AI’s trillion-dollar disruption: From efficiency booster to full-scale labour replacement

The economic consequences of artificial intelligence may extend into the trillions of dollars in labour value, as AI shifts from being a productivity enhancer to a direct substitute for human work. In his latest memo, AI Hurtles Ahead, Oaktree Capital Management co-founder Howard Marks argues that the current phase of AI development marks a structural turning point for global labour markets.

Autonomous agents

Marks describes a progression in AI capability that fundamentally changes its economic role. Early-stage AI functioned primarily as “Chat AI,” answering questions and saving research time. It then evolved into “tool-using AI,” capable of executing specific tasks when instructed. Today, however, AI has entered what he describes as the era of “autonomous agents” — systems that can be given an objective and independently complete multi-step assignments, check their own work, and deliver finished outputs.

This distinction is critical. A tool that makes a worker 20% more productive captures only a fraction of that worker’s salary in economic value. But a system that performs the entire task end-to-end represents labour replacement, not assistance. Multiply that substitution effect across software engineers, legal associates, financial analysts, consultants, compliance officers, and other structured knowledge roles, and the scale quickly expands into a multi-trillion-dollar labour market.

AI’s advancement

Marks notes that AI’s advancement is occurring at unprecedented speed. Unlike previous technologies such as computers or the internet, which required decades to mature and diffuse, generative AI has moved from early adoption to mass enterprise deployment within just a few years. Hundreds of millions of users and a majority of corporations are already integrating AI into workflows.

A further acceleration dynamic is now in play: AI is beginning to assist in its own development. Recent model releases have reportedly been used to debug training processes, manage deployment, and refine evaluations — creating a feedback loop in which intelligence contributes directly to building more advanced intelligence. That recursive improvement cycle compresses innovation timelines and amplifies economic disruption.

AI going ahead

However, Marks is careful to emphasize nuance. AI excels in domains rich with historical data and recognisable patterns. It synthesizes vast quantities of information, processes quantitative inputs without emotional bias, and can operate at scale. Yet it remains weaker in truly novel situations where intuition, qualitative judgment, and context-specific insight are required.

For investors, the implications are profound. If AI can process publicly available quantitative information more efficiently than humans, the traditional edge derived from data analysis narrows. Future alpha may increasingly depend on qualitative interpretation, assessing management quality, or forming judgments in unprecedented environments.

Marks ultimately concludes that AI is real, transformative, and likely underestimated in its long-term potential. But he also cautions that transformative technology does not automatically justify every valuation. The prudent approach, he suggests, lies between extremes: neither dismissing AI’s structural impact nor committing capital indiscriminately.

The labour substitution effect is no longer theoretical. The question is not whether AI will alter economic structures, but how quickly institutions, investors, and workers can adapt to the scale of change already underway.