The Ebb and Flow (and Ebb) of Entry-Level Work | American Enterprise Institute
Released last month, Google’s AI and Economy ATLAS sorts 14.6 million Gemini conversations into jobs and associated tasks. The headline is reassuring: AI is assisting people, not replacing them. Its findings on the role of expertise may contradict this hopeful analysis and tell us something very different about the future of work, especially at the entry level.
A little background on how jobs and tasks get sorted is in order before launching into analysis. The U.S. Department of Labor’s O*NET database breaks every occupation into specific task statements. Some of those tasks demand rare, specialized knowledge, while others do not. A study from last year sorted tasks by the language they are written in, dividing them into those with unusual, field-specific terms and those described in ordinary ones. Google’s ATLAS applies the same sorting framework to Gemini conversations.
ATLAS shows how AI use concentrates in cognitive work, and, within it, skews toward the nonspecialist end. Tasks in the bottom quarter of the expertise ranking are the most overrepresented relative to their share of the task universe, especially the non-routine cognitive ones.
AI adoption also climbs with pay. ATLAS counts conversations per worker in each occupation, then compares occupations against each other. A 1 percent rise in an occupation’s median pay goes with a rise of more than 2.5 percent in AI use per worker. The typical American job pays about $62,000. The typical job behind a Gemini work conversation pays about $83,000.
Combine the task and salary data, and it looks like well-paid workers are using AI on the least expert tasks inside their own jobs. AI use absorbs junior-level tasks, helping to illustrate why entry-level jobs and workers appear to be the most vulnerable at this stage of the AI revolution; people who used to assign routine work to assistants are assigning it to AI instead.
This is how skill-biased technological change operates inside knowledge work and how it impairs developmental pathways for less experienced workers. A firm that assigns routine work to a model reduces its need for junior workers, capturing labor savings that flow immediately to the bottom line. (It also increases senior workers’ control over work product development, a different story.) The cost arrives years later and disperses across the economy, as the pool of junior workers able to move up to higher-level tasks begins to dry up.
This is a classic example of the tragedy of the (skills) commons: long-term human capital development is expensive and its benefits crucial but diffuse. Shorter-term business incentives lean toward cutting costs by reducing junior-level employment and increasing work intensity among more senior workers.
The value of this pattern for quarterly financial reports is indisputable. The human capital development case cuts the other way, and on a pay-later basis. Fixing incentives in the labor market structure to sustain talent pipelines will require policy changes for government and business that bring a sharper focus on the importance of long-term talent development.
