government uses AI.

Six years ago, I called for a government with heart. Heart is understanding how a public decision lands in someone’s life, and having the capacity to do something about it.

America.gov has opened the door. A national model will help us build the next era of government behind it.

Amanda Renteria is CEO of Code for America.