Before the mayor of San Jose, California, arrives at a ribbon-cutting ceremony for a new business, his aides ask ChatGPT to help draft some talking points.
“Elected officials do a tremendous amount of public speaking,” said Mayor Matt Mahan, whose recent itinerary has taken him from new restaurant and semiconductor startup openings to a festival of lowriding car culture.
Other politicians might be skittish admitting a chatbot co-wrote their speech or that it helped draft a $5.6 billion budget for the new fiscal year, but Mahan is trying to lead by example, pushing a growing number of the nearly 7,000 government workers running Silicon Valley’s biggest city to embrace artificial intelligence technology.
Mahan said adopting AI tools will eliminate drudge work and help the city better serve its roughly one million residents.
He’s hardly the only public or private sector executive directing an AI-or-bust strategy, though in some cases, workers have found that the costly technology can add hassles or mistakes.
“The idea is to try things, be really transparent, look for problems, flag them, share them across different government agencies, and then work with vendors and internal teams to problem solve,” Mahan said in an interview. “It’s always bumpy with new technologies.”
By next year, the city intends to have 1,000, or about 15%, of its workers trained to use AI tools for a variety of tasks, including pothole complaint response, bus routing and using vehicle-tracking surveillance cameras to solve crimes.
One of San Jose’s early adopters was Andrea Arjona Amador, who leads electric mobility programs at the city’s transportation department. She has already used ChatGPT to secure a $12 million grant for electric vehicle chargers.
Arjona Amador set up a customized “AI agent” to review the correspondence she was receiving about various grant proposals and asked it to help organize the incoming information, including due dates. Then, she had it help draft the 20-page document.
So far, San Jose has spent more than $35,000 to purchase 89 ChatGPT licenses — at $400 per account — for city workers to use.
“The way it used to work, before I started using this, we spent a lot of evenings and weekends trying to get grants to the finish line,” she said. The Trump administration later rescinded the funding, so she pitched a similar proposal to a regional funder not tied to the federal government.
Arjona Amador, who learned Spanish and French before she learned English, also created another customized chatbot to edit the tone and language of her professional writings.
With close relationships to some of the tech industry’s biggest players, including San Francisco-based OpenAI and Mountain View-based Google, the mayors of the Bay Area’s biggest cities are helping to promote the type of AI adoption that the tech industry is striving for, while also promising guidelines and standards to avoid the technology’s harms.
San Francisco Mayor Daniel Lurie announced a plan Monday to give nearly 30,000 city workers, including nurses and social workers, access to Microsoft’s Copilot chatbot, which is based on the same technology that powers ChatGPT. San Francisco’s plan says it comes with “robust privacy and bias safeguards, and clear guidelines to ensure technology enhances — not replaces — human judgment.”
San Jose has similar guidelines and hasn’t yet reported any major mishaps with its pilot projects. Such problems have attracted attention elsewhere because of the technology’s propensity to spew false information, known as hallucinations.
ChatGPT’s digital fingerprints were found on an error-filled document published in May by U.S. Health Secretary Robert Kennedy Jr.’s “Make America Healthy Again” commission.
In Fresno, California, a school official was forced to resign after saying she was too trusting of an AI chatbot that fabricated information in a document.
While some government agencies have been secretive about when they turn to chatbots for help, Mahan is open about his ChatGPT-written background memos that he turns to when making speeches.
“Historically, that would have taken hours of phone calls and reading, and you just never would have been able to get those insights,” he said. “You can knock out these tasks at a similar or better level of quality in a lot less time.”
He added, however, that “you still need a human being in the loop. You can’t just kind of press a couple of buttons and trust the output. You still have to do some independent verification. You have to have logic and common sense and ask questions.”
Earlier this year, when OpenAI introduced a new pilot product called Operator, it promised a new kind of tool that went beyond a chatbot’s capabilities. Instead of just analyzing documents and producing passages of text, it could also access a computer system and schedule calendars or perform tasks on a person’s behalf. Developing and selling such “AI agents” is now a key focus for the tech industry.
More than an hour’s drive east of Silicon Valley, where the Bay Area merges into Central Valley farm country, Jamil Niazi, director of information technology at the city of Stockton, had big visions for what he could do with such an agent.
Perhaps the parks and recreation department could let an AI agent help residents book a public park or swimming pool for a birthday party. Or residents could find out how crowded the pool was before packing their swim clothes.
Six months later, however, after completing a proof-of-concept phase, the city didn’t buy a full license for the technology due to the cost.
The market research group Gartner recently predicted that over 40% of “agentic AI” projects will be canceled before the end of 2027, “due to escalating costs, unclear business value or inadequate risk controls.”
San Jose’s mayor remains bullish about the potential for these AI tools to help workers “in the bowels of bureaucracy” to rapidly speed up their digital paperwork.
“There’s just an amazing amount of bureaucracy that large organizations have to have,” Mahan said. “Whether it’s finance, accounting, HR or grant writing, those are the kinds of roles where we think our employees can be 20 (to) 50% more productive — quickly.”
