> ## Content Index
> Fetch the complete content index at: https://www.vitalcitynyc.org/llms.txt
> Use this file to discover other available public pages before exploring further.

# Between Hype and Reality: What AI Can and Can’t Fix in City Hall
- URL: https://www.vitalcitynyc.org/ai-city-government-services-hype-vs-reality/
- Published: 2026-11-18T17:00:00.000Z
- Updated: 2026-11-18T17:47:50.000Z
- Description: Local governments are already putting AI to work, but that progress is only as strong as the foundation of workers and operations that it stands on.
- Author: Tiffany Chu
- Tags: #ai-and-the-city, Artifical Intelligence, City Government, The Big Reboot

In 1970, if you wanted to file a complaint about a pothole, you went down to a municipal office, spoke with a frontline worker and filed some paperwork. Shortly after, they’d follow up with a phone call. The process was generally the same for food stamps, welfare programs and any other administrative service the government provided. When it worked, it was highly human, and highly personal. 

In the following decades, we’ve moved many paper-based processes onto computers in the name of efficiency and modernization. We replaced counters with websites, paper forms with PDFs and databases, and one-off phone calls with 311 systems. Cities launched open-data portals — for example, [New York City’s offerings](https://www.nyc.gov/opendata?ref=vitalcitynyc.org) — and redesigned websites, while technical infrastructure aged and workforces shrank. Slow-moving, incremental improvements in service delivery have been approached as one-time capital projects, not as continuously improving systems. What was new is already becoming technical debt, and more people than ever are churning out of critical services because of missed administrative steps ([procedural disenrollment](https://www.linkedin.com/pulse/could-llms-reduce-procedural-disenrollment-snap-program-rouault-nxqdc/?ref=vitalcitynyc.org)), not ineligibility.

In 2026, the government is doing even more with less. The workforce is graying precisely at the moment when [personal agents](https://ai.engineer/topics/personal-agents?ref=vitalcitynyc.org) are starting to flood government services. [Roughly 38% of local government employees are expected to retire](https://www.berrydunn.com/news-detail/the-silver-tsunami-and-the-future-of-local-government-advice-for-a-resilient-workforce?ref=vitalcitynyc.org) within five years. Dedicated public servants struggle with vacancies — [for every person hired, about four jobs remain open](https://www.nlc.org/article/2024/10/21/attracting-and-retaining-talent-in-city-government-become-a-great-place-to-work/?ref=vitalcitynyc.org), a number that is 166% higher than the private-sector ratio. Though AI is transforming many [other industries in New York](https://pfnyc.org/research/artificial-intelligence-is-already-reshaping-new-york-city?ref=vitalcitynyc.org), basics like processing time and other administrivia still bog down the delivery of government services to residents. While large language models (LLMs) have potential to dramatically improve the user experience of interacting with government services, that potential is not yet realized. Simultaneously, city services, like the rest of the internet, must now cope with a high volume of non-human web traffic and [AI-written requests](https://techcrunch.com/2026/09/10/ai-agents-are-flooding-public-services-with-new-requests/?ref=vitalcitynyc.org). The term “[agentic flooding](https://arxiv.org/pdf/2608.16603?ref=vitalcitynyc.org)” describes how municipal services may struggle to field a high volume of requests as non-human web traffic increases steadily. As of August 2026, [50% of internet traffic is non-human](https://cpl.thalesgroup.com/about-us/newsroom/2025-imperva-bad-bot-report-ai-internet-traffic?ref=vitalcitynyc.org), with a more than 1,700% increase from agents specifically. 

### A tool, not a magic wand

In the near term, technology like LLMs won’t transform government overnight, but it can help us catch up. As former Baltimore Chief Data Officer Justin Elsasz [recently said](https://govex.jhu.edu/podcast/episode-109/?ref=vitalcitynyc.org), *“There is so much to do. AI is just a way for us to break even on workload for cities.”* 

As an example, Boston faced a daunting backlog of 13,000 open sidewalk repair cases plagued by poor data quality and mismatched numbers across departments and systems. Chief Information Officer Shin-pei Tsay’s team used AI to reconcile and resolve cases, identify outstanding issues, prioritize work and start clearing the backlog. (Disclosure: One of us, Tiffany, was chief of staff to the mayor when this effort began.) That was essential. But the harder task remains: Public Works, Neighborhood Services and Budget are working to design, test, agree on and follow a new workflow to handle [$800 million worth of capital requests](https://www.boston.gov/departments/public-works/sidewalk-repair-program?ref=vitalcitynyc.org#faq), assess exactly what kind of repair is needed, and prioritize resources against competing neighborhood demands — something AI can’t do today, and that we probably won’t trust it to do anytime soon.

In Denver, a multimillion-dollar budget shortfall led to hiring cuts and reduced staff hours in its 311 (non-emergency) call center. To adapt, the city launched Sunny, an AI-powered chatbot aimed at answering common informational questions (e.g., library hours or trash pickup). Sunny now handles about 20% of transactions, allows residents to call 24/7 and frees up staff for complex cases. That said, the [Denver Auditor report on 311](https://www.denvergov.org/files/assets/public/v/3/auditor/documents/audit-services/audit-reports/2025/denver-311%5Faugust-2025.pdf?ref=vitalcitynyc.org) found that the coordination between Sunny and city departments remained residents’ biggest complaint. While Sunny could create tickets faster, residents still did not know how long resolution would take. The auditor’s key recommendation was written service-level agreements to set expectations for timelines and accountability (“a pothole will get fixed within 72 hours”) — a challenge for numerous cities long before AI’s arrival. 

This reinforces a lesson that we have learned from previous modernization projects: when you speed up a part of a process that wasn’t the bottleneck to begin with, the core issue still remains.

But there are clear ways AI can deliver genuine leaps in performance.

Cities have a lot of data and information that is inaccessible or unusable for everyday people that AI can help unlock. 

For example, the Boston City Council votes on hundreds of items in meetings every year. The City Clerk gives each item a four-digit docket number, which is hard to remember, and records all council votes using dry, formal legislative language, which is confusing to understand. To promote transparency, Boston used AI to translate 19 years of council meetings into [concise, 20-word, plain language summaries](https://www.boston.gov/departments/city-council/city-council-roll-call-votes?ref=vitalcitynyc.org) that residents could easily read and see how individual councilors voted. Public records only create accountability when people can actually read them, and a vote no one can decipher is a vote no one can hold a councilor to.

Washington, D.C.’s municipal government has 1,500 open datasets, but most residents don’t have the technical skills to use the raw data. [DC Compass](https://opendata.dc.gov/?ref=vitalcitynyc.org) is an AI-powered assistant within the city’s open data portal that helps residents use natural language to get their questions answered directly, search and summarize data, and create maps, instead of drowning in large, open but impenetrable datasets. 

> When you speed up a part of a process that wasn’t the bottleneck to begin with, the core issue still remains.

While making information more accessible has become easier, linguistic translation at scale is still hard. Cities often pride themselves on providing translation for their website and materials into numerous languages — Boston supports [11](https://www.boston.com/news/local-news/2022/10/04/boston-311-app-11-languages-update/?ref=vitalcitynyc.org) through its 311 app; Denver [72](https://www.denvergov.org/Sunny?ref=vitalcitynyc.org) through its AI chatbot; New York City, [175](https://www.nyc.gov/content/oti/pages/testimony/nyc311-testimony-before-nyc-council-committee-on-technology?ref=vitalcitynyc.org) via live phone interpretation. Languages using the Latin alphabet are typically more reliable, but once they use other alphabets (Arabic, Mandarin, etc.), the quality declines and residents can lose trust. City governments are often trying to communicate complex rules and policies, and the stakes for people and businesses can be very high. Putting simple trust in AI is not the answer, at least not yet.

AI also struggles to reflect humans’ social judgment. In Oklahoma, a state law changed in 2024 to allow municipalities to remove discriminatory language from plats — documents and maps that show how land is divided into parcels and governs its development — such as historic racially restrictive covenants. With over 6,000 plats in Oklahoma City, many residents suggested using AI to make the process more efficient. After deliberating on issues such as how to extract record data from across multiple counties, and how to train the model on what language is racist and isn’t, Chief Innovation Officer Kelly Williams and her team decided that “there’s no magic easy button.” The city later issued [an update](https://www.okc.gov/News-articles/OKC-plat-amendment-project-moves-forward?ref=vitalcitynyc.org) that “for a project of this size, human reviewers remain the most efficient and reliable option.” While government is known for countless rules and regulations, the reality is that many decisions are made (how should this intersection be designed; where should our next school be built), importantly, with human judgment.

This is not to throw cold water on every bot. Indeed, AI can do impressive things and help teams leapfrog when used in the right way.

In Rome, garbage fires have long plagued the city’s outer suburbs. Recently, [their waste management agency, AMA,](https://kurrant.com/kurrantly-news/rome-turns-to-satellites-and-ai-to-catch-illegal-dumping-under-esa-ama-pilot/?ref=vitalcitynyc.org) found they could use computer vision and AI at three levels — satellites, drones and mobile cameras on trucks — to quickly spot debris accumulation and take action before a fire erupted. Officials also developed a predictive fire model using 10 variables like wind, crime and previous incidents. According to AMA’s Head of Commercial and Industrial Waste Management Giuseppe Morone, since 2023, trash fires across Rome are down by 33%. AI use will need to thoughtfully balance concerns around surveillance and privacy, but it’s also clear that these tools unlock new ways to tackle seemingly intractable problems. Whether it’s trash reduction, potholes, streetlights — there are numerous on-the-ground issues in the biggest U.S. cities that could be leapfrogged with AI. From former Boston CIO [Santi Garces](https://datasmart.hks.harvard.edu/agentic-ai-city-hall?ref=vitalcitynyc.org): “What can we do now that we could never do before?” 

[As Conway’s law tells us](https://en.wikipedia.org/wiki/Conway%27s%5Flaw?ref=vitalcitynyc.org), cities often design around their organizational charts rather than residents’ needs. Boston, like many cities, had a notoriously maze-like, cross-departmental process for permitting. Using AI to analyze 25 years of records (roughly 250,000 permits), the Permitting team identified and ranked 320 permit types, finding that bathroom and kitchen renovations were most common, all for about $20 in tokens. Reorganizing permitting around what people are actually trying to do, rather than which department owned which form, sounds obvious but transformed the approach: years of debate over where to start became a three-month prototype, paired with in-house product managers who broke the project into modular pieces instead of one massive procurement. The new platform is still rolling out, but already city residents can [search by intent](https://www.boston.gov/boston-permitting?ref=vitalcitynyc.org#what-are-you-trying-to-do) (“open a restaurant”) instead of having to hunt down specific approvals they need and get routed from department to department. 

> Public records only create accountability when people can actually read them, and a vote no one can decipher is a vote no one can hold a councilor to.

### What it all adds up to

Countless [articles](https://www.gao.gov/blog/federal-efforts-update-old-it-are-years-behind-schedule-we-looked-impacts-delays?ref=vitalcitynyc.org) and [studies](https://www.sciencedirect.com/science/article/pii/S2664328624000792?ref=vitalcitynyc.org) show how far government IT has fallen behind, how government services remain costly and difficult to use, and how attempts to modernize legacy systems have failed, and how vulnerable these aging systems are to cyberattacks. We have a long, long way to go. Teams like [NYC’s Public Interest Technology crew](https://www.nyc.gov/content/pitcrew/pages/?ref=vitalcitynyc.org) are starting to address the return on investment on these IT investments and pursue strategies to fundamentally rethink how we deliver value to our residents, but [even they](https://www.vitalcitynyc.org/mamdani-pit-crews-government-tech-problems/) are [seriously limited](https://www.vitalcitynyc.org/better-government-technologies-strategies-nyc/). 

LLMs have the potential to reshape the way the public interacts with government services and the way those services are bought or built. In the international development space, this shift is referred to as “technological leapfrogging,” where an entity skips intermediate stages of development to catch up to current practice. One [classic example](https://www.allianzcare.com/en/employers/business-hub/hr-blogs/ngo-leapfrogging-bridging-the-technological-divide.html?ref=vitalcitynyc.org) is how some areas in sub-Saharan Africa never deployed landline telephone networks, but have cellphones today. However, only if we are simultaneously mindful of AI’s power and its limitations can we harness its potential to truly transform government. 

For instance, it has never been easier or cheaper to build a website or translate content into multiple languages. Imagine a world where information is instantly translated into the language someone speaks at home, where questions are answered verbally in real time, at 10:30 p.m. after your shift ends, on a Saturday. Public-sector staff could spend more time designing or delivering better programs, improving policies, measuring efficacy of those programs and making actual judgments in their area of expertise instead of solving last-mile problems. 

As cities around the world start to publish their policies and guardrails around using AI, we propose four principles for local governments to think through AI for city services.

1. **The need drives the use.** Start with what residents need, not what AI can do. In the world of AI, the role of government is to ensure that technology serves the public interest, and not the other way around.
2. **Augment and redirect human workers, don’t replace them.** AI tools can make workers more productive, but without knowledgeable supervision they can fail — and sometimes produce even larger messes than individuals do when they make mistakes.
3. **Set guardrails, not roadblocks.** Some cities are so worried about AI they are [broadly banning its use](https://www.nyc.gov/mayors-office/news/2026/09/mayor-mamdani-and-chancellor-samuels-put-students-first-with-nat??ref=vitalcitynyc.org). That’s unwise. Enable [responsible experimentation](https://www.vitalcitynyc.org/nyc-public-school-classroom-ai-policy/) instead of focusing on restrictions. Test with pilots, then build for scale, and keep accountability with the human.
4. **More transparency leads to more trust.** If allowed to stay in a black box, AI can increase confusion and frustration. Instead, the tech can be an opportunity to build trust through [operational transparency](https://hbr.org/2019/03/operational-transparency?ref=vitalcitynyc.org): share useful disclosures, documentation and risks. Rules are important, but trust is built by seeing what the government actually does, communicated plainly.

[Ready or not](https://govex.jhu.edu/ai-capacity-quiz/about/?ref=vitalcitynyc.org), AI tools are changing city governments within bureaucracy and writ large in society. But they can’t magically dispel existing, deep-rooted problems like poorly designed processes or atrophied infrastructure. As with every new technology, progress depends on a deep understanding of these systems as they are, and practicing people’s judgment about when, where and how to apply it.