AI moved the bottleneck

I read Atlassian's State of Teams 2026 report recently. It's based on surveys with more than 12,000 knowledge workers plus Fortune 1000 executives, so there's some weight behind it.

The short version: AI makes individuals faster, but it doesn't automatically make teams faster. Atlassian calls the time lost in between the fragmentation tax - duplicated work, conflicting priorities and coordination overhead eating up whatever the speedup brought.

The numbers are striking. 89% of executives say AI has made work faster. Only 48% say it has improved collaboration. And just 6% can point to clear AI returns across their organisation.

That doesn't surprise me. Developers can produce way more code than before, product people turn ideas into clickable prototypes in an afternoon, designers explore five flows in the time one used to take.

But producing something isn't finishing it. Somebody still has to review it, decide whether it actually solves the right problem, integrate it, test it and release it. If that part stays just as slow, more output mostly means a longer queue.

The fragmentation tax

Atlassian splits work into execution and coordination, and AI is getting much better at the first than the second:

87% of knowledge workers say they often lack the time or capacity to coordinate, because everyone is busy executing.
84% deal with unclear or conflicting goals and priorities.
71% discover other teams working on duplicate projects or initiatives.

The report also has a name for low-quality AI output that creates extra work for somebody else: workslop. Almost half of those surveyed say their AI output isn't reliably high-quality.

I found this the most useful part of the report. AI is basically a stress test for how a team already works. If reviews were slow, you now get a longer queue. If nobody owned a decision, you get more unfinished work floating around. And if the context you feed it is bad, you get output that sounds plausible but doesn't fit your product.

AI use itself is already everywhere. 85% of knowledge workers use it, but only 29% have embedded it into their daily work in some structured way. More than half of the executives say AI is widening the gap between teams.

So a few power users being very fast isn't the same as the company getting better at delivering.

Teams and roles will change

The report expects teams to become more horizontal. 77% of executives expect fewer layers, 73% expect more blended roles.

I think that's plausible. When building something yourself gets cheap, it also gets cheaper to be useful outside your official role. A product person can build a rough prototype instead of writing another spec. A designer can test an interaction with real data. An engineer can have an opinion on copy.

That still doesn't make everybody interchangeable. The expertise decides whether any of it is good. But the first step into somebody else's area becomes cheaper, so a team can walk around the thing instead of arguing about it abstractly for three meetings.

The downside is that there are more dependencies to keep track of. And once agents join the mix, people become what the report calls digital air traffic controllers, manually keeping all the work in sync.

What the best teams do differently

Atlassian found that 14% of teams are getting both speed and better coordination out of AI. They focus on context, workflows and culture - which honestly sounds like slide material until you make it concrete.

Write down what good looks like. Keep a useful CLAUDE.md. Record architecture decisions somewhere people can find them. Make acceptance criteria and ownership clear, so review isn't a surprise party.

Then let agents do the boring coordination work: status summaries, release notes, pull request descriptions, checking things against acceptance criteria.

The main point of the report isn't that teams need more AI. Most people already use it. They need to redesign the work around it.

That leaves me with a much better question than "how fast can I produce the first draft?"

Did we actually ship anything sooner?