Nobody Decided What to Protect: AI, Enjoyable Work, and the Work Design Gap

17 hours ago

5 min read

AI was supposed to take the tedious parts of the job and leave people the work they find meaningful. New research from Atlassian's Teamwork Lab finds that it is taking both: 82% of AI users say AI now handles at least some of the work they enjoy, the same share that say it handles work they dislike.

That is not a technology failure. It is a work design failure. When nobody explicitly decides which work goes to AI and which work stays with people, the decision gets made by default, one convenient handoff at a time. Work management exists to make those decisions on purpose.

What the research found

The deeper AI is built into a team's workflows, the more of people's favorite work it absorbs. Atlassian surveyed 1,013 full-time U.S. knowledge workers between September 28 and October 1, 2026, and asked the 949 AI users how AI is changing their work. The headline numbers:

Finding	Result
AI has taken at least some work users enjoy	82%
AI has taken at least some work users dislike	82%
AI has taken all of the work users enjoy	10% (vs. 3% for disliked work)
Moderate-or-greater share of enjoyed work handled by AI, teams where AI is standard	71%
Same measure, occasional AI users	24%
Director-level+ leaders saying AI handles a moderate-or-greater share of work they enjoy	60% (vs. 42% managers, 36% ICs)
AI users who expect AI to create more specialists	54%
Want humans leading high-risk decisions	78% (only 3% would let AI decide alone)

The pattern flips as adoption deepens. Occasional users mostly hand off the chores. Once AI becomes the team's standard way of working, it takes a bigger share of the enjoyable work than the disliked work.

Source: Atlassian Teamwork Lab, "AI was supposed to take on the drudgery. Workers say it's taking on the good stuff, too." (November 4, 2026)

Why it happens: delegation by default

AI absorbs enjoyable work because nobody owns the decision about what it should absorb. Molly Sands, who leads the Teamwork Lab, put it plainly: the outcome is what happens "when no one decides what to protect."

In most organizations, AI adoption follows the path of least resistance. A tool is available, a task is easy to hand off, and the handoff sticks. Repeat that across a team for six months and the workflow has been redesigned without anyone designing it.

This is the same failure pattern behind most broken workflows, just faster:

No structural clarity. Nobody has mapped which steps exist, so nobody can see which ones moved to AI.
No explicit handoffs. Work shifts from person to AI without a defined point where ownership changes.
No decision transparency. The choice to automate a task is never recorded, so it is never revisited.

These are three of the 7 Workflow Architecture Standards. AI does not create new work design problems. It exposes the ones that were already there.

The people redesigning work are the ones whose own work has changed most. Leaders at director level and above are far more likely than individual contributors to say AI handles their enjoyable work, and Atlassian found they are twice as likely as managers to be heavy AI users.

They also see expertise differently. 48% of VP-and-above leaders say AI lets non-experts make specialist-level calls. Only 22% of everyone below VP agrees.

That gap matters because leaders set the norms. A trade that feels fair to a director, who built their judgment years ago, may cost an early-career employee the very reps that would build theirs. Work design decisions made only from the top will consistently underestimate what the people doing the work stand to lose.

Match the human role to the stakes

Workers already know where the line should sit: AI leading on volume, humans leading on judgment. Nearly everyone wants AI involved, but who leads changes with the stakes.

Type of work	Who workers want leading	Share
Reviewing large volumes of data	AI (alone or with human oversight)	68%
Routine multi-step actions	AI (alone or with human oversight)	66%
Novel or complex problems	Humans	58%
Deciding what to prioritize	Humans	66%
Deliverables requiring deep expertise	Humans	67%
High-risk decisions	Humans	78%

The problem is that most teams never write this split down. It lives in individual habits, so it drifts.

WMI's AI Workflow Governance model treats this as a design responsibility with three components:

Explicit Delegation. Every step in a workflow has a named performer (human, AI, or human-checked AI), decided on purpose and recorded.
Reference Alignment. AI works from the same approved sources, standards, and definitions the team does.
Drift Detection. Someone owns the signal that tells you when the actual split has moved away from the designed one.

Atlassian's data shows what happens without the third component. Delegation creeps from chores into craft, and no one notices until people say the work they liked is gone.

Protect the work that builds judgment

The work people enjoy is often the work that made them experts. Drafting the analysis, wrestling with the messy problem, making the first call and getting it wrong: that is how judgment forms.

Workers seem to sense this. 54% of AI users expect AI to create more specialists, not fewer, because expert judgment will matter more for guiding AI and catching its mistakes. Heavy users believe it most strongly (60% vs. 50% of lighter users). Yet those same heavy users are handing off the most enjoyable work.

That is a contradiction a team can only resolve by design. If expert judgment is the scarce resource of the AI era, the practice that produces it is a workflow asset. Treat it like one:

Identify the steps in each workflow that build skill, not just the ones that produce output.
Keep a deliberate share of those steps with people, especially early-career team members.
Use AI as a reviewer or second draft on those steps rather than the first performer.

Five questions before AI takes a task

Before any step in a workflow moves to AI, a team should be able to answer five questions. If it can't, the handoff isn't designed yet.

What is the stakes level? High-volume and routine can be AI-led with human checks. Novel, prioritization, expert, or high-risk work stays human-led.
Does this step build a skill the team needs? If yes, decide who keeps practicing it.
Who did we ask? Get input from the people doing the work, not only the leaders redesigning it.
Where is the handoff recorded? The delegation should be visible in the workflow, with a clear point where ownership changes.
Who owns the drift signal? Name the person who will notice if AI's share of the work grows beyond what was designed, and review it on a set cadence.

These map directly to the IDEAS Workflow Ownership Model: set the intent, design the split, execute, check alignment, and assign a signal owner.

The real job is deciding

AI will keep taking on more work. The question Atlassian's research raises is whether teams choose what it takes or let convenience choose for them.

Every knowledge worker has two jobs: the work of their function, and the work of managing that work. AI is getting very good at the first. The second, deciding what work exists, who or what performs it, and what must be protected, is becoming the job that matters most.

The most important skill of the AI age will not be using AI. It will be managing work.

Further reading from the Work Management Institute: