What Do We Want to Make Better?

When I was chief innovation officer in San Francisco’s mayor’s office, part of my work was helping staff in other departments find better ways to serve the public. I had my own responsibilities as a city employee, too. Across both, I saw the same expectations: people should receive an answer they can understand, get help without unnecessary delays, and be served by public servants who have enough time to exercise sound judgment.

Those expectations have stayed with me. They are a large part of why I care about AI in government.

Today, working in AI, I spend a lot of time thinking about what the technology makes possible. My experience in government gives me a practical way to assess those possibilities: would this help an employee do their job better, and would the person they serve notice the difference?

Consider someone trying to find out whether they qualify for a public service. They read the requirements but aren’t sure how a rule applies to their household. They call for help, leave a message, and wait. When they finally speak to someone, they learn that they need another document.

I want that person to understand what they qualify for, what they need to provide, and what happens next. AI could help explain requirements in plain language or prepare a checklist based on their circumstances. The explanation would need to match the agency’s rules, and the person would need a way to reach staff if it was wrong or their situation was complicated.

I think just as much about the employee answering the question. They may need to check a policy, locate a record, or ask another department how it handles an exception. Each step takes time, and other people are waiting for help.

An AI tool that finds the relevant guidance and shows its source could help that employee respond sooner. A summary of the case could reduce the time they spend searching through records. The employee would still need to check the information and decide how to proceed, but they could begin with more of what they need in front of them.

It could also give staff more time for cases that require judgment. If someone’s documents conflict or their circumstances do not fit the usual process, the employee needs time to investigate, ask questions, and explain the options. I would want AI to help make that time available.

Efficiency matters, particularly when resources are limited. But I want us to be specific about what we hope to do with the time saved. Fewer repeat calls, fewer applications returned for missing information, and more time for complicated cases would tell me more than how many people had started using an AI tool.

My experience in the mayor’s office taught me to start with a problem people needed solved. Through Civic Bridge, we brought outside partners together with city staff to work on challenges the city had identified. Staff understood the service and its constraints. The partners needed to learn about both before proposing changes.

I want to take the same approach with AI. If people repeatedly misunderstand a requirement, we should find out why. Perhaps the instructions need rewriting. Perhaps departments are giving conflicting answers. AI might help, but we should understand the cause before choosing a tool.

For me, this is also about trust. When someone receives a clear answer, knows what to do next, and gets help when something goes wrong, they have a reason to believe government can serve them well. AI is worth exploring where it can help employees deliver that experience more consistently.

That brings me to the capacity to put AI to work. Agencies need people who can test these tools against real questions, check the answers, and help staff use them. That work is worthwhile when it leads to a better service.

When I think about AI in government, I picture the colleagues I worked alongside. I want them to spend less time searching for information and more time helping someone understand their options or resolve a difficult problem. That mattered to me when I worked in city government. It still does.

Written with drafting and editing assistance from Codex. The experiences and final editorial decisions are my own.

I work at OpenAI. The views expressed here are my own and do not necessarily reflect those of my employer.