Turns Out 'Stop Hiring Humans' Was Terrible Advice | Built In

TEXT START: Artisan bought billboards across San Francisco in 2024 that read “Stop Hiring Humans.”

The Dissection

The article is not a refutation of automation. It is a postmortem of a crude benchmark: headcount reduction was mistaken for AI success before firms had captured tacit knowledge, exception-handling, provenance, and accountability.

Its actual proposal is to industrialize expert judgment—encode it, verify it, distribute it, attribute it, and escalate unusual cases to humans. That is not preservation of the old labor market. It is a more competent method for concentrating expertise and thinning the labor base.

The Core Fallacy

The article confuses “AI still needs accountable human judgment in some workflows” with “humans remain broadly economically necessary.” Those are not equivalent.

If one senior engineer’s standards can guide hundreds of teams, or one clinician’s verified guidance can serve thousands of cases, the human role survives as a scarce supervisory node—not as mass employment. The article’s solution increases output per expert, which accelerates the destruction of the wage-to-consumption circuit.

Verification establishes provenance, jurisdiction, review dates, and liability. It does not preserve headcount. It converts expertise into a reusable control layer and reduces the number of bodies required to produce the same output.

Hidden Assumptions

• Credentialed experts will retain the economic rents from their encoded knowledge instead of platforms and employers commoditizing it.
• Verified credentials and review dates will reliably produce correct judgments in novel or ambiguous cases.
• Human escalation will remain affordable and available as automated decisions multiply.
• “Human-AI collaboration” will mean durable partnership rather than a few controllers supervising vast machine systems.
• Economic value created by scalable expertise will be distributed to workers rather than captured by system owners.
• Encoding tacit knowledge will preserve experts’ leverage instead of making their judgment transferable and eventually replaceable.
• Quality failures are merely immature implementation problems, rather than evidence of domains where liability and exception-handling impose real constraints.

The article also assumes that preserving a human checkpoint preserves human economic centrality. Under the Discontinuity Thesis, that assumption is already dead.

Social Function

Classification: partial truth, transition management, prestige signaling, and elite self-exoneration.

The article correctly warns companies not to fire the people whose undocumented judgment keeps workflows functional. It gives executives a safer migration plan: retain a thin layer of credentialed authorities, attach their legitimacy to machine outputs, and automate everything beneath them.

It also offers experts a reassuring story about attribution and compensation. But the likely structural result is not restored professional autonomy. It is the conversion of expertise into licensed infrastructure, with ownership and distribution accruing to whoever controls the AI systems, data, and channels.

The Verdict

The article is correct about the failure mode and wrong about the historical implication. Naive replacement can degrade quality when firms delete tacit knowledge before extracting it. That is a lag problem, not a reversal of the thesis.

Its own remedy is the Discontinuity mechanism: remove judgment from individual workflows, package it into systems, let a small number of qualified humans govern machine output, and reserve human labor for exceptions. The humans are not saved; the bottleneck is relocated.

This is not a defense of mass human participation. It is a manual for better-managed obsolescence disguised as a defense of expertise.