AI isn't a philosophy towns need to "adopt." It’s just software, a tool in the toolbox meant for specific, targeted jobs.
But knowing what tasks AI should touch is only the first step. The true test for municipal boards and IT directors is knowing how to deploy it safely, effectively, and without blowing the budget.
The standard I hold software to
Good software solves a problem once. You get the logic right, the code right, and the problem is handled; you move on to the next one. That payoff is the entire reason I've been building software since I was writing AppleScript in fifth grade, and it's the standard AI should be held to in a town hall.
Most municipal work fails that standard today, through no fault of the people doing it. It's the same task, redone by hand, every board, every meeting, every month. AI brings a new layer of flexibility to what software can handle. To figure out where it belongs, I run tasks through three simple filters:
-
Filter 1: Recovery is easy. A simple question: what happens if this is wrong? Even the best AI systems are prone to making small errors. If an error can be easily fixed or doesn't hurt the overall outcome, AI can net big productivity gaines. If a simple mistake would be costly, though, AI isn't yet reliable enough to put in charge.
-
Filter 2: A human remains the final decision-maker. While AI offers flexibility, it also introduces randomness. Think of AI as a drafting engine or a junior intern: the work gets done, but it strictly requires an experienced, senior eye to review and approve it.
-
Filter 3: The "Is anyone doing this right now?" test. Software usually speeds up work that is already happening. But local government is full of tasks that should be done, but simply aren't because no one has the bandwidth. In these cases, the math changes. If a task currently yields zero output, AI doesn't need to be 100% perfect to create massive value. It just needs to be better than nothing. As long as it passes Filter 2 and a human verifies the final output, an imperfect AI draft beats the status quo of zero every single time.
Any deployment that passes all three filters will free your experienced staff to put their time where it actually creates value for residents.
AI isn't a crutch
"Garbage in, garbage out" is a common phrase especially in software development. For AI, it is 100x true. If your town doesn't have clear, reliable, and structured data, an AI deployment won't succeed.
If your town struggles to send out timely communication, is constantly behind on updating the website, or generally doesn't adopt a "digital-first" mindset, AI won't make any of that better. A town must have a strong base of data systems for an AI deployment to have any chance of not making things more difficult. If you know your town struggles with digital information, stop here, do not pass go.
Starting without betting the town
When tasks get large, our ability to pay attention drops and mistakes get through. Pick one narrow, high-recurrence task and find a solution that improves it. Do not get mesmerized by plans of a grand "AI strategy"—you will learn more from doing one repetitive task well than from a sweeping initiative that touches everything, costs way too much, and requires rollbacks later.
-
Demand real output from your own material. A demo is built to impress. Any AI demo must reflect your town, your boards, your information. AI deployments fail because the demo was trained on data specifically made for it, not the unique way your town operates.
-
Name the human who will own the process. Confirm with them that it will make their work easier. Managers too often deploy tools to help "productivity" without consulting the actual workers.
-
Give staff a sanctioned tool. Your employees are very likely already pasting things into consumer chatbots. A town-controlled tool for the specific job is a security measure disguised as a productivity one, pulling that activity back where it can be governed.
Why narrow tools beat "everything" platforms
One more lesson from the failures, and it cuts against where most of the industry is pushing. The government AI disasters share a shape: a broad, general-purpose deployment that fails at every turn. AI was billed as "magic," the magic turned out to be a sloppy sleight of hand, and any potential benefits are lost to the weight of the overall failure.
If you think of AI as what it is—software—it's easier to understand. You don't buy "town" software; you buy bill pay software, GIS software, HR software, assessment software. Software is narrow, built around the real structure of that job. That's a builder's instinct, and it's the one I trust: solve the actual problem well and resist the sprawl. A tool that claims to do everything for your town will be a master of none.
That's the philosophy behind what I build, stated plainly so you can discount for bias. MuniMins turns a board's meeting recording into formal, review-ready minutes, and then gets out of the way while the clerk approves them. It doesn't manage agendas, doesn't run meetings, and decides nothing on the town's behalf. Judged by the three filters: it handles a repetitive task (or one sitting at zero), requires a human to verify, and keeps data in a purpose-built tool. It sits exactly where AI belongs, and it stays there on purpose.
Towns need to lean into technology—including AI—but they must do so deliberately. We need to apply the right tools to the right tasks, keep narrow software in its lane, and ensure experienced humans remain at the helm. That is how we unlock real value, protect our residents, and make local government run a little smoother.
Making government easier for everyone through thoughtful deployment of technology...So moved...