Escalating AI costs challenging large employers: reports | Human Resources Director

TEXT START: Companies generate 'an enormous amount of volume, noise and waste' through AI, says expert

THE DISSECTION

This article presents itself as a practical HR digest on "managing AI costs," but it is actually a death rattle diagnostic buried in operational language. Every data point in this piece—cost scaling curves, productivity failures, governance gaps, informal employee adoption, governance committee formation rates—constitutes evidence that the human labor integration circuit is breaking down at scale. The article is describing the symptoms of a system beginning to reject its own architecture. The framing of "we just need better governance" is the anesthetic.

THE CORE FALLACY

The article treats the symptom as the disease.

The core error: framing escalating AI costs and uneven productivity as a management problem rather than a structural inevitability. The thesis that AI "costs don't scale linearly" and that organizational productivity gains aren't materializing is presented as a solvable governance challenge. It is not. It is the mechanism of dissolution itself.

The argument goes: if we can just align spending, governance, and workforce design, we can capture the value. This is the exact same logic that said "if we manage the mortgages better, housing will remain affordable." The system is not failing because of poor management. It is failing because AI is making the human labor circuit progressively optional, and no governance framework reverses that. The cost spiral, the productivity disconnect, the informal shadow adoption—all of this is what happens when a technology that can substitute human cognitive labor scales faster than institutions designed around human cognitive labor can adapt.

The article even cites that 88% of organizations use AI in 2025, up from 55% in 2023. That is not a success story. That is an adoption curve describing the speed at which human labor dependency is being severed.

HIDDEN ASSUMPTIONS

1. Human employment remains the primary value delivery mechanism. The article never seriously entertains the possibility that AI doesn't need to "augment" workers to deliver value—it can simply replace the function. The framing is always human-plus-AI, never human-vs-AI.

2. Productivity gains are the legitimate metric. The article treats it as obvious that if organizations can't show productivity improvements, something is wrong with the implementation. It never questions whether the productivity frame is the wrong metric entirely—AI creates value by eliminating the need for human labor, not by making human labor more productive.

3. Governance can manage the transition. The repeated calls for "great governance" assume there exists a managed transition path. The DT framework says institutional coordination cannot preserve stable human-only economic domains at scale. The article's earnest invocation of oversight committees and workforce planning structures is precisely the lag defense that will fail.

4. Mental health and reskilling concerns are the core risk. The article treats worker displacement as a human wellbeing problem. The DT framework treats it as a systemic structural collapse problem. These framings are not equivalent. Mental health framing keeps the solution inside the system. Systemic collapse framing requires acknowledging the system is dying.

5. Return on investment is achievable at scale. The article cites "sectors could see returns of up to 9.5 times their investment." This number is a statistical anomaly from early adopters or specific verticals, not the general case. The Stanford HAI data showing 70% of organizations using generative AI with productivity gains not materializing at team scale is the actual signal.

SOCIAL FUNCTION

Ideological anesthetic with transition management utility.

This article serves the function of making the dissolution of the post-WWII employment order feel like a manageable operational challenge. It tells HR professionals they have a critical role in "governance" and "workforce design" precisely when the structural evidence suggests those functions will be automated out alongside the workers they are meant to manage. It is the corporate version of telling intensive care nurses that the patient is making progress while every monitor flatlines.

It is not propaganda in the sense of deliberate deception—it is lag defense literature. It performs the institutional function of buying time by making the crisis sound like a process problem rather than a structural rupture. Every call for better governance is a delaying action.

THE VERDICT

The article accidentally publishes its own autopsy report. Escalating AI costs that don't produce productivity gains are not a governance problem—they are what the system looks like when the technology begins to decouple value creation from human labor. The informal employee AI adoption rate (half introduced tools independently) describes a workforce that has already begun bypassing the institutional structures meant to manage its own obsolescence. The desperate search for "people that can help us do really great governance of AI" is the sound of organizations realizing they don't know how to manage the end of their own reason for existence.

The lag defense is active. The lag defense is failing.

This article is an artifact of Phase 2 institutional paralysis—organizations that have adopted AI faster than they can adapt their own structures, spending more than they can justify, and increasingly unable to answer the question of what human labor is for at scale. The fact that it is framed as an HR challenge rather than a civilization challenge is precisely what the lag defense is designed to do: keep the conversation inside the building while the building is being dismantled.