Corporate America's Great Flattening Has a Judgment Problem

Impulsive job cuts to the middle of the org chart can shortchange progress. Pricing and judgment are keys to organizations maintaining advantage in the future of work.

Across corporate America, companies are making heavy cuts into the middle of their org chart. Crunching numbers on the spreadsheet, the math behind this change looks irresistible. Except that leaders may be failing to consider the long-term costs involved, coupled with a clear trail of warnings.

Here’s the backdrop. The Bureau of Economic Analysis puts American wages at roughly $13 trillion a year. Within this, more than $3 trillion pays for management and coordination (M&C). That’s nearly one dollar in four paying for activities that include routing work, finding problems, helping decide what matters, and much more. Recently, noted MIT economist Daron Acemoglu named judgment-heavy middle management as the next layer where agentic AI will matter most. Moreover, Goldman Sach's economists already see AI trimming monthly US job growth by 10,000 to 15,000 positions, with an estimate of about 15 million Americans changing roles as AI roots more firmly in the next few years.

There’s no argument that the middle layer is being exposed and targeted. Most organizations today price the M&C layer as pure cost, as a line to be arbitraged against a cheaper AI input with costly implications. For example, in late May 2026, Axios reported that one company with no internal usage limits ran up $500 million in AI token costs with a single vendor in just 30 days. That’s just north of $11,000 for every single minute, 24 hours a day for seven days a week. Separately, Uber pushed staff to use AI heavily, where they added employee ranking incentives on a leaderboard based on AI use. Later, as reported by Fortune, the company burned through its entire 2026 budget for AI coding tools in just four months.

Once you read past the headlines, the diagnosis gets more useful. In the case of the first company mentioned above, the $500M went to licenses that nobody capped—more of a governance failure rather than a technology one. Separately, Sophia Velastegui, former chief AI officer at Microsoft, named the widespread underlying error: most organizations default to automating the tasks they dislike rather than the tasks that carry the most value. One example was a different CTO in the same reporting, who found staff burning enterprise tokens to check the weather.

As it happened, Microsoft cancelled most of its Claude Code licenses partly over cost. Ali Ansari of Micro1 calls the broader pattern tokenmaxxing, the act of consumption that’s mistaken for progress. This is not a failure of models, but rather poor decisions about where to point AI and agent spending, made by leaders with no mechanism for making that decision well.

Those stories get told as budget cautionary tales. Actually, they’re evidence of a challenge, where current thinking runs up against a shift in the nature of work and workforce.

Whether paid in the form of a human salary or tokenization of AI, a dollar buys the same commodity of work aimed at an outcome. The return shows only when the work coordinates, and when what an AI agent finds reaches a person who can act on it in time to matter. In the evolution from humans alone to a blended workforce of humans, agents, and robotics, companies swapped a payroll line for a token line. One challenge being the assumption that coordination would naturally fall into place—and it has not.

The evidence for better thinking in this space has been emerging. PwC's 2026 Global AI Jobs Barometer, built on more than a billion job postings, found the most AI-exposed companies have tripled their productivity growth lead over the least exposed since 2022. This, while growing headcount and wages.

Taking a deeper dive, it’s worth studying the spread inside that leading group. Companies in the most-exposed quartile are growing productivity at 33.5% against a 2018 baseline. The top 20% of that same quartile, the ones PwC calls “superstars,” reach 163%. Five times the return, running technology every competitor can buy at the same price. Set that against PwC's CEO Survey, where only 8% of chief executives report AI generating more than a slight revenue increase in the past year.

The superstars are also not the ones cutting. At the most AI-exposed companies:

• Headcount grew 52.2%, against 35.7% at the least exposed

• Also at the most exposed, wages were up 24.4% against 16.6%

PwC's appears to distinguish leaders from laggards in their use of AI to pursue growth, across sector lines, instead of hunting efficiencies. Instead of using the excuse of exposure, large company “compounders” shared the discipline of deciding to focus enterprise attention.

Nobel-winning scholar Herbert Simon called this in 1971, warning that a wealth of information creates a poverty of attention. Intelligence became a commodity once every company and its competitors could rent the same models. The input that never got cheaper was the ability of leaders to catch the right signal, to inform the right decision, in time to act.

And this is key to what we are seeing in workforce reduction, specifically heavy cuts in the middle. Leaders may be forgetting that managers and middle-org professionals learn judgment by doing coordination activities. Historically, coordination activities have also helped serve as an “in the trenches” apprenticeship for building judgment. Rapid replacement of humans by AI, has the very real potential to thin out your company’s next generation of deciders. Companies making signficant middle org cuts today, could be leaving themselves hamstrung for the trust in decisions across the enterprise, they will need most in 2031.

PwC's Global CEO Survey found 49% of chief executives expect AI adoption to decrease junior hiring over the next three years, against 12% who say the same about senior hiring. While researchers from Stanford have already measured a 16% decline in entry level roles across AI-exposed fields. The bottom rung is being pulled up.

Now look at what employers want from their hires. In AI-exposed entry level jobs, employers are seven times more likely to demand traditionally senior skills than in the least exposed ones; PWC lists them:

• motivational leadership

• team building

• stakeholder management

• mentorship

• data-driven decision making

In the U.S., entry level roles that added ten or more such skills grew 35% between 2019 and 2025, while the roles that did not shrank by 10%. New tasks in the most AI-exposed roles are 2.5 times more likely to demand what MIT's Loaiza and Rigobon classify as human-intensive capability, a set of abilities that includes judgment and ethics under ambiguous conditions.

Hold those two findings together and the bind is obvious. The market has started asking for senior judgment at the front door during the same years companies are dismantling the layer where judgment was learned. PwC states the resulting obligation plainly: companies and educators must rethink how they train, mentor, and scaffold early career pathways so junior workers build senior skills far earlier than before. Nobody has yet explained where that scaffolding comes from once the middle is gone.

No one budgeted for that. It appears on no spreadsheet.

Organizations who take a knife to the middle must change their thinking to recognize and treat enterprise attention as an explicit budget. Where critical signals, buried in an organizations data, are identified with context, urgency, and an ability to inform more efffective and defensible decisions. Such signals having reasoning and time-to-act attached, so leaders can take action and have such action and outcomes saved in institutional memory to have the enterprise, and its decisionmakers at all levels, learn from its own history.

None of this is meant to argue against AI in our workforce. The argument is about where the freed value lands. Boards should ask a harder question than how many layers came out this year. When the machines route everything, who decides what deserves to be noticed? And who is building the people who will know how?