Agentic market $10.8B and climbing  ·  editor@gaasnews.com
Sections
HomeWhat is GaaS?PlatformsPricingGlossaryOpinionAboutContact
HomeLabor & WorkforceHarvard working paper
Labor & Workforce

Harvard paper warns agentic AI is eating the entry-level rung

The Mossavar-Rahmani Center paper documents falling employment and job postings for early-career workers and coins Intellectual Atrophy as a compounding risk.

AJ
Andrew Jamerson
Founding Editor
Jul 21, 2026 · 4 min read
A new Harvard Kennedy School working paper maps agentic AI onto the early-career job ladder. // GaaS News
TL;DR
  • A new Harvard Kennedy School working paper argues AI is the first technology able to automate the entry-level work through which expertise forms.
  • It documents a 6 to 13 percent employment decline for workers aged 22 to 25 in the most AI-exposed roles and a 15 to 20 percent drop in graduate-level postings.
  • The paper coins Intellectual Atrophy as a compounding risk and proposes Workforce 4.0 frameworks, though it is one institution's analysis.

A new working paper from Harvard Kennedy School argues that agentic AI is the first technology capable of automating the entry-level cognitive work through which professionals actually learn their trade. The paper, M-RCBG Associate Working Paper No. 276, "Future of Work in the Age of Automation, Augmentation, and Agentic AI," is circulating in July 2026 from the school's Mossavar-Rahmani Center for Business and Government. It is written by senior fellow Solmaz (Sol) Rashidi, a former enterprise Chief AI and Data Officer, and represents a single institution's analysis rather than settled consensus.

The numbers on young workers

The paper's central claim is empirical. It documents a 6 to 13 percent employment decline among workers aged 22 to 25 in the roles most exposed to AI, a 15 to 20 percent drop in graduate-level job postings, and a 14 percent decline in the monthly rate at which young workers enter those occupations, per the paper. Read together, those figures describe a bottom rung being pulled away rather than a broad collapse in demand for labor. Senior work continues. The tasks that used to season a new hire are the ones going first.

That is the twist that separates this argument from a standard automation story. Earlier waves took over routine manual or clerical work. This one, the paper argues, targets the developmental cognitive work, the drafts, the first-pass research, the low-stakes analysis, that a junior professional does badly at first and learns from.

Intellectual Atrophy and Workforce 4.0

To name the second-order risk, the paper introduces "Intellectual Atrophy," a compounding erosion of skill that follows from over-reliance on AI tools. If agents handle the reps, the argument goes, neither the newcomer nor the mid-career worker keeps the muscle sharp, and the loss accrues quietly over years. Against that, Rashidi proposes "Workforce 4.0" frameworks meant to redesign roles and training so that humans retain the judgment agents cannot yet supply. The figures and the frameworks are the paper's own, and the piece reads as a warning aimed at the enterprises doing the deploying.

What this means for the agent economy

The GaaS value proposition and this pipeline problem point in opposite directions. Vendors sell agents precisely because they can absorb the repetitive, entry-level tasks that clog a team. If those same tasks are how junior workers become senior ones, then every deployment that lands as promised also quietly narrows the funnel that produces future expertise. The bill does not arrive this quarter. It arrives when the current crop of seniors retires and the bench behind them was never trained.

Our earlier coverage of how firms are redesigning entry-level jobs offered the more optimistic counterpoint, that roles can be rebuilt rather than deleted. The Harvard paper lands as the caution against assuming that happens on its own. It arrives the same week Gallup reported that a majority of US workers now use AI on the job, which is the adoption curve that makes the pipeline question urgent rather than theoretical. Enterprises buying autonomy will have to answer who trains the humans once the agents take the training tasks.

AJ

Andrew Jamerson

Founding Editor, GaaS News

Andrew Jamerson is the founding editor of GaaS News, covering the economics of the agent era. He started the publication to cover Agentic AI as a Service as a dedicated beat and edits every article on the site.

Be on the list when the beat breaks

One email when a platform ships, a round closes, or the ground shifts under the software stack.