How will AI automation hit — like a crashing wave or a rising tide? | MIT Sloan

TEXT START: AI performance is improving broadly across many workplace tasks, not in sudden shocks.

The Dissection

The article reframes AI displacement from a sudden occupational collapse into gradual task-level saturation. Its measurements show genuine capability gains, but its interpretation converts systemic danger into an administrative planning problem: workers and institutions supposedly have time to adapt while jobs are merely reorganized.

The Core Fallacy

It confuses gradual capability improvement with gradual economic impact. Firms do not need near-perfect AI to cut labor. A system that completes 50%–75% of text tasks without edits can already reduce headcount, compress wages, and make one worker supervise the output of several.

The rising tide is not safer than the crashing wave. Under the Discontinuity Thesis, it is more dangerous because it erodes tasks across entire occupations without creating a single dramatic event that triggers resistance. “Jobs will be reorganized” can mean the same work performed by fewer people, with the remaining human tasks degraded into low-bargaining-power servitor roles.

The study’s incomplete information, regulatory, and error-cost caveats are lag defenses. They slow deployment; they do not reverse the underlying substitution mechanics.

Hidden Assumptions

• Employment remains viable as long as an occupation retains some tasks beyond current language-model capability.
• Firms will wait for near-perfect performance instead of accepting adequate output plus human review.
• Productivity gains will produce worker adaptation rather than layoffs, wage compression, or labor-force exclusion.
• A transition window automatically gives workers meaningful power to respond.
• “Minimally sufficient” task performance maps cleanly onto production-ready automation.
• Legal, institutional, and data-integration barriers will remain durable enough to prevent competitive adoption.
• Human–AI task reorganization preserves productive participation rather than concentrating ownership and control in AI-capital holders.

Social Function

Classification: partial truth serving transition management and ideological anesthetic.

The empirical core is not worthless. AI capability is improving broadly, deployment is not instantaneous, and task-level performance does not equal immediate occupation-level elimination. But the article uses those truths to smuggle in a comforting conclusion: that time is equivalent to safety. It is not. Time may simply give firms more opportunity to automate before workers understand that their bargaining position has already been amputated.

The Verdict

The article gets the shape of the threat largely right and the consequence badly wrong. AI may arrive as a rising tide, but a rising tide still drowns the beach. Gradual saturation can kill the postwar employment–wage–consumption circuit without a spectacular crash, leaving the majority economically unnecessary while Sovereigns capture the productive surplus. The warning is therefore not “prepare to avoid being blindsided.” It is: the inundation has already begun, and adaptation without ownership is managed decline.