Training Your Replacement, One Keystroke at a Time - OnLabor

TEXT ANALYSIS: Training Your Replacement, One Keystroke at a Time

The Dissection

The article documents a specific extraction mechanism in real-time: employers using worker activity as simultaneous production input and AI training data. It identifies the phenomenon correctly — keystroke capture, call transcript reuse, AI-assisted editing at degraded pay — and frames it as a labor rights failure requiring legal and contractual remediation. The argument is structurally coherent and factually grounded. It is also, in the most operationally important sense, wrong about what kind of problem this is.

The Core Fallacy

The article treats the automation-training-as-work problem as a correctable abuse within an otherwise functioning employment system. This is the lullaby classification at its most textbook. The framing implies: identify the harm, apply legal or contractual limits, restore appropriate boundaries.

The hidden assumption is that the mass-labor-to-AI-capital transition is a policy problem rather than a structural phase transition. The article's proposed solutions — constructive dismissal doctrine, collective bargaining, legislative safeguards, "compensation for automation training contribution" — treat symptoms with the vocabulary of a system that cannot support those remedies at scale or speed.

Constructive dismissal doctrine cannot track whether keystroke data is being used for training versus monitoring. At-will employment in the U.S. renders individual refusal structurally suicidal. Legislative response operates on political timescales that AI deployment does not respect. Collective bargaining is the strongest of these mechanisms, but it still operates on a per-firm, per-negotiation basis against systems designed for economy-wide labor replacement.

The article identifies the wound. It proposes bandages that require a patient with functioning circulatory capacity.

The Hidden Assumptions

1. That worker leverage through law or contract is a viable counterforce to AI deployment economics. It is not. When AI training data extraction is cost-negative relative to productivity gains, legal friction is a rounding error.

2. That automation training is a discrete function that can be separated, priced, and compensated. Under P1 (Cognitive Automation Dominance), every cognitive work output becomes simultaneously a production act and a training act. You cannot invoice for "contributing to your own replacement" because the replacement is the point, not a side effect.

3. That there exists a stable, definable "job" to protect. The article assumes the employment relationship has sufficient structural integrity to be reformed from within. The Discontinuity Thesis operates on the premise that this relationship is dissolving as a mass-phenomenon, not merely being abused.

4. That the dignity and wage-theft framing generates sufficient political or legal momentum. It is morally correct and operationally inert. Dignity arguments have not historically stopped capital from automating; they have, at best, delayed it while extracting maximum training value.

Social Function

This is labor's grief response to structural dissolution, dressed in the language of legal remedies. It is written by people who understand the mechanism accurately but cannot — or will not — name the terminal prognosis. The article does real investigative and analytical work. Its conclusions are the work product of someone who has seen the autopsy results and is arguing about whether to try CPR.

The Verdict

The article correctly diagnoses stealth wage theft as the operative mechanism. It incorrectly assumes the theft is actionable in a way that preserves the underlying employment structure. Under DT logic, the extraction is not a bug in an otherwise-functional system — it is the primary function of the transition. Workers producing output while simultaneously training their own replacement is not an abuse requiring remedy. It is the mechanism by which the post-WWII employment-wages-consumption circuit is severed. The law, bargaining, and legislation the article proposes are lag defenses at best, relevant to small populations of highly organized, high-skill, well-represented workers, and irrelevant to the structural math governing the majority.

Partial truth. Lullaby in legal clothing. The wound is real. The bandage will not hold.