Bet on the last mile of new technologies.
In May, New York City’s comptroller, Mark Levine, put hard numbers to a worry that has been hanging over the city’s economic landscape. His report modeled five ways, ranging from optimistic to dire, that artificial intelligence could reshape the local economy. In the worst case, the private sector could lose over 100,000 jobs by 2027, which would cost the City billions in foregone revenue.
Levine is right to focus on New York City’s unique exposure to AI. The city is focused on knowledge-intensive white-collar work, and many workers do the exact same routine knowledge tasks AI is learning to master. But the report also points out something more hopeful: Levine notes the city is already home to hundreds of firms competing to make New York “the capital of applied AI.” This is the real opportunity. New York City will not win the battle for AI by simply bracing for impact or racing to the bottom to attract firms through tax breaks for data centers. The city is unlikely to overtake San Francisco in the race to develop frontier models or Northern Virginia in being home to GPU clusters.
New York City can capture the highest-value AI jobs by becoming the key hub for what the industry calls the “application layer” for AI. This is where AI’s capabilities meet the demands of real-world industries, most of all in regulated sectors. Think using AI to reshape law, finance, medicine, real estate, media, insurance and other sectors. These are all sectors where New York City already has an unmatched level of density and agglomeration of talent, where the economic value of AI will ultimately be realized. The opportunity for New York lies in becoming the primary destination for firms seeking to build the plumbing and deployment for AI across these different domains. This is where the high-paying jobs of the future are likely to be located, and hence New York City’s future tax base.
Applications, not wrappers
The hard part about building the economy around AI applications is avoiding what technology experts refer to as the “wrapper” trap. A wrapper is a company that only provides a minimal interface bolted on top of a technology built by OpenAI or Google. These firms have multiplied since the arrival of ChatGPT: Think of the wave of “chat with your document” tools or one-click “AI assistant” applications.
These companies tend to have no real defense against future innovation. The moment a major lab improves its base model to tackle the same use case, or a hobbyist spins up a free open-source version over a weekend, the wrapper’s business model collapses.
New York’s strategy should explicitly not be about attracting or subsidizing these firms. There’s a more durable type of company that uses AI to build something genuinely valuable.
Which firms survive as the underlying AI models keep getting better and better? There are two defensible advantages: regulation and proprietary data.
Regulation sounds like a burden for firms, but here it can be a critical advantage. The best AI models can offer great legal and medical advice. But they cannot offer attorney-client privilege or prescribe a drug an insurer will reimburse. Industries built on licensing, liability and compliance have built-in natural barriers that a better AI model alone is not going to take down.
The second key advantage is having access to unique or niche data. A firm that sits on its own information AI labs cannot touch now possesses a critical advantage. These can consist of their own records, information about customers or their own history and decisions over time. These firms are in a position to build a huge advantage by putting AI to work on data competitors do not have access to.
These two advantages compound. A firm that successfully runs AI inside its own high-stakes operation throws out still more internal data in the process, which regulation then keeps out of the hands of a competitor. This lead will compound over time.
What do these opportunities look like?
Here’s how this might play out in a field like finance. Private equity firms, hedge funds and investment banks are already deploying novel AI applications that interact with their trade executions, deal decisions, risk management and regulatory reporting. These are all areas where errors have enormous consequences and deep access to proprietary internal data — like the firm’s trading history, the record of how they have accepted or turned down deal opportunities, and private client data — is the baseline cost of entry. The firms that win the AI race will combine AI with human judgment to decide on execution and catch errors, taking advantage of data no one else has. The blend of data and talent to do this exists in New York and almost nowhere else.
New York has many other centers of excellence that can follow this same format. In medicine, the city is home to many forefront medical centers, among them NYU Langone, Mount Sinai, NewYork-Presbyterian/Weill Cornell and Memorial Sloan Kettering. These hospital systems are all experimenting with ways to incorporate AI intelligently in their clinical workflows. The role of regulation here — everything from the FDA’s regulatory requirements to malpractice insurance requirements to federal privacy laws — will help ensure that these jobs and tasks are going to remain durable and not be completely replaced. Even radiology, a job many AI experts predicted would become obsolete as technology got better and better at reading X-rays, has remained resilient as medical professionals remain a credentialed specialty. Instead, these systems are likely to expand to take on new roles and tasks that require both technical and medical expertise.
Legal and compliance firms, a stronghold in the city, are another example of the potential for AI enhancement. The white-collar firms that survive will be able to build upon their existing legal datasets to map and understand legal precedents and vulnerabilities as well as tackle novel issues around cybersecurity and alignment risks. The people who can build and validate these systems should also live overwhelmingly in New York.
A common thread through these examples is the deployment of AI in high-stakes, regulated settings to generate more data and more accurate models that drive future business value. Every time a clinical AI application processes a case and receives clinical feedback, it fuels a virtuous cycle. Better deployments lead to better algorithms, which generate more feedback. This loop compounds as firms deploy more AI inside of protected, proprietary environments, which ultimately generates a deeper competitive moat. It’s a promising path for AI as a competitive advantage for New York City.
What the City should do
To some degree, AI in New York City will evolve along the lines I’ve described whether or not government does anything. So what should the City and State do?
The critical context here is that New York is looking at stagnation in well-paying white-collar jobs. As the state comptroller has shown in a recent report, nearly all job growth in New York since the COVID-19 pandemic has been concentrated in a single sector: “health care and social assistance,” which generated 253,100 net new jobs. If you remove these positions, New York City’s private-sector job growth was negative. Two of the highest-paid sectors, “finance and insurance” and “information,” combined generated just 26,000 new jobs. This makes the city strategically vulnerable around a valuable set of white-collar jobs.
Together with Stijn Van Nieuwerburgh, I have referred to the urban doom loop as the challenge of coping with dynamic responses from a smaller set of employed workers. While the full scope of AI-related impacts on white-collar employment is yet to be seen, the risk is that AI simply automates a large chunk of professional work just as we saw with the automation and outsourcing of blue-collar work, which led directly to the fiscal crisis of the 1970s.
Just as the city came out of that period with a renewed focus on new jobs and occupations (at that time, the focus was on the very white-collar professions that are now squarely in the targets of AI companies), today we should be proactive in advance of technological shifts to secure a future for novel economic tasks and occupations. Securing the AI application layer is a strategy that leverages New York’s existing advantages to ensure they remain relevant in a more AI-oriented world.
The City’s policy should be oriented around three approaches: strengthening access to domain-specific data, strengthening access to talent and space and helping institutional customers adopt new tools.
What will those strategies look like?
Data-sharing frameworks. New York City alone generates an enormous amount of structured data — building inspection reports, permit applications, transit operations, public health surveillance and so forth. While some of these data are made available through the City’s Open Data platform, much of it remains inaccessible or in formats that are challenging to use.
So the industrial policy for New York starts with the City itself. It should first make publicly accessible as much data as it can, in the form of a Model Context Protocol (MCP) server that agents can access. In layman’s terms, MCP is a bridge between AI models and external data sources, making it easy for them to talk to one another. Additionally, the City should make available other controlled-access data, modeled, for instance, on the U.S. Census Bureau’s Federal Statistical Research Data Centers, which let AI firms work with city data under privacy restrictions. This would provide the raw data for application-layer products in government technology, property technology and urban analytics.
The City should then lead the way in extending this approach to private-sector data. The City government can facilitate data-sharing agreements between academic medical centers (for clinical AI), between financial and fintech firms (for compliance AI) and between housing agencies and proptech firms. There is no need here for the City to build its own tools or models. Instead, the government can simply lower the transaction costs of data access that helps firms develop and scale their solutions. In medicine, this might mean that the City helps negotiate an agreement that lets startups train diagnostic tools on de-identified records pooled across multiple hospital systems. These are the kind of niche data that provide a critical edge, but that privacy rules alone might keep locked up in distinct silos upon which no one hospital is able to fully capitalize.
Talent programs. The growth of the application layer requires people who can combine technical fluency with domain knowledge. New York City’s universities are well placed to produce this talent, but the existing pipelines are too narrow. To prepare for future jobs, the City should pair with and nudge local universities to offer additional joint degrees, certification programs and lifelong-learning initiatives. There may also be a need for mid-career programs for displaced white-collar workers to retrain and retool themselves. Imagine a degree or certificate in AI and law or AI for clinical operations.
Application-layer firms will also need space in which to operate. Since Manhattan’s office vacancy remains elevated in the wake of the COVID-19 pandemic and the rise of the work-from-home economy, the City should help convert Class B and Class C office space into subsidized lab and incubator space for firms located near industries they serve, modeled, for instance, on the M-CORE program, which supports upgrading of commercial office space. The New York City Economic Development Corporation’s NYC AI Nexus program, which provides venture support and accelerator programs for early-stage AI firms, offers another model of how to foster collaboration and integration between AI firms, investors and others.
City procurement. Finally, there is room here to use City procurement much more intelligently. The government is one of the largest institutional buyers in the area. Committing to AI-enabled procurement in building inspection, permit review and benefits administration, among other functions, would ensure a guaranteed first customer for application-layer startups. To do this would require that the City also remove certain roadblocks and prohibitions to the deployment of AI in government and partner with employees and labor unions to do so. While challenging, being a leader itself in the deployment of AI can help push the City’s ecosystem in this direction.
Even as it helps some firms scale up, the City should work hard not to subsidize the wrong firms, the dreaded “picking of winners and losers.” A range of firms is likely to succeed in the application layer. This includes OpenAI, Anthropic and Google themselves, as they add functionality that directly connects their models to enterprise customers. It also includes the large companies that themselves find new ways to monetize existing proprietary datasets and workflows. And of course, many new startups will successfully build additional innovative AI-first strategies to grow and scale, as some will fail. The City should encourage all of these sets of firms to grow.
Regulatory frameworks. Finally, another dimension in which the City can help establish a framework for AI firms to flourish is through the development of regulatory guidelines. The City was an early mover here, releasing the AI Action Plan in October 2023 through the New York City Office of Technology and Innovation. A clear set of regulatory requirements can actually help firms facilitate further growth and innovation, if the regulations provide clear safe harbor against litigation risk. For example, if the City set a clear bias and auditing standard for the use of AI in lending and gave firms that meet that standard some protection against discrimination suits, lenders would know the rules before they build, rather than having to guess where that legal line is.
The bigger picture
The competition for AI jobs is sometimes viewed as a frantic fight to pay billion-dollar salaries for the very highest-paid talent in frontier labs, or an attempt to build wrapper models on top of foundational models. Neither approach is sustainable for a city of the scale and depth of New York. There will ultimately only be a few thousand AI researchers working to build the latest frontier models, but in doing so, they will continue to disrupt the wrapper firms.
The potential for New York is instead to focus on the firms with the best strategies to meaningfully deploy AI in the sectors of finance, health care, law, insurance, real estate and media. The regulatory restrictions and existing data moat will ensure that firms that succeed in these areas are going to require hybrid expertise of domain as well as institutional knowledge. This is the sweet spot in which New York City is well poised to excel.
In seizing this opportunity, New York has the potential to remain a global center for white-collar and professional work. The AI economy is unlikely to be geographically distributed the same way the pre-AI knowledge economy was. If foundational models continue to absorb routine cognitive tasks, the work that remains will concentrate in a handful of cities that already have the institutional density to support it. There may only be room for a few such hubs globally. New York is the natural candidate to be one of them, if we play our cards right.
