Is Your Community AI Ready, or Just AI Curious?

Imagine a resident calling city hall because her heat has been shut off. The employee who answers has two choices: spend twenty minutes hunting through scattered policy documents, or use an approved AI assistant to find the right program in seconds. The technology can make the search faster. It cannot hear the worry in her voice, weigh the circumstances of her family, or take responsibility for what happens next.

That is the real promise and challenge of artificial intelligence in public administration. AI can help governments translate information, summarize meetings, identify infrastructure problems, draft routine communications, and make public records easier to search. Yet government is not a customer service app, and residents are not data points. When a decision touches housing, employment, public safety, education, health, or benefits, speed matters, but so do dignity, fairness, privacy, and the right to ask a human being, “Why?”

The question for community leaders is no longer whether AI will enter public service. It already has. The better question is whether leaders will shape its use deliberately or allow convenience, vendor promises, and scattered experimentation to shape it for them.

AI Literacy Is a Leadership Skill, Not a Coding Course

AI literacy does not require every mayor, agency head, supervisor, or new public employee to become a machine-learning engineer. It requires them to know what AI can do, where it can fail, what information should never be entered into an unapproved tool, and which decisions demand meaningful human review.

That distinction matters because enthusiasm is running ahead of readiness. A National League of Cities report cited a survey in which 96 percent of mayors expressed interest in generative AI and 69 percent said their municipalities were exploring or testing it, but only 2 percent were actively implementing it. The gap between curiosity and responsible use is where leadership belongs.

Traditional training often misses that mark. A slide deck on definitions may explain what a large language model is, but it will not prepare a housing director to question an AI-generated eligibility summary or help a recent hire recognize that a confident answer can still be wrong. The OECD has warned that the supply of training may not be keeping pace with the growing need for general AI literacy, especially training that is accessible and inclusive.

Effective learning should feel less like a technical lecture and more like a flight simulator. Give staff realistic scenarios. Ask one group to use AI to simplify a dense public notice, another to check the result for missing legal details, and a third to rewrite it for a resident with limited English proficiency. Let participants compare outputs, find errors, discuss risks, and decide when the work needs an expert.

Leaders can build this habit through short, recurring practice rather than a single annual seminar. A fifteen-minute “AI reality check” at a staff meeting can examine one useful prompt, one flawed output, and one policy question. Early-career employees can contribute fresh tool knowledge, while experienced staff contribute the institutional memory and community context that technology lacks. That exchange turns AI literacy into a team sport instead of a top-down mandate.

Build Guardrails Before Buying the Guard Dog

An AI framework should not be a decorative policy stored in a shared drive. It should tell employees what they may use, what data they may share, who approves new tools, how residents will be informed, how outputs will be checked, and what happens when a system causes harm.

The National Institute of Standards and Technology offers a useful starting point. Its voluntary AI Risk Management Framework is designed to help organizations incorporate trustworthiness into the design, development, use, and evaluation of AI systems. The U.S. Government Accountability Office organizes AI accountability around four complementary principles: governance, data, performance, and monitoring. In plain language, leaders should know who is responsible, whether the information feeding the system is fit for purpose, whether the tool performs as promised, and whether anyone is still watching after launch.

A workable local framework can begin with a simple rule: the greater the potential impact on a person’s rights, safety, livelihood, or access to services, the stronger the review must be. Using AI to brainstorm a meeting agenda is not the same as using it to flag a family for a fraud investigation. Drafting a social media caption is not the same as ranking job candidates. Treating every use case alike either creates paralyzing red tape or dangerous shortcuts.

Before approving a tool, leaders should be able to answer a short chain of questions in complete sentences. What public problem are we solving? Why is AI more appropriate than a simpler process change? What data does the system use? Who could be harmed if it is wrong? Can a resident receive an explanation and challenge the result? Who has the authority to pause or shut it down?

Infrastructure matters too, but the most expensive platform cannot rescue poor data or an unclear purpose. Agencies should first examine whether their records are accurate, current, secure, and representative. They should also budget for training, cybersecurity, accessibility, evaluation, and ongoing staff time. The purchase price is only the front door. Responsible maintenance is the rest of the house.

Put Bias Under a Spotlight, Not Under the Rug

AI does not arrive as a neutral referee. It learns from human-created information, and that information can contain old patterns of exclusion. A system trained on yesterday’s inequities can reproduce them tomorrow at digital speed.

Consider a hiring tool that favors language found in the résumés of previous successful candidates. If past leadership pipelines excluded certain neighborhoods, schools, age groups, or disability communities, the tool may quietly reward the same narrow profile. A pothole-prediction system can create a different problem if it relies heavily on complaint data. Neighborhoods with more time, connectivity, English proficiency, or trust in government may report more problems, making need appear greatest where reporting is easiest.

Fairness therefore cannot be reduced to a final software test. It begins with deciding what problem to solve, whose experiences count, and what evidence will be used. Community representatives, frontline employees, civil-rights advocates, accessibility specialists, data professionals, legal counsel, and the people most likely to feel the consequences should be involved before a system is purchased, not invited only after criticism begins.

Boston offers a useful picture of this people-powered approach. In 2024, the city used generative AI to create descriptive titles for more than 1,100 City Council vote records spanning sixteen years, making public information easier to navigate. Its accompanying civic AI work emphasized that people must validate outputs, correct errors, add context, and build trust with communities. The example is modest by design. It improves access to information without handing a machine authority over a resident’s future.

Every high-impact system should be tested across relevant demographic and geographic groups before launch and at regular intervals afterward. Results should be documented in language residents can understand. If a tool performs unevenly, leaders should not hide behind the phrase “the algorithm decided.” Algorithms do not hold office, supervise teams, or answer to the public. People do.

Turn Collaboration Into a Working Prototype

Government does not need to solve every AI challenge alone. Universities can help evaluate systems. Community organizations can reveal barriers that a technical team may miss. Private partners can bring specialized tools and implementation experience. Peer jurisdictions can share policies, procurement language, and lessons from pilots that did not work.

Collaboration becomes valuable, however, only when it is organized around a public need. “We want to use AI” is not a project goal. “We want residents to find accurate sanitation information in their preferred language at any hour” is a project goal. The first statement begins with a product. The second begins with a person.

Small, time-limited pilots can turn that goal into evidence. Start with a low-risk, high-friction task such as summarizing internal meetings, categorizing service requests, drafting plain-language explanations, or searching approved policy documents. Establish a baseline for time, accuracy, resident satisfaction, staff experience, and equity before the pilot begins. Then compare the new process with the old one.

There is promising evidence that carefully supported tools can reduce routine work. In a 2025 United Kingdom government trial involving more than 20,000 civil servants, participants reported saving an average of 26 minutes per day while using generative AI for tasks such as drafting documents, summarizing meetings, and updating records. The figures were self-reported, so they should be treated as an encouraging signal rather than a universal guarantee.

The most important question is not simply, “How many minutes did AI save?” It is, “What did people do with those minutes?” If a caseworker uses the time to listen more carefully to a resident, a planner spends it resolving a difficult permit issue, or a supervisor uses it to coach a new employee, the technology has increased public capacity. If the time savings merely produce higher quotas and thinner human contact, the public may experience a faster system without experiencing better government.

Move From Pilot Fever to Public Value

Government innovation can get trapped in an endless loop of demonstrations. A tool looks impressive in a conference room, a pilot generates excitement, and then no one can explain whether it improved an outcome that residents actually value.

Leaders can interrupt that cycle by defining success before selecting a vendor. A resident-facing chatbot, for example, should not be judged only by the number of conversations it handles. Leaders should also examine whether its answers are accurate, whether residents can reach a person, whether use differs across languages or neighborhoods, whether the source information is current, and whether the tool resolves problems instead of merely deflecting calls.

The same discipline should shape procurement. Contracts should preserve government access to its data, require vendors to disclose meaningful system changes, establish testing and incident-reporting obligations, and provide a realistic exit plan. Staff should know when a model has been updated, because a system that behaved well last month may behave differently after a change.

Public transparency should be practical, not performative. Residents do not need a fifty-page technical manual to understand every automated tool. They do need a clear explanation of where AI is being used, what it contributes, what information it relies on, who remains accountable, and how to request human review. Some cities have begun creating algorithm registries and public AI inventories for precisely this reason. Sunlight does not slow responsible innovation. It helps keep innovation worthy of public trust.

Prepare the Workforce for AI Without Losing the Human Work

The future of public administration will not be a contest between people and machines. It will be a test of whether institutions can combine computational speed with human judgment, local knowledge, empathy, and accountability.

For senior leaders, that means modeling responsible curiosity. Use approved tools, disclose when AI materially contributes to a product, ask difficult questions in vendor meetings, and reward employees who surface risks early. A culture that punishes every failed experiment will hide problems. A culture that celebrates every experiment will repeat them. Mature leadership creates room to test, measure, learn, and stop.

For managers, it means redesigning work with employees rather than imposing tools on them. Ask which repetitive tasks drain time, which decisions rely on professional judgment, and where residents already struggle. Clarify which outputs require review and make sure that review is real, not a hurried click on an approval box.

For people beginning their careers, AI literacy can become a professional advantage when it is paired with public-service judgment. Learn to write precise prompts, verify claims, protect sensitive information, document how a result was produced, and explain a tool’s limitations without jargon. Just as important, learn the policies, histories, and lived realities behind the data. The employee who can use AI and challenge it will be more valuable than the employee who simply accepts its first answer.

Continuous development should include role-specific practice, peer coaching, office hours, and communities of learning. Training should reach procurement staff, program managers, communications teams, legal counsel, analysts, frontline workers, and executives. AI governance fails when responsibility belongs only to the technology department. Public value is everyone’s job.

A Practical 90-Day Starting Strategy

During the first thirty days, map what is already happening. Ask each department where employees are using AI, including free tools that may never have passed through procurement. Identify the tasks, tools, data, expected benefits, and possible risks. Do not begin with punishment. Begin with visibility. Staff are more likely to report informal experimentation when leaders frame the inventory as a chance to build safe pathways rather than catch people breaking rules they may not understand.

At the same time, establish a small cross-functional AI working group. Include technology, legal, procurement, human resources, communications, accessibility, frontline operations, and community engagement. Give the group authority to review proposed uses, recommend interim guidance, and identify one low-risk pilot. Select a problem that is irritating enough to matter but contained enough to manage, such as searching approved policies, translating routine information, summarizing internal meetings, or drafting plain-language notices.

During days thirty-one through sixty, define the pilot before turning on the tool. Record the current time, cost, accuracy, staff burden, and resident experience. Decide what information the system may access, who will review outputs, how errors will be reported, and what result would justify continuing. Create a short staff guide using real examples, then run a practice session in which employees deliberately try to make the tool fail. A team that knows how a system breaks is better prepared to use it responsibly.

During days sixty-one through ninety, test with a limited group and review results every week. Track not only speed but also accuracy, accessibility, employee confidence, resident satisfaction, and uneven effects across communities. Publish a plain-language summary of what was tested, what worked, what did not, and what will happen next. At the end of the pilot, choose deliberately among expanding, revising, pausing, or ending it. Stopping a weak pilot is not a failure. It is evidence that governance is working.

By day ninety, municipal leaders should have more than a new tool. They should have an inventory, an accountable team, an interim policy, a trained pilot group, baseline measures, a public explanation, and a repeatable way to evaluate the next idea. That foundation is far more valuable than a rushed citywide rollout.

Empower Leadership for AI-Driven Transformation

Visionary leadership is not the ability to predict every new tool. It is the ability to establish values sturdy enough to guide decisions when the tools change.

Those values should be visible in daily practice. Efficiency should create more room for service, not less room for people. Innovation should begin with a public problem, not a sales pitch. Transparency should arrive before controversy. Human review should mean the power to question, change, and overturn an automated result. Equity should be measured in outcomes, not merely mentioned in a policy.

AI can help government move faster, but leadership determines the direction. The next step does not require a massive platform or a perfect strategy. Choose one frustrating, low-risk process. Invite the employees who do the work and the residents who experience it. Define what better would look like. Test an approved tool, measure the result, publish what was learned, and keep a human hand on the wheel.

The ball is now in your court. Will AI become one more system that happens to communities, or a carefully governed tool that helps communities thrive? Pick the first process, gather the right people, and make your next experiment worthy of the public’s trust.

References

Department for Science, Innovation and Technology. 2025. “Landmark Government Trial Shows AI Could Save Civil Servants Nearly 2 Weeks a Year.” June 2, 2025. https://www.gov.uk/government/news/landmark-government-trial-shows-ai-could-save-civil-servants-nearly-2-weeks-a-year.

National Institute of Standards and Technology. 2023. “AI Risk Management Framework.” January 26, 2023. https://www.nist.gov/itl/ai-risk-management-framework.

National League of Cities. 2025. AI in Cities. Washington, DC: National League of Cities. https://www.nlc.org/wp-content/uploads/2025/01/AI-in-Cities-Report.pdf.

Organisation for Economic Co-operation and Development. 2025. Bridging the AI Skills Gap: Is Training Keeping Up?Paris: OECD Publishing. https://www.oecd.org/en/publications/bridging-the-ai-skills-gap_66d0702e-en.html.

U.S. Government Accountability Office. 2021. Artificial Intelligence: An Accountability Framework for Federal Agencies and Other Entities. GAO-21-519SP. Washington, DC: U.S. Government Accountability Office. https://www.gao.gov/products/gao-21-519sp.

Williams, Sarah, Sara Beery, Christopher Conley, Michael Lawrence Evans, Santiago Garces, Eric Gordon, Nigel Jacob, and Eden Medina. 2024. People-Powered Gen AI: A Responsible Approach to Generative AI in the Public Sector. Boston: City of Boston. https://www.boston.gov/sites/default/files/file/2024/07/People-Powered-Gen-AI-pre-publication.pdf.