Generative Artificial Intelligence, Productivity, and the Future of Work - TD Economics

TEXT START: Conversations around how generative artificial intelligence (AI) will reshape the future of work have emerged as a common discussion point in both academia and corporate boardrooms.

The Dissection

This is a delay memo disguised as an economic analysis. It correctly distinguishes technical exposure from immediate automation, then quietly converts that limited observation into a broader claim that jobs will mostly survive. It measures layoffs, aggregate employment, and labor-market churn while avoiding the decisive variables: ownership of AI capital, labor’s shrinking bargaining power, hiring collapse, wage compression, and the loss of economically necessary work.

The report treats implementation cost, oversight, imperfect reliability, and gradual adoption as durable barriers. Under the Discontinuity Thesis, these are lag defenses. They slow the process; they do not reverse it. The article’s “job apocalypse” is defined as synchronized mass layoffs, allowing quieter forms of displacement to disappear from view.

The Core Fallacy

The central error is mistaking delay for negation.

AI does not need to automate every task in an occupation before employment collapses. It only needs to remove enough valuable tasks for firms to stop hiring, compress teams, intensify workloads, or pool the remaining human work among fewer workers. The report’s focus on jobs with more than half their tasks fully automatable uses the wrong threshold.

It also assumes productivity gains will recycle into wages, consumption, and new employment. That requires the old wage-consumption circuit to remain intact. If AI increases output while ownership and income concentrate around AI-capital holders, productivity growth does not restore mass participation. It merely produces more output with fewer economically necessary humans.

The report therefore mistakes P1 barriers in the present for a refutation of P1, P2, and P3 in the trajectory. Reliability improves. Costs fall under competition. Firms redesign workflows when rivals gain an advantage. Human institutions cannot permanently preserve human-only cognitive labor markets at scale.

Hidden Assumptions

• Firms must achieve fully autonomous, end-to-end job replacement before reducing headcount.
• Retraining will create jobs rather than merely redirect workers into a shrinking queue.
• New AI-infrastructure employment will absorb displaced cognitive workers at comparable scale and quality.
• Productivity gains will flow to wages and demand instead of primarily to capital owners.
• Aggregate unemployment and layoffs are adequate measures of individual economic viability.
• Stable labor-market churn can absorb AI displacement without permanent cohort scarring.
• Human “complementarity” will persist even as AI takes over coordination, analysis, drafting, and judgment-heavy tasks.
• Policymakers can preserve full participation without confronting control of AI capital.
• Current adoption speed is a reliable guide to the eventual competitive equilibrium.

Social Function

Primarily ideological anesthetic and transition management, with a legitimate layer of partial truth and elite self-exoneration.

The article tells businesses to retrain, workers to improve their skills, and policymakers to use the available time. That framing converts a structural ownership problem into an individual adaptation problem. It reassures institutions that gradual exclusion is orderly and manageable, while treating early-career workers who cannot enter exposed occupations as statistical softness rather than the first casualties of productive participation collapse.

Its valid point is narrow: capability does not instantly equal deployment. Its political function is much broader: use that delay to normalize the transition without confronting who owns the replacement system.

The Verdict

Technically competent on friction, structurally evasive about the destination. The report measures the corpse’s pulse and calls the organism healthy.

Its own evidence—slower hiring in exposed occupations, disproportionate damage to younger workers, and substitution through non-hiring rather than mass firing—is early-stage P3. The system is already withdrawing access to economically necessary work, quietly enough to evade an unemployment headline.

This is not evidence that AI will mostly change jobs instead of eliminating them. It is evidence that elimination begins before the firing notices: fewer entry lanes, fewer workers per function, weaker wages, and increasing dependence on whoever owns the machines. The article describes hospice care for the post-WWII employment order and labels it adaptation.