AI was supposed to take on the drudgery. Workers say it’s taking on the good stuff, too.

The promise of AI at work has been consistent: let it handle the tedious parts of the job so people have more time for the work they find meaningful. But new research from Atlassian’s Teamwork Lab suggests the reality is more complicated.

In a survey of 1,013 U.S. knowledge workers, those who use AI say it’s taking on the work they enjoy at about the same rate as the work they dislike. And more troublingly, the deeper AI is built into how teams work, the more of the enjoyable work it’s taking.

AI is absorbing the good parts of the job as much as the bad

The common pitch for AI at work is that it frees people from tasks they’d rather not do. And for most AI users, that’s happening: 82% say it’s taken on at least some of the work they dislike, and more than half (53%) say it’s taken a moderate amount of that work, or more.

But the same percentage (82%) say AI has taken at least some work they enjoy, and more than half (51%) say it’s taken a moderate amount or more.

In the same vein, 10% of AI users say that AI has taken all of the work they enjoy, compared with just 3% who say that it’s taken all of the work they dislike.

“Getting the workforce on board with AI requires continuous, credible proof of how it benefits people,” says Molly Sands, future-of-work expert and head of the Teamwork Lab. “We promised AI would clear the busywork so people could focus on what they love. Instead, it’s taking the work they love, too. Nobody designed it that way. It’s just what happens when no one decides what to protect.”

The more deeply the team uses AI, the more likely they are to hand off the enjoyable work

How much AI takes on depends on how deeply people use it. Occasional users, those who use AI as a “quick fix” for one-off tasks, mostly hand off the work they dislike: 31% say AI handles a moderate or greater share of it, versus 24% for the work they enjoy.

Once AI becomes standard in a team’s shared workflows, the pattern flips. Among these workers, 58% say AI handles a moderate or greater share of the work they dislike, while 71% say the same about the work they enjoy.

Leaders are also far more likely to say AI has taken on all of their most enjoyable work. 6 in 10 leaders at the director level and above (60%) say AI handles a moderate or greater share of the work they enjoy, compared with 42% of managers and 36% of individual contributors.

Part of that split reflects how deeply leaders use AI; they’re twice as likely as managers to be heavy users. But this trend also means the people with the most say over how work gets redesigned are the ones whose own favorite work has shifted to AI the most.

“Leaders have handed off more of the work they enjoy than anyone else in the organization, and they’re the ones setting the norms for everyone else,” says Sands. “But it’s important to consider that what feels like a fair trade to a director may not feel that way to someone earlier in their career, who’s still building the skills that work develops.”

AI is lowering the barrier to specialized work, but expertise still matters

AI is lowering the barrier to expert work, but it’s not erasing the need for specialized expertise. Around six in ten AI users say AI has made it easier for people without deep expertise to analyze information (60%), review or improve work (60%), or produce a first draft (59%). That echoes earlier Teamwork Lab research showing that heavy AI users are twice as likely to handle specialized tasks without looping in an expert.

But the further the work gets toward the final call, the fewer people are comfortable with AI assistance. Forty-one percent say AI helps non-experts contribute meaningfully to final deliverables, and 28% say it helps them make decisions that used to require a specialist. 

When asked directly, workers assert they don’t expect expertise to fade. Contrary to popular speculation that AI will turn everyone into generalists, a majority of AI users (54%) expect it to create more specialists over time, saying expert judgment will matter more to guide the work and catch mistakes. That view is strongest among heavy AI users: 60%, compared with 50% of lighter users.

And yet here, too, the view from the top is different. Forty-eight percent of senior leaders at the VP level and above say AI lets non-experts make specialist-level calls, compared with 22% of those below VP.

“From the outside, it’s easy for leaders to assume AI lets anyone do an expert’s job,” says Sands. “But the people doing the work know which parts take real judgment. Before you decide what to hand off to AI, ask them. They’ll tell you which skills you can’t afford to lose.”

People want AI involved, but want to keep the judgment

Workers aren’t trying to keep AI out of their work. In every scenario we tested, most want AI to play a role, including about 9 in 10 for reviewing data, carrying out routine multi-step actions, and solving novel problems. Even for decisions with significant business, customer, legal, or people risk, two-thirds (66%) want AI involved.

What changes is who leads. For high-volume work, about two-thirds want AI in the lead, acting on its own or with human oversight: 68% for reviewing large volumes of data and 66% for routine actions. As the work calls for more judgment, that flips. Most want humans leading on:

• Novel or complex problems (58%)
• Deciding what to prioritize (66%)
• Deliverables that require deep expertise (67%)
• High-risk decisions (78%)

Few would hand any of it over completely. Just 3% would let AI make a high-risk decision on its own, and even for routine actions, only 18% would. This holds among the heaviest users: where AI is the team’s standard, 62% want humans leading on high-risk decisions, and 53% on deliverables that require deep expertise.

What teams can do

AI will keep taking on more of the work. The question is whether teams choose what it takes, or let efficiency decide by default. Three places to start:

• Make engagement and skill-building part of the design. When deciding what AI takes on, weigh whether the work keeps people engaged and builds skills the team needs, alongside speed and quality. Ask the people doing the work, not just the leaders redesigning it, which tasks matter most to them. If you give some of the enjoyable work to AI, it should be a deliberate choice.
• Match the human role to the stakes. Let AI lead high-volume work with human checks. Keep humans leading on novel problems, priorities, expert work, and high-risk calls, and make that split a team norm.
• Protect the practice that builds judgment. Some of the work AI can do is also how people build expertise. Keep enough of it with people, especially early-career employees, to sustain the skills the team will need.

“The heaviest AI users are the most convinced that expertise matters, and they’re handing AI the most work they enjoy,” says Sands. “But the work people enjoy is often how they built that expertise. If teams want it to last, they need to decide which work stays with people.”

---

Methodology

Between September 28 and October 1, 2026, Atlassian’s Teamwork Lab conducted a double-blind survey of 1,013 full-time U.S. knowledge workers on AI use, how AI is changing the work people do, who should lead different kinds of work, and what AI means for expertise. Respondents spanned industries, functions, seniority levels, and generations. Questions about AI’s role in work were asked of the 949 AI users only.