San Francisco’s big swing at putting artificial intelligence on thousands of city desktops has landed with a mixed thud. Some teams swear the new tools are saving serious time, a few power users are racking up eye-popping activity, and a sizable chunk of the workforce barely touches them. What started as a promise of across-the-board efficiency has turned into a tricky test of how a self-proclaimed AI capital actually governs the tech in its own house.
As reported by the San Francisco Chronicle, the city rolled out Microsoft Copilot to roughly 30,000 employees across more than 40 departments beginning in July 2025. Between July 2025 and April 2026, city records show workers exchanged more than 1 million messages with Copilot. The Chronicle also found that Copilot interactions climbed by more than 200% in that period, even as the share of employees using the tool held steady at around half of the workforce.
Training and early returns
The city did not flip the AI switch overnight. Before Copilot became widely available, about 2,000 staff participated in a six-month ChatGPT pilot, and OpenAI hosted eight training sessions that drew more than 3,000 attendees overall. According to a report by the San Francisco Civil Grand Jury, 70% of survey participants in that pilot said they saved up to five hours per week using ChatGPT.
Who’s actually using Copilot
Copilot activity has clustered in white-collar corners of city government rather than in frontline fieldwork. Since December, the Department of Police Accountability has posted the highest Copilot use rate among major departments, while the Department of Public Health has logged the largest raw number of interactions. According to the Chronicle, analysts, managers and attorneys in desk-based roles tend to be the most frequent users, while firefighters and patrol officers show much lower uptake. One Port collections officer alone recorded more than 1,200 Copilot interactions in a five-month stretch.
Rules and red lines
City officials have tried to keep some guardrails in place as use grows. San Francisco’s Generative AI guidelines, updated in July 2025, require employees to review and fact-check AI-generated material before it is used in public-facing or sensitive work. The rules explicitly prohibit creating images, audio or video that could be mistaken for real people. The San Francisco Generative AI Guidelines also tell departments to disclose AI use in certain communications and to list approved tools in the city’s Chapter 22J inventory.
City response and roadmap
In its official reply to the Civil Grand Jury, the mayor’s office said it agrees with many of the jury’s findings, backs the Copilot rollout, and stated that work is underway to consolidate AI procurement and develop a formal citywide AI strategy, according to the city’s official responses. Earlier coverage of Hoodline highlighted the administration’s early claims about time savings, and city leaders now say governance and training will be central in the next phase.
What to watch next
Planning and budget documents lay out the near-term AI agenda. Officials expect to publish a citywide AI strategy, create a try-before-you-buy experimentation framework, and build out both a Citywide AI Workspace and an AI Champions program through 2026. Those projects appear in the city’s budget and ICT planning materials as the administration seeks to align broader access to AI tools with clearer ground rules and more structured procurement processes.
