After three years overseeing Vermont’s technology systems, Denise Reilly-Hughes says the challenge isn’t simply keeping up with artificial intelligence. It’s understanding what government needs it to accomplish—and ensuring people remain responsible for the results.
Denise Reilly-Hughes spent more than two decades working in private-sector technology, including at Microsoft, before joining Vermont state government in 2023. On October 16, she concluded her tenure as secretary of the Agency of Digital Services and the state’s chief information officer, returning to the private sector.
In an interview with Compass Vermont on her final day, Reilly-Hughes offered a perspective on artificial intelligence that was less about what the technology might eventually accomplish and more about what government should expect from it.
Technology Is Not the Fix for a Problem
“Technology is not the fix for a problem. Technology is the tool that we use to resolve the problem,” she said.
For an agency responsible for supporting the state’s computer systems, protecting its information, and helping government deliver services, that distinction shapes decisions about what technology to adopt, when to use it, and when not to depend so heavily on it.
Reilly-Hughes said the agency spent much of the past three years trying to change how it approached those decisions. Technology, in her view, should help government fulfill its responsibilities rather than become an objective in itself.
The rapid development of AI has made that distinction increasingly important.
You’re Probably Using AI Without Knowing It
Artificial intelligence can refer to something as familiar as a recommendation algorithm on social media, a system that summarizes documents, or software that supports—or makes—decisions. Those applications involve different capabilities, risks, and levels of human control.
Reilly-Hughes recalled meeting someone who insisted they didn’t use AI. When she asked whether the person used social media, the answer was yes. Her response was that the person was already using AI, even without knowingly choosing to interact with it.
It illustrates a common misunderstanding. AI isn’t one technology doing one thing.
A tool that helps draft a letter is different from one that evaluates a person’s eligibility for a government benefit. And neither is necessarily comparable to a system that can carry out a series of actions independently.
Those distinctions matter when government decides what safeguards are needed—and what risks the public should be concerned about.
Who Knows Enough to Make the Rules?
Reilly-Hughes believes legislators serving on committees responsible for specialized subjects should be expected to develop expertise in those fields, just as they do with energy, transportation, or the environment.
But AI creates a difficult problem. Its capabilities change rapidly, while lawmakers must make decisions about privacy, public spending, accountability, and what government should be permitted to do with the technology.
They cannot be expected to master every development. Yet relying on specialists raises another concern: How does government obtain the expertise it needs without handing oversight to the people and organizations whose work it is regulating?
Industry executives, government technologists, researchers, and advocacy organizations all bring useful knowledge. They may also bring professional interests or deeply held views about the technology.
Expertise doesn’t guarantee impartiality. But independence doesn’t automatically guarantee expertise, either.
The challenge is drawing on both while keeping elected officials responsible for the rules.
Using AI Isn’t the Same as Letting AI Decide
Vermont has established an oversight framework.
Its AI Code of Ethics, updated in June 2026, sets standards for the use of artificial intelligence in state government. The state’s Division of Artificial Intelligence also has statutory responsibilities for reviewing AI systems and their potential effects on Vermonters.
Reilly-Hughes said these safeguards are intended to allow government to use the technology responsibly, with rules governing where and how it is used.
But using AI to organize records is very different from letting it influence decisions about public benefits, legal obligations, or the delivery of government services.
Vermont’s updated AI Code of Ethics requires a human to be accountable for every decision made by a state AI system. It permits narrowly defined autonomous decisions when an error would have only a small, temporary impact, subject to additional safeguards, including review or appeal mechanisms and closer monitoring.
Reilly-Hughes said the state’s approach keeps people accountable for government decisions while allowing AI to assist with appropriate tasks.
Yet human responsibility on paper isn’t necessarily meaningful oversight in practice.
If an employee routinely accepts an AI recommendation without understanding or examining it, a human may technically have the final say while exercising little independent judgment.
That’s why training, disclosure, and the ability to explain and review decisions matter. Government must be able to demonstrate that its safeguards work, not simply point to policies saying they exist.
What Vermont’s Floods Taught Government About Technology
Reilly-Hughes sees AI governance and cybersecurity as closely connected. Both require government to understand and protect the systems and information it depends on.
But she also pointed to a distinctly Vermont lesson from Tropical Storm Irene, more recent flooding, and the pandemic.
What happens when government employees can’t access the technology they normally use? Do they still know how the underlying processes work? Do they understand the rules governing their responsibilities?
Those questions become more important as public services grow increasingly dependent on digital systems.
Technology can make government faster and more adaptable. But it can also create dependencies that become apparent only when something fails.
Reilly-Hughes said those experiences helped shape her agency’s approach to technology investments, including recognizing when new systems would help and when relying too heavily on them could create problems.
The goal, she said, was better government—not simply more technology.
The Test Isn’t How Much AI Vermont Uses
Vermont has established rules for government AI. The harder task is ensuring they remain effective as the technology changes.
That requires knowledgeable employees, informed legislative oversight, transparent decisions, and safeguards that work in practice.
It also requires recognizing what AI cannot do. Technology may help agencies process information, reduce repetitive work, and deliver services faster. But it cannot determine, on its own, what constitutes good government.
That responsibility remains with people.
Reilly-Hughes’s message is not that Vermont should embrace every new technology or fear what comes next. It’s that government should understand the problem it’s trying to solve before deciding whether technology is the answer.
And the ultimate measure of success isn’t how much artificial intelligence Vermont adopts.
It’s whether Vermonters are better served—and whether the people responsible for serving them remain accountable for the results.