Study Followed AI Investment Dollars, Finds Surprising List Of At-Risk Jobs

Why AI Startups Are Going After Office Clerks Before Judges and Surgeons

Fears that AI will upend medicine, law, and other high-skill professions have dominated the conversation in recent years. A new study says the investment data tells a different story. Venture-backed startups, it turns out, are far more focused on the people who manage files, organize data, and keep offices running than on the elite professionals most people assume are first in line.

Researchers developed a new tool to measure which jobs are actually being targeted by funded AI startups. Rather than asking what AI could theoretically do, the team asked a sharper question: where is the money going? Published in the journal PNAS Nexus, the study introduces what the researchers call the AI Startup Exposure index, or AISE, designed to capture not just what AI is technically capable of, but what entrepreneurs and investors are actively building and funding right now.

‘Follow the Money’: How Researchers Built the AI Startup Exposure Index

To build the index, the team analyzed descriptions of AI products from startups funded by Y Combinator, a U.S.-based startup accelerator that accepts only a small fraction of applicants. That selective threshold was central to the study’s logic: if a startup cleared that bar, investors had already done serious work evaluating whether the product could realistically succeed.

Researchers then used an AI language tool called Llama 3, an open-source model made by Meta, to compare those startup product descriptions against standardized descriptions of hundreds of occupations drawn from O*NET, a U.S. government jobs database. When a startup’s product closely matched the core tasks of a given job, that job received a higher exposure score. Results were also checked against a separate database of European startups, producing highly correlated findings, though that comparison is better understood as a consistency check than a full cross-market confirmation.

Which Jobs Are Most Exposed to AI Startup Investment?

At the top of the exposure ranking sits a role that rarely makes headlines in AI anxieties: the general office clerk. Filing systems, database management, sorting records, handling administrative tasks: these are the kinds of duties that funded startups are actively building tools to handle. One startup in the dataset developed an AI-powered event planning tool for large-scale corporate events; another built multilingual AI assistants for answering calls and processing organizational information.

Other high-exposure roles included data scientists, computer and information systems managers, and market research analysts. What these jobs share is a heavy reliance on information processing, programming, and organizational planning.

At the bottom of the list are athletes, judges, and pediatric surgeons. Athletes rely on physical performance that carries intrinsic human value. Judges handle decisions so ethically charged that funded startups in this dataset were not strongly targeting the core tasks of judging, even though AI may be technically useful for related tasks. Pediatric surgeons require a combination of manual skill, deep expertise, and extremely low tolerance for error that creates similar hesitation among investors.

AI Startup Investment Reveals a Gap Between ‘Can’ and ‘Will’

Perhaps the most revealing finding is the gap between which jobs AI could theoretically affect and which funded AI startups are actually targeting. Researchers compared their new index against an existing capability-based measure that ranks jobs by whether current AI technology could perform their tasks. By that standard, highly educated professionals rank near the top. Lawyers, judges, and doctors all score high. Under the startup-investment measure, those same jobs often fell toward the bottom.

Take lawyers versus database administrators. Both require similar cognitive abilities: logical thinking, processing large amounts of information, drawing conclusions from complex material. A capability-only measure treats them as nearly equally exposed. But in the startup investment data, a database administrator scored roughly 0.8 on the exposure scale while a lawyer scored around 0.05. Automating how a company stores its data raises few ethical alarms. Automating legal judgment raises many. That gap does not mean doctors, lawyers, or teachers will be untouched by AI, but it does suggest investors are not yet prioritizing the replacement of their core duties.

A Selective Pressure, Not a Flood

Geographically, AI startup exposure tracks closely with tech industry density. Metro areas built around knowledge work, led by the San Francisco Bay Area, registered the highest average exposure, while Midwestern cities anchored in manufacturing and agriculture showed the lowest. At the industry level, finance, insurance, professional services, and information technology ranked as most exposed, while construction, agriculture, and food preparation ranked at the bottom. Healthcare and education landed in a middle zone, fields where AI appears more likely to assist workers than to displace them.

One additional finding adds a note of caution. When researchers examined startups combining AI software with physical robots, the exposure picture shifted. Several jobs that scored low on standard AI exposure showed meaningfully higher exposure when robotics entered the equation, suggesting that the combination of AI and physical automation could eventually extend disruption beyond office work.

For now, the picture is one of selective pressure rather than wholesale disruption. Funded AI startups are not an indiscriminate wave moving across the workforce equally. Office clerks and data analysts may face more near-term pressure from AI products than surgeons or judges, and the startup investment data shows where that activity is already concentrated.

---

Disclaimer: This article is based on a published academic study and is intended for informational purposes only. The findings reflect startup investment patterns at a specific point in time and do not constitute predictions of job loss, career advice, or economic forecasting.

---