Your AI Rollout Is Creating More Work, Not Less. Here's How to Fix AI Workload Creep.

A product manager on your team finishes a deck in half the time. Instead of taking a breath, she opens a backlog ticket she's been ignoring for weeks. Then she starts drafting a product brief that wasn't even assigned to her. By 4 p.m., she's done more work than any Thursday in memory. By month six, she's exhausted and her output quality is slipping.

I've watched this pattern play out in every AI rollout I've been close to. And now a UC Berkeley study put numbers to it. Researchers Aruna Ranganathan and Xingqi Maggie Ye spent eight months inside a 200-person tech company and found something that should make every leader pause: AI tools didn't reduce work. They consistently intensified it.

The researchers call it "workload creep." Employees absorbed more tasks, worked the same hours or longer, and eventually hit a wall of cognitive fatigue and lower-quality output. By month six, the initial productivity momentum had given way to burnout and decision paralysis.

This is a concern I've carried since day one of doing this work. And I think most leaders are walking right into it with the best of intentions.

The "phone scrolling" assumption is costing you AI workload creep you can't see

Here's a belief I hear from leaders constantly: "If we give people AI tools and they finish work faster, they'll just scroll on their phones. We won't actually get value out of it."

I get the instinct. But it misses something fundamental about how people relate to their work.

Most people don't want to be bad at their jobs.

When AI makes them faster, they don't reach for their phones. They reach for the next thing on the list. The proposal they've been putting off. The inbox they've been ignoring. Product managers in the Berkeley study started writing their own code. User researchers picked up engineering tickets. People filled knowledge gaps and widened their own job scope, with no formal expectation adjustment from anyone above them.

That's not laziness. That's ambition running without guardrails. And without a plan, it becomes a liability.

What the Berkeley researchers found hiding inside AI adoption

The study surfaced three specific mechanisms driving AI workload creep, and they're worth understanding because they're probably already happening on your team.

Task expansion without boundary adjustment. AI made new skills accessible. So people used them. The problem? Nobody recalibrated what "their job" actually included. Scope widened quietly and without any formal adjustment.

The disappearance of micro-breaks. One employee captured it perfectly: "With AI, the solution is always just one prompt away." People started prompting during lunch, between meetings, right before logging off. The small pockets of mental recovery that used to happen naturally just... stopped.

Multitasking overload. Workers managed multiple simultaneous threads: debugging one script while generating another via AI while running test suites in the background. It felt productive. It was actually a recipe for cognitive drain that compounded over weeks.

Here's what made this research land for me: the competitive advantage won't belong to the companies that use AI to run the fastest. It belongs to the ones that use it to run the longest.

Three things managers can do this week to prevent AI workload creep

This doesn't have to play out like the Berkeley study. But it does require you to be intentional about something most rollouts skip entirely: what happens after AI saves time.

1. Define what "time saved" is actually for. Before your team leans into AI tools, have an explicit conversation: "When AI saves you an hour, here's what we want you to do with that hour." Is it deeper work on existing priorities? Strategic thinking? Actual rest? If you don't answer this question out loud, your team will fill the gap with more tasks by default. Every time.

2. Protect the capacity you free up. If AI saves your team 30 minutes a day, resist the urge to immediately backfill that time with new deliverables. When savings always get re-consumed, adoption starts to feel like a treadmill instead of a benefit. Your people will notice. And they'll stop telling you about the time they're saving.

3. Check in on capacity, not just completion. Your one-on-ones shouldn't only ask "did you get it done?" Add a standing question: "Is your workload sustainable right now?" This surfaces creep before it becomes crisis and signals that you're paying attention to the human, not just the output.

Try This: A 10-Minute Capacity Check

In your next one-on-one with each direct report, ask these three questions:

1. "Since you started using AI tools, has your workload gone up, down, or stayed the same?"
2. "Are you taking on tasks that didn't used to be part of your role?"
3. "When AI saves you time on something, what are you doing with that time?"

Write down the answers. If more than half your team says their workload went up, you have workload creep. The good news: now you can see it. And you can start making intentional decisions about what "time saved" actually means for your team.

AI adoption without structure doesn't give people more capacity. It fills existing capacity with more work until something breaks. AI workload creep is predictable, and it's preventable. But only if you treat "time saved" as something worth protecting, not something to fill.

What would change on your team if you made that decision out loud this week?