AI moved the bottleneck

I read Atlassian's State of Teams 2026 report and one thought stuck with me:

AI makes parts of a company faster. It does not automatically make the company faster.

That sounds obvious, but I think this is where a lot of the disappointment comes from.

A developer can create a lot of code in an afternoon now. A product person can turn a vague idea into a prototype. A designer can get a first UX flow without starting from a blank page. A founder can build something that looks like an MVP before anyone has opened Figma.

Great.

But the work is not done when the first version exists.

Someone still has to understand it, review it, test it, integrate it, release it, explain it, and sometimes undo it. So if one part of the system gets much faster and the rest stays the same, the bottleneck has not disappeared. It has just moved.

Maybe AI does not only make teams faster. Maybe it also makes the broken parts of the project lifecycle more visible.

Roles get blurrier

One prediction in the report is that teams will become more horizontal, with fewer layers and more blended roles.

I buy that.

AI lowers the bar for being useful outside your official role. Not great. Useful.

Why wait three days for a first UX idea if Claude can help you get a half-good one in a minute? Why hand over only a Figma design if Lovable lets you build a working prototype that engineering can actually click through? Why should a product person only write tickets if they can also explore the data, draft the copy, and test a few flows?

This does not mean expertise matters less.

The designer is still better at knowing whether the flow is good. The engineer is still better at knowing whether the prototype can survive production. The product person is still better at knowing whether this solves the right problem.

But the rough first step between roles gets cheaper. That should change how teams work.

Managers should stop acting like task routers

Atlassian says 87% of knowledge workers feel teams lack the time or capacity to coordinate because everyone is in execution mode.

That feels very believable.

If everyone can execute faster, managers probably need to spend less time assigning sequential tasks and more time making the direction painfully clear.

The "what" can often be explored with AI. Write the draft. Build the prototype. Generate options. Create the first version.

The manager's job is to make sure all of that activity points somewhere useful.

So maybe every person on the team should be able to answer four boring questions before work starts:

What problem are we solving?
What change are we trying to drive?
How will we measure success?
Who is responsible?

This sounds basic because it is basic. But it becomes more important when everyone has faster tools.

If the goal is vague, AI gives you more vague output.

If the metric is missing, everyone can be busy and nobody can tell whether anything improved.

If ownership is unclear, agents will not fix it. They will just produce more stuff for nobody in particular.

The useful boring work

The report also talks about context, workflows, and culture. I usually dislike this kind of three-word list. It sounds like something that should be on a slide.

But the practical version makes sense to me.

Write down what good looks like. Keep a useful CLAUDE.md. Maintain review checklists. Document product principles. Capture decisions so the same conversation does not happen every three weeks.

And let agents handle the annoying coordination work where possible.

Status updates. Ticket summaries. Release notes. Pull request descriptions. "What changed since yesterday?" checks. Comparing a branch against the acceptance criteria. Noticing that the docs no longer match the product.

None of this feels exciting. But it is often the difference between "done" and actually delivered.

So maybe the question is not: how do we make every individual faster?

Maybe the better question is: how does the whole lifecycle need to change now that creating the first version is cheap?

If the lifecycle is unclear, AI will mostly create more unclear work. If the review process is weak, it will create more work waiting for review. If the context is bad, it will produce something that looks right and still does not fit.

So that is the part I want to think more about.

Not "how do we make developers faster?"

"What happens after the first draft exists everywhere at once?"