AI Tools Can Intensify Work Instead of Reducing It

Here's a finding that should change how you think about AI adoption:

At a company where employees adopted AI tools voluntarily, work got more intense, not lighter.

That's not from an anti-technology think tank. It's from an eight-month field study by UC Berkeley Haas researchers Aruna Ranganathan and Xingqi Maggie Ye at a U.S. tech company of about 200 employees, published in Harvard Business Review in February 2026. They used on-site observation and more than 40 interviews, not a survey. Employees worked at a faster pace, took on a broader scope of tasks and extended work into more hours of the day.

The people who leaned into AI took on more, not less.

The Paradox in the Data

Other data points in the same direction:

• 77% of employees using AI say it has added to their workload (Upwork survey, July 2024)
• 41% rise in mentions of "fatigue" on the Glassdoor Community in 2025 (Glassdoor)

These are different kinds of evidence (a survey and a count of forum posts, alongside the field study), so they don't prove a single cause. But they point the same way: AI adoption often comes with more cognitive burden, not less.

The tools work. The outputs are often good. But something else is happening alongside the productivity gains.

What AI Actually Adds to Your Day

Every AI tool you adopt adds a layer of decisions that didn't exist before:

The email gets written. Maybe faster, maybe not. But either way, you've made 5 decisions that didn't exist before.

Multiply that across every AI-assisted task in your day.

Hours worked and task volume don't capture this. The number of decisions required per day may matter as much.

The "Quiet Burnout" Pattern

There's an emerging term for what heavy AI users experience: quiet burnout.

Unlike traditional burnout — which comes from sustained overwork and shows clear symptoms like cynicism and detachment — quiet burnout is subtler:

• You're technically productive
• Your output looks fine
• You're not working excessive hours
• But you're exhausted by 2 PM
• You avoid starting complex tasks
• You default to whatever requires the fewest decisions

Quiet burnout doesn't trigger the alarms that traditional burnout does. Your manager sees normal output. Your calendar looks manageable. But your cognitive capacity is depleted from a thousand micro-decisions you never consciously registered.

Why "Just Use Fewer Tools" Doesn't Work

The obvious advice — reduce your AI toolkit — misses the point.

The problem isn't individual tools. It's the cumulative decision layer that AI adds to knowledge work. Even a single well-chosen AI assistant adds:

• Prompt decisions: How to frame what you need
• Evaluation decisions: Whether the output meets your standard
• Integration decisions: How to blend AI output with your work
• Trust decisions: When to accept AI judgment vs. override it

A 2026 NBER survey of nearly 6,000 senior executives in the US, UK, Germany and Australia found that about nine in ten reported no impact of AI on productivity or employment at their firms over the previous three years.

The survey doesn't say why. One possibility worth testing is that output gains are being absorbed by the overhead of managing AI interactions.

What the Research Points Toward

The UC Berkeley researchers recommend that organizations build an "AI practice": intentional pauses, sequencing work instead of reacting to every AI output, and protected time for human connection. Our own reading is that this also means measuring the cognitive cost of AI adoption, not just the output gains.

Most AI ROI calculations count:

• Time saved per task
• Output volume increases
• Error reduction rates

Almost none measure:

• Decision load added per AI interaction
• Cognitive overhead of tool switching
• Recovery time between AI-assisted tasks
• Cumulative mental fatigue from evaluation decisions

This is a measurement gap. The positive side of AI is measured carefully. The cognitive cost side is barely measured at all.

The Measurement Question

If AI tends to intensify work, and decision load is a plausible mechanism, then the practical question becomes: can you measure your own decision load?

Not in a clinical sense. Not as a diagnosis. But as a simple baseline number — how many decisions are competing for your attention right now?

The concept isn't new. Researchers like Baumeister, Danziger, and Gloria Mark have studied decision fatigue for decades. What's new is the urgency: AI is accelerating the problem faster than most people can adapt.

What We'd Suggest

If any of this resonates:

1. Count your AI decisions for one day. Every time you prompt, evaluate, or override an AI tool, note it. Most people are surprised by the number.
2. Notice when you avoid complexity. If you're defaulting to simple tasks by mid-afternoon, that's a signal — not laziness, but depleted decision capacity.
3. Get a baseline. You can measure your current decision load in about 5 minutes. No signup, no email required. Just a number that tells you where you stand.

References

CTE Research explores the intersection of cognitive load and knowledge work. All statistics are attributed to their original sources.

Corrections, 28 September 2026: An earlier version said a UC Berkeley/HBR study found 83% of AI power users reported increased workload; the study (Ranganathan & Ye, HBR, February 2026) is a qualitative field study at one company of about 200 employees and contains no such figure, so the title, headline and figure were changed. The Upwork figure was corrected from 80% (2026) to 77% (July 2024). The WalkMe 71% figure and the Deloitte claim that decision fatigue is the number-one burnout indicator could not be traced to a primary source and were removed. The Glassdoor figure was corrected to refer to Glassdoor Community posts in 2025, not reviews. The NBER study was described as showing zero executive productivity impact exactly offset by AI overhead; it is a survey of about 6,000 executives in which about nine in ten reported no firm-level productivity or employment impact, and it draws no conclusion about AI overhead. The UC Berkeley team's recommendation was restated to match what they wrote.