The Pothole That Learned to Report Itself

In San José, California, a pothole no longer waits for a resident to notice it, dial 311, describe the exact block, and hope somebody dispatches a truck. A camera mounted on a city vehicle sees it first, tags it with 97 percent accuracy, and files the ticket before a tire ever hits it.[^1] That is not science fiction. That is a Tuesday. And it is a small, unglamorous preview of a much bigger shift already reshaping local government from the sidewalk up.

Ninety-five percent of English councils are now using or exploring artificial intelligence, up ten points in a single year.[^2] In the United States, roughly forty-six percent of state and local government employees say they are already using AI tools at work, most often to draft, summarize, and process documents.[^3] The technology is no longer arriving. It has arrived, in the break room, on the laptop, and inside the permit-review queue. The only real question left for cities, counties, and towns is whether their people will shape how AI shows up in the public square, or watch it show up without them.

AI Literacy Is a Culture, Not a Course

There is a tempting way to approach AI literacy that goes something like this: buy a training package, run everyone through a two-hour webinar, print certificates, declare victory. It fails almost every time. Skills learned in a stuffy conference room on a Thursday afternoon evaporate by Monday morning, especially when the tools change every quarter.

What actually works looks less like a class and more like a habit. A community of practice where a records clerk can show a planner the prompt that cut her weekly report in half. A monthly demo lunch where the parks department shares what worked and what embarrassed them. A shared prompt library, living on the intranet, that grows the way a good recipe box does, with dog-eared favorites and honest notes about what to skip.

Partnerships accelerate this culture. A community college down the road, a research lab at the state university, a civic-tech nonprofit, or a vendor willing to co-design instead of just sell — each one brings expertise that most municipalities cannot afford to hire full-time. Bloomberg Philanthropies' City AI Connect platform, launched to link local governments experimenting with generative AI, is a working example: peer cities swapping playbooks on multilingual chatbots and transit optimization, so nobody has to invent the wheel alone.[^4]

For those just starting out, the on-ramp is simpler than it looks. Pick one recurring task you hate. Try one AI tool on it this week. Tell one colleague what happened. That is literacy. Everything else is scaffolding.

The Fear in the Room Is Real, and It Deserves an Answer

Seventy-one percent of Americans worry that AI will permanently displace workers.[^5] Inside public-sector organizations that have already adopted AI, sixty-nine percent of employees say "AI anxiety is real" in their workplace, driven by concerns about job security, ethics, privacy, and misinformation.[^6] Any leader who walks into an all-staff meeting waving a shiny pilot without acknowledging that fear is going to lose the room before the slides load.

The honest answer is not a pep talk. It is a promise, backed by policy. When Seattle rolled out its generative AI policy, it did not lead with hype. It led with a vetting process, disclosure requirements, and rules about attribution and sensitive data — the boring, trust-building infrastructure that tells employees the city is treating this seriously.[^7] San José went a step further and required employees to log their generative AI use through an internal reporting form, creating a public record of what was tried and what worked.[^8]

Three moves consistently soften resistance without pretending the concerns are silly:

• Name what will not change. "Nobody is being laid off because of this pilot" is a sentence worth saying out loud, in writing, more than once.

• Put staff in the driver's seat. The clerk who processes 300 permit applications a week knows exactly where an AI assistant would help and where it would create chaos. Ask her before the vendor demo, not after.

• Reward the sharing of failure. A pilot that flopped and taught the team something is more valuable than a pilot that was quietly buried.

The PwC 2025 Global Workforce survey found that daily generative AI users report higher productivity, higher pay, and higher job security than their peers.[^9] That is worth telling employees plainly. The people who learn to work alongside these tools are not the ones being replaced. They are the ones being promoted.

Start Where the Paperwork Hurts

The instinct in government is often to swing for the fences — a citywide AI transformation, a fifty-page strategic plan, a task force with a name that fits on a coffee mug. The instinct works better in reverse. Start with the smallest problem that annoys the most people.

San José did not launch its road-safety AI as a citywide moonshot. It bolted cameras onto a handful of city vehicles, tested pothole detection, hit 97 percent accuracy, and only then expanded to street sweepers scanning for bike-lane obstructions and construction debris.[^10] Buenos Aires did not build a citywide digital twin. It built "Boti," a WhatsApp chatbot for residents, and now handles roughly five million conversations a month through it.[^11]

Good first candidates for AI look almost boring on paper:

• Tasks that repeat hundreds of times a week with only small variations, like triaging 311 tickets or extracting data from scanned forms.

• Work that already generates enormous datasets nobody has time to read, like inspection reports, meeting minutes, or public comments.

• Front-door services where residents wait too long for simple answers in the language they actually speak at home.

Whatever the first project is, define what "good" looks like before turning anything on. Not "we deployed AI." Something measurable: response time cut by X, backlog reduced by Y, resident satisfaction held steady while staff hours dropped. A pilot without a scoreboard is a demo, not a decision.

Data Is the Foundation, and Trust Is the Roof

An AI tool is only as good as the data it is fed, and public data is often held together with duct tape and institutional memory. Records live in three different systems, use four different naming conventions, and are guarded by the one employee who remembers why. Cleaning that up is not glamorous, but it is the difference between an AI pilot that dazzles and one that quietly hallucinates its way into a lawsuit.

Privacy sits right next to data quality, and residents are watching. Half of public servants who use AI at work are reaching for publicly available consumer tools, which is exactly how sensitive constituent data walks out the door.[^12] A serious local government AI program answers three questions in writing, in plain English, and in public:

• What data can be fed into which tools, and what absolutely cannot.

• How residents are told when AI touched their case, and how they can opt out or appeal.

• Who inside the organization reviews AI outputs before they affect a person's benefits, permit, or safety.

Transparency is not a compliance checkbox. It is the reason a parent trusts a school district's AI-assisted enrollment system, or does not. Get the plain-English version right first. The legal version will follow.

Break the Silos Before the Silos Break the Project

The best AI ideas rarely come from the IT department alone. They come from the moment a librarian, a code-enforcement officer, and a data analyst end up in the same room and realize they have been solving three versions of the same problem for a decade. Cross-departmental working groups, hackathons that invite frontline staff, and standing "show and tell" sessions where teams demo what they built last month are cheap, and they compound.

External collaboration matters just as much. Neighboring jurisdictions almost always face the same headaches, and a joint pilot with a shared vendor contract can cut costs while doubling the learning. Community groups, especially those representing residents who are historically underserved by digital services, need to be at the design table early, not surveyed after launch. An AI chatbot that works beautifully in English and stumbles in Spanish or Mandarin is not a chatbot. It is a new form of exclusion.

The Real Question Is Not Whether. It Is Who Decides.

AI in local government is not coming. It is here, in the drafting of memos, the routing of complaints, the flagging of potholes, the summarizing of council packets, the answering of the phone at two in the morning. The choice in front of every city manager, department head, first-year analyst, and new hire is not whether to engage. It is whether to shape what engagement looks like, or to inherit whatever the vendors, the headlines, and the loudest voices decide for them.

If you are a leader, the question this week is not "should we do AI." It is "what is the smallest, most honest pilot we can start on Monday, with the people most affected in the room?" If you are early in your career, the question is not "will AI take my job." It is "which repeatable, tedious part of my week can I hand to a tool by Friday, so I can spend Monday on the work that actually needed a human?"

The pothole is already reporting itself. The permit is already being pre-checked. The chatbot is already answering in three languages. The only remaining variable is you.

Pick the one thing. Try it this week. Tell somebody what happened.

That is how a city learns.

References

[^1]: City of San José. 2025. "San José AI Road Safety Pilot Data Shows 97% Accuracy." News release, September 29, 2025. https://www.sanjoseca.gov/Home/Components/News/News/6926/4699.

[^2]: Local Government Association. 2025. State of the Sector: Artificial Intelligence — 2025 Update. London: Local Government Association. https://www.local.gov.uk/publications/state-sector-artificial-intelligence.

[^3]: MissionSquare Research Institute. 2025. Artificial Intelligence in the Workforce: A Survey of State and Local Government Employees. Washington, DC: MissionSquare Research Institute. https://research.missionsq.org/content/media/document/2025/5/Artificial_Intelligence_StateLocalEmployees.pdf.

[^4]: World Economic Forum. 2024. "How 5 Global Cities Are Innovating with Generative AI." July 24, 2024. https://www.weforum.org/stories/urban-transformation/generative-ai-smart-cities/.

[^5]: Reuters. 2025. "Americans Fear AI Permanently Displacing Workers, Reuters/Ipsos Poll Finds." August 20, 2025. https://www.reuters.com/world/us/americans-fear-ai-permanently-displacing-workers-reutersipsos-poll-finds-2025-08-19/.

[^6]: KPMG in Canada. 2025. "Half of Public Servants Turn to AI, Raising Risks." September 2025. https://kpmg.com/ca/en/media/2025/09/half-of-public-servants-turn-to-ai-raising-risks.html.

[^7]: Smart Cities Dive. 2023. "Generative AI Is Here. This Is How 5 Cities Plan to Manage Its Use." November 16, 2023. https://www.smartcitiesdive.com/news/artificial-intelligence-how-cities-plan-tech/699906/.

[^8]: Smart Cities Dive. 2023. "Generative AI Is Here. This Is How 5 Cities Plan to Manage Its Use." November 16, 2023. https://www.smartcitiesdive.com/news/artificial-intelligence-how-cities-plan-tech/699906/.

[^9]: PwC. 2025. Global Workforce Hopes and Fears Survey 2025. London: PwC. https://www.pwc.com/gx/en/issues/workforce/hopes-and-fears.html.

[^10]: StateScoop. 2025. "AI Now Powers Street Hazard Detection in San Jose, Calif." September 29, 2025. https://statescoop.com/san-jose-ai-street-hazard-detection/.

[^11]: Tomorrow.City. 2025. "How Generative AI Is Shaping Today's Cities." February 3, 2025. https://www.tomorrow.city/how-generative-ai-is-shaping-todays-cities/.

[^12]: KPMG in Canada. 2025. "Half of Public Servants Turn to AI, Raising Risks." September 2025. https://kpmg.com/ca/en/media/2025/09/half-of-public-servants-turn-to-ai-raising-risks.html.