AI to Shred Government Paperwork
Stuart Franklin / Magnum Photos

Employees need not get bogged down by red tape.

The impacts of data centers cannot be dismissed, and the need for regulation of AI frontier companies is imperative as their technology gets ever more powerful. Still, the AI tools that are currently available offer tremendous potential to make New York City government work better and faster. While a lot of potential uses exist, one the Mamdani administration should rush to embrace is quite simple: using AI to streamline the many internal and interagency forms and documents that too often determine how quickly New York City government functions.

When the public thinks about bureaucratic slowness, they unsurprisingly think of issues where they themselves are on the receiving end: the time it takes to respond to 311 complaints, to issue a permit, to make government decisions.

Invisible to the public is the bulk of government paperwork: the internal processes, usually involving forms or documents, that shape how quickly government works. These are easy to dismiss as “check the box” exercises, which they are, but they serve valuable purposes: ensuring that a vendor or potential new hire has the right qualifications for a job, ensuring that transparency rules have been followed, protecting the City from liability, and ensuring that contracts are written fairly.

These internal approvals are a source of massive delay — ranging from weeks to several months — which is not a surprise when you consider the number of people and agencies that get involved even in some relatively simple efforts. The front-line work of city government is done by the operating bureaus of city agencies, ranging from a police precinct to one of the Department of Education’s schools to the Department of Environmental Protection’s Bureau of Water Supply. These are the places where the money is spent, the contractors and staff are hired, and bills are paid. When one of them needs to do something, they have to submit paperwork to another division of the same agency: the human resources office to hire someone, the procurement office (and legal counsel) to buy something, the audit office to pay a contractor. 

Then, once that office approves it, the paperwork goes to another agency: the Office of Management and Budget (OMB) and the Department of Citywide Administrative Services (DCAS) to approve and process a hire, the Law Department and the Mayor’s Office of Contract Services (and then the Comptroller’s office) to approve a contract, OMB again to pay a bill. Simply keeping track of where each document sits is a big challenge; when I was at DEP, we’d keep data on this both to understand where we were and, when necessary, to complain to other agencies when they were causing delay.

At each of these steps, if a piece of information is missing or unclear, the paperwork is sent back. Sometimes it’s because there’s a real problem: an unqualified vendor or hire, an incomplete invoice. That’s government doing its job; you don’t want those forms waved through.

More often, though, it’s because the document was incomplete, poorly worded or didn’t meet the standards of whoever had to approve it next. Either way, the delays are huge: Often the person or team whose work was rejected doesn’t understand the problem (no doubt they tried earnestly to get it right, but these are often complex), and then they need to go back and redo the paperwork and get it in the review queue again. I remember times when we struggled to understand OMB’s reason for rejecting a form, and when we found that different people at a given agency had slightly different standards. Weeks, sometimes months, can be lost this way.

And that, in turn, creates a real cost to New Yorkers: jobs left unfilled, vendors waiting to be paid, projects delayed, equipment and supplies that don’t arrive in time. All of that leads to lower-quality government.

The goal is not to replace human government with government by AI, but to use AI to check forms before they are submitted.

The good news is that there are patterns here — records that can be used to train an AI tool to spot mistakes and omissions before they become a wrench in the gears. While there are lots of delays and failed documents, there are also thousands of successful ones: every City employee is the result of a successful hire memo, every City contract the result of a successful procurement.

Mayor Mamdani has spoken repeatedly about the need for government to function better; from Public Interest Technology (PIT) Crews to his Commission on Government Efficiency, he seems eager to explore better ways of doing things and to hold agencies accountable. For example, he’s reimposed a deadline on how long OMB can take to review hiring proposals from agencies, one that the Adams administration also imposed but which was never really followed by OMB. But such deadlines require action, not approval, and errors and incomplete forms will still bog things down.

Here’s one straightforward way AI could help this agenda.

The mayor should ask every agency to catalog the internal approval processes where delays occur, either internally or where they face delays with other agencies’ approvals. Then he should task each approving agency or office with delivering to the Office of Technology and Innovation (OTI) a complete set of forms or applications that it had received for that process in the last, say, three years — both those approved and those rejected. He should task OTI with taking those documents and training an AI tool to figure out what a successful form looks like, so it can prescreen all forms meant to accomplish a given task. 

The goal is not to replace human government with government by AI, but to use AI to check forms before they are submitted. It’s basically a well-trained version of what lots of people already use AI for: “Think like the person who will receive this draft document, and tell me how you’d react to it.”

Just as spell-check isn’t infallible, the tool will not be either — but, just as spell-check does catch errors and typos, a system like this should improve the success rate of document approvals. If we assume that each rejected form costs several days to several weeks’ delay, across thousands of annual uses that could add up to a meaningful improvement in the overall speed with which city government functions.

Those concerned about AI will have two fears related to this. One is privacy; many of these forms contain personal or privileged information. OTI should have agreements that protect the City’s information from being used to train agents beyond those the City is creating; this is standard for large corporate AI accounts.

The second is whether this creates a slippery slope that could replace civil servants with AI tools. I’d argue that’s unlikely because — as with the spell-check example — there are likely to be exceptions and novel situations that the tool won’t recognize. That’s what experience is for, and where New York City’s many skilled and dedicated civil servants really provide value, not in filling out paperwork but in using their judgment, experience and understanding of the public’s needs to provide services.


Great! You’ve successfully signed up.

Welcome back! You've successfully signed in.

You've successfully subscribed to Vital City.

Success! Check your email for magic link to sign-in.

Success! Your billing info has been updated.

Your billing was not updated.