AI may not eliminate costs but redistribute: HR leaders

TEXT START: The economics of AI are becoming more complex, with spending shifting from payroll to computing infrastructure, model licensing, data preparation, cybersecurity, governance, integration and the people needed to manage the technology.

The Dissection

The article is performing transition management. It recodes automation from labor elimination into workforce redesign, task decomposition, reskilling and redeployment. Its cost analysis is real at the deployment stage, but it treats transitional overhead as evidence that human labor remains structurally necessary.

The Core Fallacy

It confuses the cost of adopting AI today with the long-term competitive consequence of AI. Infrastructure, licensing, governance and integration costs do not preserve the mass employment circuit; they relocate expenditure from wages to capital, vendors, energy and specialized labor.

The decisive question is not whether an AI workflow is immediately cheaper than a human workflow. It is whether firms can become less dependent on broad human labor. Once AI improves, scales and diffuses, competitive pressure rewards organizations that produce the same output with fewer workers. Productivity gains become headcount compression unless demand expands fast enough to absorb the surplus—and the article supplies no basis for assuming that expansion.

The claim that AI-related hires exceed AI-linked layoffs is also a weak comparison. It ignores hiring avoidance, reduced hours, wage compression, displaced non-AI roles, concentration of new jobs among a small technical elite and the possibility that one AI specialist enables the removal of many routine workers. The denominator is doing the public-relations work.

Hidden Assumptions

• Reskilling will occur at the speed of capability improvement and will produce economically necessary jobs rather than merely more employable surplus workers.
• Human judgment, creativity, empathy, communication and ethical reasoning will remain difficult for AI indefinitely. Under the Discontinuity Thesis, these are residual categories at the current frontier, not permanent moats.
• Redeployed workers will be needed in sufficient numbers. Creating capacity is not the same as creating labor demand.
• Firms will retain workers when competitors can use automation to lower dependence on them.
• AI operating costs will remain high enough to protect human labor, despite scale, learning and centralized infrastructure.
• Institutional knowledge must remain embodied in employees rather than being captured, codified and automated.
• The current human-plus-AI arrangement is a stable endpoint rather than a bridge to deeper substitution.

Social Function

Primary classification: transition management. Secondary classifications: ideological anesthetic and elite self-exoneration. The article contains a partial truth—AI deployment has substantial costs and creates real technical, governance and integration work. But it uses that truth to conceal the structural result. It reassures workers that their jobs are being redesigned while normalizing the disassembly of jobs into automatable tasks. It reassures managers that layoffs are not the objective while giving them a framework for identifying exactly which tasks can be removed.

The language of human strengths is especially useful ideological packaging. It converts temporary human residue into supposedly durable human advantage and presents reskilling as an absorption mechanism. In practice, it can function as a selection filter: a smaller number of workers are elevated into oversight, relationship, liability and orchestration roles while the broader workforce loses bargaining power.

The Verdict

This is a competent description of the lag phase and a misleading forecast of the destination. AI does redistribute costs—but from wages broadly distributed across society toward concentrated ownership of models, infrastructure, energy, data and platforms. The article mistakes the scaffolding of automation for a rebuttal of automation.

Under P1, cognitive capability keeps advancing. Under P2, institutions cannot permanently cordon off stable human-only economic domains. Under P3, the majority can lose access to economically necessary labor even while firms retain some humans for oversight, trust, liability and transition work. Human-plus-AI is not the rescue architecture. It is the bridge over which the workforce is reduced.