What Stalls AI Adoption in Local Government?

The first month ChatGPT was public, I asked each Think Tank group whether anyone had heard of it. Most of them heard about it first from me, in that question. We were nearly all standing in the same unknown place, looking at the same thing from the same distance.

Those same kinds of gatherings now hold a larger range. Some members are daily users with real institutional implementations underway. Others rarely touch it and have not needed to. Watching that spread widen, month after month, with the same people in the same seats, has taught me something about adoption that the tools themselves do not explain.

The People With the Least Need Could Hold the Most Influence

Last week on LinkedIn, Micah Gaudet, Deputy City Manager for Maricopa, Arizona, asked “What do you see as the biggest barrier to AI adoption in local government?” I couldn’t help but consider the situations where the people with the most influence over how (or whether) a team adopts AI are the ones with the least personal need for it. When you do not need a tool, you rarely build fluency with it, and it is human to feel a step behind. In local government, it doesn’t take long to go from feeling a little behind on technology to feeling exposed.

So what happens in local government organizations where the elected body and/or the appointed officials sitting on the dais don’t have a need to use AI? Could there be an unconscious resistance bias stemming from the lack of developed skill or proficiency?

Micah wrote about the responses he received here.

When Caution Is Wisdom, and When It Becomes a Blind Spot

I want to be careful here, because caution about AI in government is frequently the right call. Protecting what feels stable when the ground is moving is not a flaw. It is instinct, and instinct is worth noticing in ourselves, because it can quietly shape policy before we have named what we are protecting. That unnamed thing is the blind spot. Not a failure of attention or character. A natural consequence of where we sit and what the position asks of us.

This is the part worth saying plainly. A blind spot is rarely incompetence. It is usually instinct doing its job a little too well, in conditions the instinct was not built for. The same depth of experience that makes a manager effective also creates stable patterns, and stable patterns have edges where perception narrows. The people who can see past those edges are lower in the organization, and honest upward feedback does not flow easily in any organization, least of all in government, where FOIA exposure, political dynamics, and career risk all shape what people are willing to say.

AI Lands on a Desk That Is Already Full

None of this is happening in calm conditions either. AI is arriving on top of pressures that were already heavy. The political environment has shifted under local leaders in ways the 2019 playbook did not account for. Americans still trust local government far more than they trust Washington, by a wide margin, and yet most local officials now feel the spillover of national polarization in their council chambers and public meetings. We collected that research in The Trust Paradox, and the numbers there describe a daily reality, not a headline.

The workforce has shifted too. Many employees are not quitting, but they have quietly disconnected. Fewer than half can say they know what is expected of them, a pattern we gathered in The Expectation Gap. For a lot of the long-tenured managers and administrators I talk with, the last two years have been harder to navigate than the pandemic, when at least everyone knew they were in the unknown together.

What the Research Names, and What I Hold Loosely

The AI piece sits inside all of that, and the research names the gap directly. The widest version of that finding comes from broad workforce surveys rather than local government, so I hold it loosely, but its shape matches what I watch happen in the Think Tanks. We wrote about it in The AI Anxiety Gap. What the surveys measure is not really the technology. It is whether the conditions exist for people to say what they actually think, and that is a Face the Truth question first.

Why the Caution Runs Deeper Than the Tools

There is a quieter reason the caution runs deep, and it is a good one. Using a consumer AI tool means handing your information to a third party whose policies decide what happens to it. A federal court said as much in writing this year, the first opinion of its kind, confirming what many government attorneys had been cautioning all along. I wrote about what that ruling means for organizations built on self-governance. So when a manager hesitates, some of that hesitation is wisdom. The work is to separate the wisdom from the instinct that is protecting something we have not yet named.

A Blind Spot Is Hard to See Alone

That separation is hard to do alone, which is the whole reason our Think Tanks exist. A blind spot named by yourself is useful. A blind spot explored with other people who have sat in the chair is something else. It is why the approaches that built careers do not always survive changed conditions, a pattern I have watched closely enough to write about in its own right. The managers who navigate this well are not the ones with the best technology strategy. They are the ones who have built, around themselves, the conditions where the honest version of a thing can be said.

We have been building a careful answer to the careful instinct, an AI tool that runs on a locally hosted model so the conversation never routes through a consumer platform. It is described, with its privacy architecture, on The Shift Mirror page. I mention it less as a thing to go do and more as a sign of how we think the technology should be handled here, which is the same way the work itself proceeds. Slowly, privately, and with the person’s own judgment kept at the center.

What Is Shaping Adoption Where You Sit?

What is actually shaping adoption where you sit? In some places it is moving faster than the room is ready for, in others it has stalled, and plenty are somewhere in between and hard to read. The reason is rarely just the tools. It might be the policy, the workload, or something closer to home that is harder to name. If any of this lands, I would rather hear your version of it than guess at it. Start a conversation with us, and we can think it through together, which is the only way I have ever seen it done well.