Feature
White Paper
The Future of Leadership Development: AI-Driven Manager Enablement
The promise was efficiency. AI would perform the tedious work, the first drafts, the synthesis, the boilerplate code and hand the time back so employees could think.
For a growing share of workers, the trade has run the other way. AI has not removed labor so much as swapped one kind for another. The new work is slower, less satisfying and harder to name. Workers still look productive on the dashboard, but the judgment, craft and meaning that made the work worth doing are being stripped out from the inside.
Researchers have started measuring this change. A Harvard Business Review survey of 1,488 full-time US workers found that 14% of those using AI reported a specific kind of mental fatigue the authors called "brain fry," defined as mental fatigue from overseeing AI tools beyond one's cognitive capacity. The rate climbed in AI-heavy functions: 26% in marketing, 19% in HR and 18% in engineering.
A separate eight-month study by UC Berkeley Haas, also published in HBR, embedded with a 200-person tech company and found that generative AI did not free up time. It expanded what workers felt able and willing to take on. Prompts seeped into the pauses: lunch, the minutes before meetings and the evening.
Both studies have nuances. The Berkeley work covers a single unnamed company and is still in progress. The brain-fry survey, conducted with co-authors from Boston Consulting Group, also notes that workers using AI agents reported less burnout than colleagues who used none. The researchers suggest this is because burnout tests measure emotional and physical exhaustion, not the sharper, more acute cognitive strain that AI oversight produces. That complicates any direct line from "AI use" to "workers suffering."
AI Oversight Is Labor Too
What the studies do establish is direction, and practitioners who build these systems describe the same arc. "AI stops being a tool the moment you spend more energy supervising it than it saves you," said Mariano Facundo Scigliano, a founder of VisionTech Solutions who deploys AI agents in production. "A collaborator gives you time back. A liability takes your time and hands you the risk on top of it."
That liability typically isn’t quantified, because oversight does not register as work. People who validate outputs, catch hallucinations and carry the blame when something slips through are doing labor that never shows up in a usage count.
When Companies Reward the Wrong Metric
Yet usage is what a growing number of large employers now reward, leading to "tokenmaxxing." Companies including Amazon and JPMorgan, Meta and Disney have deployed AI usage leaderboards, in some cases prompting workers to burn through token budgets to climb them. One Disney employee prompted an AI assistant 460,000 times in nine days. Meta has formalized the pressure, announcing that "AI-driven impact" will be part of every employee's performance review from 2026, regardless of role.
That’s the Power User Trap, said Akilah E. Kamaria, founder and CEO of Lozen Advisory, an organizational strategy firm. The employees best at cleaning up machine output get rewarded with more of it, while no one measures the burden. "People are no longer being rewarded for judgment," she said. "They are rewarded for feeding the token machine." It is the lines-of-code problem reborn. Reward volume, and you get volume, regardless of whether it was worth producing.
The Rise of Judgment Debt
Every general-purpose technology has triggered some version of this panic. Calculators would kill arithmetic. Spellcheck would gut spelling. IDEs would breed coders who could not reason about their own work. Each time, the skill that atrophied turned out to be one the job no longer needed, and the worry reads in hindsight as nostalgia for obsolete effort.
The findings are an "early warning sign" rather than a verdict, suggesting that expectations around AI productivity may need recalibrating rather than reversing, said Julie Bedard, a managing director at Boston Consulting Group and co-author of the brain-fry study. However, this leaves the question open of which skills are becoming obsolete.
The erosion that worries practitioners most is not burnout, which is visible and recoverable, but the slow loss of judgment. It’s judgment debt, which compounds like technical debt, said Julian Baldwin, founder and CEO of leadership consultancy Cohere Leadership Group.
Each time a worker signs off on an output instead of building the reasoning behind it, the underlying skill gets less exercise. "For newer workers who never built that foundation, the gap opens faster," Baldwin said. "Companies are creating a structural problem and calling it an adjustment period."
This is what is different from the calculator analogy. Arithmetic was a mechanical step, and the judgment to know whether an answer was plausible was not automated. AI targets the interpretive layer that decides whether a fluent output is correct.
"Skill is built by struggle, and AI is very good at removing the struggle, which feels like a gift until you realize the struggle was the training,” said Scigliano. The risk is not a workforce that cannot use AI, but one that can only use AI, able to wave a conclusion through but unable to tell when it is wrong.
A Management Problem Disguised as a Tech Problem
So where, exactly, does assistance become hollowing-out? The threshold is not how much AI a person uses; tokenmaxxing leaderboards measure the wrong thing. It is when a worker can no longer reconstruct the reasoning behind the output they are approving.
That makes the problem work design, not technology. Firms have added automation to existing workflows without adding time to verify, escalate or decide, then asked the same worker to move faster, review more and be responsible for errors.
“Companies that invest in both the tool and the human end up with employees who can still catch the system when it's wrong,” said Baldwin. “The ones that don't, end up with workers who can operate it but can no longer think without it."
An agent should remove work, not relocate it onto a tired reviewer, agreed Scigliano.
Engaged, But Not Convinced: What's Eroding Beneath Your Strongest Engagement Numbers
Watch Now
Retention Is a Communication Problem: New Data from 6,000+ Employees
Watch Now
Do Learning Programs Really Work? How to Turn Education Into Engagement In Healthcare
Watch Now
Fuel Your Front Line: Workforce Trends Shaping 2026
Watch Now
The broader numbers are not encouraging. Most companies that adopted AI saw no meaningful uplift, according to an MIT study. A separate survey found that 40% of non-managerial white-collar workers said AI saved them no time at all. The technology isn't dictating these results, it's what organizations choose to measure, reward and protect. That is a management problem, not an AI one.
Editor's Note: A number of people are raising flags to make sure we preserve the human skills we need. Read more on this topic:
- AI Can Give HR the Answer. Can HR Tell if It's the Right One? — AI can generate HR insights in seconds, but without data literacy, teams won't know when to challenge the output. Upskilling and AI adoption must grow together.
- The Human Advantage: What AI Disrupts – and How to Restore It — Why the age of AI needs more human judgment, not less. Part four in a four-part series. Today: how to preserve accountability when AI shapes decisions.
- Digital Fluency: Designing Intelligent, Human-Centered Workflows — Workflow redesign is the single strongest predictor of scaled value from AI. When done intentionally, it helps employees and serves the business's bottom line.