AI is not just a theoretical disruptor. It's already here—subtly augmenting some roles, quietly automating others, and rapidly altering the calculus of employment, wages and urban economic resilience.

”This Report Was Not AI-Generated” — A Wake-Up Call for Urban Labor Economies in the Age of Generative AI

by ChatGPT-4o

Introduction

In its August 2025 publication, This Report Was Not AI-Generated, the New York City Comptroller’s Office delivers a measured but data-rich analysis of artificial intelligence’s impact on the labor market, particularly within New York City. Building on proprietary usage data from Anthropic’s Claude and cross-referencing with U.S. Bureau of Labor Statistics and O*NET, the report is both a socio-economic snapshot and a strategic alert for policymakers, employers, educators, and citizens.

At a time when generative AI is reshaping not only how work is performed but also which jobs remain viable, this report underscores a central reality: AI is not just a theoretical disruptor. It is already here—subtly augmenting some roles, quietly automating others, and rapidly altering the calculus of employment, wages, and urban economic resilience.

Most Surprising Findings

1. AI Usage Is Highly Concentrated in a Few Occupations
   Nearly 37% of all Claude-related work conversations originated from Computer and Mathematical occupations, a group that only comprises 3.4% of U.S. workers. This overrepresentation factor of 10.85 reveals a dramatic skew that most public discussions on AI diffusion tend to understate.

2. Generative AI Use Is Correlated with Higher Wages
   Contrary to fears of AI hollowing out white-collar work indiscriminately, the data suggests that higher-income occupations are the most engaged with generative AI, both in augmentation and automation. Workers making $100,000+ in New York were more likely to use Claude than those below the median.

3. Claude Is Used More to Augment Than Automate Tasks
   While 57% of tasks involved some automation, 65.5% showed augmentation, with users working collaboratively with AI tools rather than offloading work entirely. Only 19% of tasks were majority-automated.

4. Occupational Claude Usage Depth Is Still Shallow
   Even in AI-heavy sectors like education and IT, only a minority of tasks within each occupation were being performed with AI assistance. For instance, only 4% of all detailed occupations used Claude for three-quarters or more of their tasks.

5. Job Growth Is Occurring in Low-AI-Usage Occupations
   Healthcare Support, Management, and Healthcare Practitioners—occupations with minimal AI uptake—are the only sectors experiencing headcount growth in NYC since 2019.

Most Controversial or Politically Charged Statements

1. Tech Industry’s Attempt to Block State AI Laws
   The report references a Senate-struck provision in President Trump’s budgetthat would have barred states from enacting AI regulations for 10 years. Supported by OpenAI's CEO but opposed by Anthropic’s CEO Dario Amodei, this sparked debate on whether AI governance should be centralized or decentralized—a fundamental question about federalism and innovation.

2. Shopify’s AI-First Hiring Policy
   Shopify’s CEO has reportedly begun requiring managers to prove a job can't be done by AI before new hires are approved. This startling policy has implications for employment rights, HR ethics, and economic inequality.

3. Wall Street Interns Replaced by Claude
   Anthropic’s new Claude-based financial analysis tool is already replacing entry-level Wall Street positions, accelerating the AI-induced decline in early-career white-collar roles.

4. Disparity in Claude Usage vs. Claude Impact
   The paradox is clear: Claude is least used in fields like Healthcare Support and Legal—yet those are sectors with high employment and potentially high AI disruption risk in the near future.

Most Valuable Insights

1. City-Specific Workforce Vulnerabilities
   NYC has a unique mix of overrepresented AI-exposed and AI-insulated occupations, making it particularly vulnerable to both job displacement and skill mismatches. This includes:

   • High exposure: Arts, Education, IT

   • Low exposure but high employment: Healthcare, Legal, Office Admin

2. Task Granularity Shows Real AI Penetration Levels
   The use of O*NET’s 19,530 tasks allows a nuanced understanding of AI's reach—not just across occupations but within them. This task-level analysis is essential for retraining policy and workforce adaptation.

3. AI Mentions in Job Postings Don't Equate to Growth
   Even as AI mentions rise, jobs in AI-heavy sectors like Software Development and Arts are not recovering to pre-pandemic levels, suggesting either displacement or a collapse in demand.

4. AI Uptake Is Not (Yet) Synonymous with Full Integration
   The report’s key takeaway is that while AI is everywhere, deep, widespread occupational integration is still rare. This lag presents a policy window for proactive planning.

My Perspective: A City on the AI Precipice

New York City, long a beacon of human talent and industrial dynamism, is now a testing ground for AI’s economic realignment. What makes this report powerful is its grounding in granular data, not just speculative narratives. Yet it also reflects institutional caution: most of the findings are descriptive, with little normative guidance.

From a strategic viewpoint, this report reflects a tension between adoption and absorption. AI is being adopted in selective high-value roles, but the broader absorption across occupations is partial, cautious, and unequal. This spells trouble for young workers, especially those without access to AI-relevant education or tools.

The fact that high-AI-use occupations are stagnating in headcount, while low-AI-use sectors grow, should give business leaders pause. It suggests AI is replacing growth, not augmenting it, at least in the short term. And yet, it is precisely during this “pause” that regulators and employers must prepare for the tidal wave to follow.

Recommendations

For Regulators

1. Enact Occupation-Specific AI Oversight
   Rather than one-size-fits-all rules, sectoral regulation of AI usage (e.g. legal, education, finance) can target risks where augmentation may slip into silent automation.

2. Reject Blanket Preemption of State Laws
   The attempt to preempt state-level AI regulation for a decade is anti-democratic and harmful to innovation governance. States must retain authority to tailor responses to local labor markets.

3. Invest in Real-Time Occupational Impact Monitoring
   Develop systems to continuously track AI usage by task and occupation, modeled after the Anthropic-O*NET approach.

For AI Developers (e.g. OpenAI, Anthropic)

1. Expand Transparency Tools Like Clio
   Make it easier for researchers, labor departments, and educators to understand how generative AI is used at a task level, without compromising user privacy.

2. Support Open Labor Impact Forecasting Models
   Collaborate with governments to develop publicly accessible models that predict job/task displacement likelihood by geography and skill.

3. Offer Free AI Literacy Tools for Entry-Level Workers
   Democratize AI literacy with free training tailored to those at highest risk of being left behind—the young, underpaid, and underrepresented.

For Large Businesses

1. Avoid “AI First” Hiring Practices Without Oversight
   Shopify’s model may become popular but should be tempered by ethical hiring policies that recognize the long-term costs of workforce hollowing.

2. Map AI Usage to Wage Inequality
   Assess how AI tool usage correlates with internal wage data—are low-wage roles being denied access to AI augmentation tools?

3. Prioritize Augmentation Over Replacement
   Incentivize internal use cases where AI enhances productivity but retains human oversight and final decision-making.

For Educational Institutions and Workforce Boards

1. Incorporate Task-Based AI Skills into Curricula
   Teach students not only how to use AI tools but which tasks are most amenable to augmentation, and how to recognize automation creep.

2. Realign Job Training Programs to Claude-O*NET Mapping
   Use the Claude-O*NET framework to design job training aligned with durable, AI-resilient skills.

3. Emphasize Creativity, Judgment, and Physical Skills
   Since these are the least represented in AI interactions, they should become the cornerstone of AI-era human capital strategies.

Conclusion

“This Report Was Not AI-Generated” is more than a title; it’s a declaration of urgency. New York City, like many global urban economies, stands at a fork in the road: one path leverages AI to build a fairer, smarter workforce, while the other allows silent displacement to undercut economic inclusion and dignity of work.

If regulators, AI makers, employers, and educators act on the insights from this report—not just the numbers but the stories they imply—New York could become the first AI-adaptive metropolis. If not, it may become the first cautionary tale.