AI in Local Government: Here We Go

Intro

We seem to be living in a transformative age, so what should feel different, and how do we make it so?¹

First let’s talk local government. Local agencies define an enormous if sometimes understated share of what it feels like to live somewhere. Housing, parks, infrastructure, transit, sewage, development, growth, public health, public safety, permits, and a very long-tailed ‘etc’. Improving how these processes operate could mean subtly or significantly improving residents’ lives, depending on the use-case. So, what can AI do about it?

As a mild disclaimer, I’m writing about this in the abstract, despite AI use in local government already being very real.² You’ll have to forgive me, it’s only because I just don’t know enough to write concretely. This paper is mostly me getting my initial thoughts straight in a very general way about how AI could change local government operations for the better, and what that might take to happen.

Inspired by the AI transformation literature³ and the general work improvement literature⁴ more broadly, three ‘paths’ local governments could take to use AI transformatively came to mind: AI automates or augments work directly, AI makes it easier to change work, or AI acts as an interface for residents directly. These options are not necessarily exhaustive, and many orgs may pursue each simultaneously or in rapid succession, but they help give me some mental structure to organize the work involved in local government AI innovation.

Like previous innovations, though, all paths to AI transformation involve costs and hurdles, from capacity and expertise, to security and procurement. And, also like prior innovations, these costs and hurdles will determine how far along any one of these paths a local government progresses.

Three Possible Paths to Local Government AI Transformation

1. AI performs or augments work directly, reducing the effort required for existing tasks.

This is already the case in many places, as AI is just very helpful for certain tasks.⁵ Not to catalog use-cases, but activities like information retrieval, file editing, coding, etc, are abundant across local government and private sector teams. As AI becomes more capable, it should, at least theoretically, be able to automate or augment a larger and larger share of even complex, high value activities.³

That said, accessing the benefits of AI in a process is rarely as easy as ‘just have AI do it.’ Even for tasks that AI is particularly suited to there are still the questions of:

1) How (i.e., through what platform or tool) does AI ‘access’ the task, and

2) What is the work of enabling AI within the task?

For example, if we want AI to help us process permits, do we write up a python script with an LLM to check permit PDFs for inconsistencies as they come in? Does a planning analyst open Claude CoWork and ask to check a folder of applications for missing documents? Does someone log into a custom chatbot to ask “how many zone 3 industrial land-use permits have been issued this year?”

And what has to happen for the user (or their boss, or their IT manager) to be able to do any of those things? What coding or configuring or setting-up is required, and by whom, for different kinds of automation or augmentation to be possible? How are outputs verified and trust built? Who has to do what for all that to happen?

Sadly, and with almost shameless abandon, I’m asking a lot of questions and leaving us both hanging on the answers. What these questions get at, though, is that even for work that AI does ‘for’ you, there is a cost to getting AI in there in the first place and potentially ongoing human involvement. If those costs come in the wrong shape or at the wrong time, they can bound AI’s transformation potential, whatever the benefit it offers.

2. AI makes transformation work easier

But, AI doesn’t just help with ‘core’ work, it helps with the work of improving work (e.g., ERP transformation, record digitization, process automation, etc), potentially expanding access to innovations from more ‘advanced’ localities. This point relies on a kind of “transformation overhang,” the idea being that innovations may pop up and improve a place, but then fail to diffuse to all the other places that could benefit.

AI, by reducing the effort of certain transformation tasks like schema mapping, process documentation, or coding itself, could help existing best-practice digital improvements that were once too much work to implement reach more agencies, or reach more deeply within an agency. Like a rural area without phone service skipping land-lines to build cell towers, could AI let agencies ‘skip-ahead’ to the same digital infrastructure as their more capable peers?

For the sake of finishing this essay we will be leaving most of these points painfully underspecified. For example, what are the modernizations in question (cloud ERP transitions, custom software, database migrations, etc)? What tasks are involved, and how specifically does AI affect them? And, if AI can affect them positively, what is the work of learning & applying AI within them?

As above, the costs (financial and not) of implementing AI, like any other technology, determine how and where it spreads. Exciting transformation initiatives are much less exciting, or effective, when no one has time to support them properly. Even if AI reduces the time or effort of some portions of these projects by some degree, we’re still left wondering whether, or for whom, those reductions are enough?

Nevertheless, the possibility is exciting. AI’s various local government applications are new entries in a long history of local government innovation. Even if AI did nothing but make successful innovations more accessible, it would be an enormously helpful technology for public agencies everywhere.

3. AI as an interface for government services.

Finally, AI can support residents directly by acting as a kind of agentic interface for government operations. For example, a resident can go to a local government website and fill a complex form by talking to an AI chatbot in plain-language.

These initiatives (i.e., agentic chatbots) still take a lot of work and a comparatively significant amount of ongoing technical expertise, and not always with great results.⁶ As developers gain experience with the process⁷, and AI gets rapidly smarter, more capable, and cheaper, the ease and reliability of agentic interface options may advance rapidly as well.

Agentic interfaces are an attractive possibility because they parallel, in a way, the ability to access government services through the internet i.e., a new channel that has been immensely helpful but can vary in quality and change dramatically over time.

Government work still involves a lot of manual paper moving and compilation, work some of which AI can do just as well if not better. And interacting with local government too often still involves maneuvering the same files, forms, and pieces of information across tools and services. Though it remains to be determined when and how specifically an agentic interface is the appropriate channel for resident-government interactions, they offer the enticing possibility of a system that pulls information, submits forms, explains clearly, and guides effectively, all with infinite patience, at any time.

Conclusion

If transformation were easy everyone would do it, and most people would do it well. But it very, very clearly isn’t, especially for organizations with limited capacity and substantial bureaucratic constraints. That has always been the case⁴, and yet local governments, not classically known for abundant resources or streamlined bureaucracies, have still managed to dramatically change how they work and what they offer.⁸

AI is a new, general purpose technological offering in this broader transformative landscape, but, because it operates not only on work itself, but the work of changing work, and the interaction with government itself, it may come to represent a broad and (almost) unprecedented upheaval across local government operations.

Ensuring that upheaval is possible, positive, and well-distributed will take a lot of work, some of which is already being pursued by local government workers and administrators across the country and the globe. This essay is a first attempt for my own edification at mapping out what that looks like, how it works, and in the end, what someone interested might do to help.