Agentic AI Could Break the Old Rules of Job Displacement
The old frameworks for automation are broken. For decades, experts have studied how automation displaces workers, relying on a model developed by Acemoglu and Restrepo. This model effectively broke down occupations into discrete tasks, evaluated which tasks AI could automate, and estimated the number of workers affected. While this approach worked for manufacturing and earlier waves of software automation, it fails to capture the fundamentally different landscape presented by Agentic AI.
Understanding Agentic AI
Agentic AI systems do not merely automate individual tasks; they orchestrate entire occupational workflows from start to finish, making numerous interconnected decisions. For instance, a judge employs a multitude of skills in their role, including reading, reasoning, and legal analysis. An autonomous workflow can integrate these capabilities, enabling a single AI system to read a case file, research precedents, draft opinions, and identify key issues without human intervention. This capability has the potential to displace entire occupations.
Introducing the Agentic Task Exposure (ATE) Score
To address this new reality, researchers have developed a new measurement system: the Agentic Task Exposure (ATE) score. This composite measure captures the risk that a single AI system executing a coherent workflow will eliminate the need for human workers in a specific job. The ATE score combines:
- AI Capability Score: Measures the proportion of tasks within an occupation that current AI systems can competently execute.
- Workflow Coverage Factor: Assesses whether automatable tasks form a coherent, orchestrable sequence.
- Logistic Adoption Velocity Parameter: Models how quickly organizations will deploy agentic systems.
Why Workflow Matters Differently than Tasks
The distinction between task automation and workflow orchestration is crucial. When specific tasks become automatable, skilled humans often remain necessary to manage the work and apply judgment. However, when an agentic system can chain tasks together autonomously, it eliminates the need for human orchestrators entirely. For example, while AI might be capable of financial modeling, evaluating credit risk, and drafting recommendations, an agentic system that can handle all these tasks in sequence makes the credit analyst redundant.
Geographic Concentration of Vulnerability
The analysis reveals that the geography of vulnerability is stark. Researchers focused on five major US technology regions—Seattle-Tacoma, San Francisco Bay Area, Austin, New York, and Boston. They found that 93.2% of 236 analyzed occupations across six information-intensive SOC categories face significant disruption risk by 2030. In cities like San Francisco, nearly every job in these categories is at risk.
Occupations such as credit analysts and judges, which involve document-in-document-out workflows with heavy reasoning components, exhibit particularly high ATE scores, indicating a very high risk of displacement.
Emerging Occupational Categories
Despite the potential for job displacement, the research identifies seventeen emerging occupational categories that may expand as agentic AI systems are adopted. These roles are concentrated in three areas:
- Human-AI Collaboration Roles
- AI Governance and Auditing
- Domain-Specific AI Operations
These emerging occupations will require different skills than those displaced, indicating the need for specialized training and potentially significant career changes for affected workers.
The Timeline for Adoption
The findings hinge significantly on the speed of adoption. The research models adoption through a logistic curve, predicting that the exponential growth phase will occur between 2025 and 2030. Factors such as financial pressure, regulatory approval, and competitive pressure could accelerate adoption, while risk aversion and technical limitations might slow it down.
Policy Implications
The research emphasizes the importance of proactive policy responses. If disruption is concentrated geographically and occurs more rapidly than previous automation waves, it necessitates immediate action from regional economic planners. This includes diversifying economies and investing in education and infrastructure for emerging roles.
For workers in high-ATE occupations, the research advocates for immediate transition planning. Understanding whether retraining into emerging roles is feasible or if migration is necessary is critical for future job security.
Conclusion
Ultimately, the research highlights that agentic AI capabilities are either imminent or already present. It shifts the discourse from “will agentic AI displace workers?” to “at what pace and under what constraints?” The visibility of these risks compels society to consider what policies and safeguards are necessary to minimize disruption while maximizing opportunities for workers transitioning into new roles.
