AI: 82% of executives already value their workers less
AI is delivering on its managerial promises before its technical ones. Contempt arrives before benefit.
The number nobody wanted to publish
82% of executives say AI has lowered the value they place on their employees. That comes from the G-P (Globalization Partners) study released this week, covering 2,850 VP-and-above managers in the United States, Germany, Singapore, Australia and France.
The same study finds that 73% of executives view AI results as "below expectations." MIT NANDA 2025 reports that 95% of enterprise GenAI pilots produce no measurable ROI. Gartner 2026 puts 42% of AI projects in the zero-ROI bucket. The promised benefit is neither measured nor demonstrated for the vast majority of organizations. 16% are even posting a clear net loss.
The devaluation, on the other hand, did not wait.
What the bosses say off the record
The G-P survey is interesting because it does not ask executives whether they are satisfied with AI. It asks how AI has changed the way they see their teams. The difference matters. A leader can find the tool disappointing and still respect the humans keeping the shop running. Or a leader can decide the tool is enough to look down on those humans, even when the tool does not work.
The second scenario dominates. At 82%, this is the order of magnitude of a cultural shift.
That is not the only awkward data point. 88% of the same executives suspect their employees of using AI "performatively rather than productively." Translation: their own teams are burning time pretending to use tools the bosses themselves mandated without any certainty that they work.
The contempt is mutual, but the asymmetry is sharp. Executives steer. Employees perform a play whose rules are dictated to them.
Amazon, or tokenmaxxing as a corporate sport
To see this dynamic playing out inside a real company, look at Amazon. The firm deployed an internal tool called MeshClaw (its in-house equivalent of OpenClaw) and set a target: 80% of developers must use AI every week. Managers track token consumption. Officially, those numbers do not feed into reviews. Unofficially, employees report the opposite.
The result, documented by Jellyfish in Q1 2026, is striking. A median developer consumes 51 million tokens per month, around $52. The top 10% of Amazon consumers explodes to 380 million tokens a month, around $700 per head.
The cost per merged pull request is even more telling. The 20% most efficient: 11 PRs merged for $3. The 20% heaviest users: 23 PRs for $1,822. For comparable PR value, the theater costs 600 times more.
One anonymous employee sums it up: "Managers look at these numbers. When usage is monitored, it creates perverse incentives and people become extremely competitive about it."
This is Goodhart's law applied to AI rollout. When a measure becomes a target, it stops being a useful measure. Ask for tokens, get tokens. Lose track of whether you are getting actual work.
Why contempt comes before ROI
The sequence is worth noting. Normally, organizations wait for an investment to deliver before adjusting their view of the resources it replaces. With AI, the order is reversed. The devaluation of workers is already locked in even though 95% of enterprise GenAI pilots show no measurable ROI, per the MIT NANDA study published in late 2025. Contempt is running ahead of productivity.
The most parsimonious explanation: the two promises do not have the same cost. The technical promise (productivity, margins, automation) requires years of integration, data quality work and redesigned processes. The managerial promise (justifying wage pressure, reassuring shareholders, accelerating restructuring) is achievable immediately.
Saying it out loud is enough. Contempt costs zero dollars. ROI costs real money.
The G-P numbers capture that timing gap turned into a stable posture. Executives no longer wait for AI to deliver before treating their employees as less essential. They have already made the call.
On Wall Street paper, this works fine. On the actual income statement, it is a different story (a subject already explored in the AI productivity paradox).
What changes for the people inside
For an employee, this means living in an organization where the boss values them less without having seen the slightest productivity gain on their own work. They carry the symbolic cost of a payoff that never arrived.
If the boss already considers them replaceable, the rational move is to do whatever it takes to avoid being replaced. Burn tokens. Perform. Become a visible consumer rather than a silent producer.
For a manager, the AI KPIs sent up the chain measure almost nothing. Token consumption, prompt counts, weekly adoption rate: three metrics teams will inflate the moment they become critical.
The manager reports theater. The CEO draws conclusions from numbers that describe nothing but the fear of those teams.
For an executive, the trap closes in. The decision to value humans less has already been made. The KPIs being tracked are theater. The tokens being paid for produce PRs that cost 600 times too much.
When ROI does not come, two options remain: admit the promise was hollow, or find a new culprit. Given the orientation of the G-P panel, the culprit has already been identified.
The current phase of AI
Time to name what is happening. This is not a productive rollout. This is a disciplinary rollout.
ROI may eventually arrive, once back-office chains are integrated and the models start delivering on their technical promises. Contempt did not wait. It is already here.
