Who Decides What Vermonters Need? A $250K Project Hands Part of the Answer to an AI Tool
The effort carries no apparent state mandate. And it raises a question worth asking out loud: can a tool built for transparency help measure need?
A federal-grant project led by UVM Extension will map what Vermonters need to thrive, drawing in part on Local Minutes, an AI tool that searches municipal meeting records. The tool's own disclaimer says it should not be the sole basis for important decisions.
The University of Vermont is trying to learn more about what Vermonters need to thrive — not in the abstract, but through what UVM Extension describes as a statewide assessment of community needs, assets and perceptions. The project is called the Vermont Assessment of Assets, Angles, and Need, or VAAAN, and UVM Extension says it will use a mixed-methods approach to explore what is needed across a broad range of basic social themes, what assets communities already have, and how needs and assets are perceived.
One named partner is Local Minutes, an artificial-intelligence service that ingests municipal meeting minutes and other public documents and lets users ask questions about them. The premise is straightforward: if you analyze what selectboards, school boards and commissions are actually talking about, patterns of community concern may emerge.
It is an ambitious idea, and in places a genuinely useful one. It is also being built mostly outside the institutions Vermonters might assume would define their needs — and on a foundation whose builders are unusually candid about its limits.
The money, and who is at the table
VAAAN is paid for by federal money routed through the university. The Leahy Institute for Rural Partnerships, funded by a $14 million, four-year grant from the U.S. Department of Agriculture’s National Institute of Food and Agriculture, awarded the project $250,000 in its third grant round. In the institute’s official list of awarded proposals, the grant is titled “Redefining Community Needs Assessment,” its UVM partner is Extension, and its named community partner is Local Minutes. The stated problem: Vermont “lacks an annual comprehensive assessment of community needs to inform planning.” The stated solution: a “living and interactive ‘Vermont Analysis of Need.’”
A note for the record: in grant documents the project is the “Vermont Analysis of Need,” abbreviated VAN; in UVM Extension’s public-facing material it is called VAAAN. The available materials describe the same project.
What the public grant materials and UVM Extension’s announcement do not show is a formal role for state government — no indication that the Legislature, Gov. Phil Scott’s office or any state agency commissioned the assessment, reviewed it or holds authority over it. The project’s own public framing points inward as well as outward: Interim Director of Extension Chris Callahan has tied VAAAN to UVM’s strategic plan and the College of Agriculture and Life Sciences’ research priorities.
VAAAN does plan to convene two advisory committees — one drawn from, in Callahan’s words, “positions of decision-making power and relative wealth of resources,” and another from the grassroots level. Whether that first committee will include state officials or legislators is not yet public.
How the project decides what counts as “need”
The scraping is not the whole method, and it would be unfair to suggest the project intends to read minutes and simply declare what towns are missing. By UVM Extension’s own account, VAAAN is a mixed-methods effort: the Local Minutes analysis is described as one part of a broader process that includes a review of literature, the advisory committees and additional validation.
The project also says it is alert to how the same need looks different in different places. Co-investigator Mariya Shcheglovitova, a human geographer and Extension assistant professor of community and economic development, has described the tool detecting that smaller rural communities discuss school closures and the squeeze between property taxes and local control, while larger towns frame education around childcare and trades.
[ANALYSIS] But the discovery layer sets the initial agenda. The validation steps described publicly appear designed in part to catch false positives — themes the AI flagged that turn out not to matter. The harder problem may run the other way: false negatives — needs that never surface because the people who carry them do not attend meetings and do not appear in the minutes. A survey of experts might backfill some of that, but the public method still leaves an important distinction: measuring civic conversation is not the same as measuring community need.
What the tool actually is
Here the project’s foundation deserves a close look, because Local Minutes is plain about what it is and is not.
The service says it obtains documents “from municipal and public body websites using automated methods” — that is, it scrapes posted minutes and agendas off town websites. Those records are public by law. Under 1 V.S.A. § 312, public meeting minutes must be available after five calendar days, and minutes must be posted within five calendar days to a website if the public body maintains or has designated one as its official website. Except for draft minutes replaced by updated minutes, posted minutes cannot be removed from the website sooner than one year from the date of the meeting.
That posting duty is conditional on having a website in the first place — which helps explain the coverage gap. A February 2026 Norwich Observer item, citing Local Minutes, reported that the site covered 179 of Vermont’s 247 municipalities, or roughly 72 percent. The towns left out are disproportionately the smallest and least-resourced — the ones least likely to maintain a usable website.
The tool was built, and has been described publicly, as a transparency aid: a way for ordinary citizens to find out what their town government did. Seven Days reported at its 2025 launch that founder and software engineer Duane Millar Barlow came up with the idea after joining Essex’s Conservation and Trails Committee and discovering how difficult it was to dig useful history out of past meeting minutes. That origin is worth keeping in view, since the tool’s original public-facing purpose was civic transparency.
And it does that job. But Local Minutes’ own disclaimer is unusually frank about the boundaries of what it can be trusted to do. The AI, it warns, “may occasionally generate incorrect or misleading information (often called ‘hallucinations’).” Responses “might contain errors, inaccuracies, or omissions.” It is “not an official government source” and “not an authoritative record.” And, most pointedly: it “should not be the sole basis for making important decisions or taking action,” and “lack of an answer is not proof of the non-existence of information.”
[ANALYSIS] None of that is a knock on Local Minutes. Those are appropriate disclaimers for a free convenience tool, and the company should be credited for posting them. The question is what happens when an instrument built with that level of caution becomes a named, grant-funded input for an assessment designed to inform planning and institutional priorities. The humility appropriate to a citizen’s lookup tool becomes a live issue when the same outputs help guide where attention and resources go.
A transparency tool, asked to do a different job
[ANALYSIS] This is the heart of it. A transparency tool answers a descriptive question: what did this body discuss and decide? A needs assessment answers a more judgment-based one: what do communities require to thrive? The first is about the record. The second is about people — including the people who never enter the record. A tool can be useful for the first and still incomplete for the second, precisely because the second depends on seeing what the documents leave out.
The risk is not hypothetical, and the same grant round documents the foil. In the cycle that funded VAAAN, the Leahy Institute also awarded $237,198 to the Vermont Professionals of Color Network for a “Vermont Community Data Assessment” — a project premised on the fact that available Vermont data leaves gaps in understanding populations not always captured in data collection. Set those two awards side by side and the tension is clear: one project exists because existing data under-represents some Vermonters; the other proposes to read need in part from meeting records, which tend to over-represent whoever shows up to meetings and whichever towns maintain a website.
Vermont has rules for government AI. This project doesn’t appear to be covered.
The state has not been casual about artificial intelligence. Act 132 of 2022 created a Division of Artificial Intelligence inside the Agency of Digital Services and an AI Advisory Council. The statute directs the division to review AI systems “developed, employed, or procured in State government.” It also directs the division to consider policies to protect Vermonters from unfair discrimination caused or compounded by government AI and to address systems “that have not been tested for bias or have been shown to contain bias.” The AI Advisory Council is scheduled to be repealed on June 30, 2027.
[ANALYSIS] That framework, by its own terms, governs AI used or bought by state government. On the public record, VAAAN is a UVM Extension project involving a private vendor and federal grant funding — not a state procurement — so it does not appear to fall under that same review structure. That is not a violation of state AI law. It is a governance question: Vermont has built a bias-conscious oversight regime for state-government AI, while a federally funded university project designed to map Vermont’s needs sits outside it.
What it’s for
The people building VAAAN describe something genuinely collaborative. Co-investigator Gianni Solórzano has framed the eventual tool as a way to break down silos — a place where someone working on, say, food security could filter by region, see who else is doing that work, and build a partnership rather than duplicate effort. That is a real problem, and access to hard-to-search municipal records is a real public good. On both counts, the instinct behind the project is sound.
The open questions are narrower, and they are the ones worth watching as the project develops over the next two years: whether anyone in state government is consulted or accountable; how heavily the AI layer is weighted against the human ones, and whether that weighting is documented; and whether a tool built to reveal what towns say can be trusted to help reveal what Vermonters need — especially the Vermonters who rarely make it into the minutes.
Compass will be following the project as its advisory committees form and its methodology is published.
Compass Vermont is independent and reader-supported.
I am not sure this article is entirely accurate.
Many of the municipal websites use CivicPortal and other similar sites to host their content. CivicPortal for certain can be accessed directly by writing a small amount of code to download the data directly. I have done this to analyze town decision making over the last decade.
The AI is predominantly in the user interface layer where people ask about local questions and issues and then LocalMinutes interrogates the data or has downloaded.
In many ways, it is a good use of AI as it makes it easier for non-software engineers to get meaning from town government records and increases transparency.