While DOGE used ChatGPT to torch federal grants with a DEI prompt, a free platform called GrantWell is doing the opposite: helping communities actually access the $1 trillion Congress already allocated.
The Summary
- DOGE staffers used a single ChatGPT prompt to flag NEH grants as "DEI", cutting the agency's budget in half and eliminating projects from Holocaust documentaries to American music scholarship
- Congress appropriates over $1 trillion annually in federal grants, but most communities lack staff who know how to apply (Massachusetts alone leaves 70% of $17.5 billion eligible funding unclaimed)
- GrantWell, a free AI platform, helps under-resourced communities actually access this money, showing what purpose-built public-interest AI looks like versus ideological automation
The Signal
The contrast here is not about AI being good or bad. It's about who builds it, for what purpose, and with whose input. DOGE's approach was pure algorithmic reductionism: feed ChatGPT a politically loaded binary question, collect yes/no answers, cut everything that trips the wire. No context. No domain expertise. No consideration of congressional intent or community impact.
The real damage is not just what got cut. It's the precedent: using general-purpose LLMs as ideological sorting machines for complex policy decisions. A model trained on internet text has no framework for evaluating whether a music history project serves the public good. It only pattern-matches words to other words. When you ask a bad question to a tool that cannot say "this question is bad," you get bad answers at scale.
"They used artificial intelligence as an ideological wrecking ball."
Now look at the problem GrantWell is solving. Over $1 trillion in federal grants gets appropriated every year. This is not speculative funding or new programs. Congress already decided this money should go to states and cities for infrastructure, schools, clean water, broadband. But 70% of eligible funding in Massachusetts goes unclaimed because small towns and under-resourced communities do not have grant writers on staff.
This is not a funding problem. It's a translation problem:
- Federal grant applications require specialized knowledge of compliance language, reporting structures, and bureaucratic formatting
- Small municipalities cannot afford full-time staff who speak this language
- The money sits unused while the bridges still need repair
GrantWell uses AI to close that gap. It is purpose-built for this specific workflow. It knows grant taxonomy, application requirements, compliance frameworks. It helps communities identify relevant funding, draft competitive applications, and navigate reporting requirements. It does the translation work that currently requires expensive consultants or simply does not happen.
The difference is design intent. GrantWell was built with input from the people who actually write grants and the communities trying to access them. It solves a real coordination failure in how public money flows. DOGE's ChatGPT prompt was built to confirm a political conclusion, regardless of accuracy or consequence.
The Implication
This matters for anyone building AI tools for government or public systems. The question is not whether to use AI in the public sector. The question is whether you build tools that help institutions function better or tools that automate bad decision-making.
If you are building agents for government workflows, study GrantWell's approach. Narrow the problem. Involve the actual users. Build for the coordination failure, not the political theater. The real opportunity in government AI is not replacing human judgment with algorithms. It is helping under-resourced humans access systems that were built to serve them but require specialized knowledge to navigate.