Why AI Makes Individual Workers Faster But Leaves Whole Companies Behind

Key points

• Atlassian's 2026 State of Teams Report surveyed 12,000 workers and found 89% of Fortune 1000 executives say AI is speeding up individuals at their companies.
• Only 6% of those same executives could point to a specific, concrete example of financial return from AI.
• Roughly 14% of teams inside those organisations had managed to turn AI use into real, measurable value.
• Teams that succeeded shared three traits: a shared store of organisational knowledge, redesigned workflows, and a culture that openly welcomed failed experiments.
• Atlassian tested "AI working agreements", written team rules about how and when to use AI, and found they improved speed, decision-making, and output quality.

Imagine a relay race where each runner gets faster shoes, but nobody agrees on the route. That is roughly what is happening inside most companies using AI right now, according to Dr. Molly Sands.

Sands leads the Teamwork Lab at Atlassian, the company behind project-management tools like Jira and Confluence. Her team of behavioural scientists studies how AI is changing the way people work together, then goes into organisations and actually changes things. She shared the lab's findings at VB Transform 2026, first reported by VentureBeat.

So why isn't AI paying off?

The short answer: companies are training individuals, not teams. The tools get faster. The coordination stays broken.

The numbers back this up sharply. Atlassian's annual State of Teams Report, which surveyed 12,000 workers worldwide and interviewed roughly 200 executives at Fortune 1000 companies (America's 1,000 largest by revenue), found an uncomfortable gap. Nearly nine in ten executives said people at their companies were working faster with AI. Just 6% could name a specific place where that speed had turned into real financial return.

About 14% of teams, though, had cracked it. What set them apart came down to three things.

Shared knowledge. Winning teams stored goals, decisions and context in tools everyone could see, rather than keeping everything in individual heads or private chat threads. Atlassian calls this a "context graph", essentially a living map of who is doing what and why, that AI tools can read and use. Without it, an AI assistant only knows what the person in front of it knows.

Redesigned workflows. The high-performing teams did not just speed up existing tasks. They rebuilt whole processes from scratch. Sands put the risk of skipping this step bluntly: make individuals faster without aligning them, and they "very quickly start to crash into each other."

A culture that allows failure. The fastest teams had leaders who made experimentation feel safe. Some experiments would fail. That was the point.

What did the winning teams actually do differently?

They imposed deliberate constraints to learn fast. One example: a team committed to writing zero lines of code by hand for a week, forcing everyone to work through AI tools. Unsustainable long-term, Sands said, but a rapid way to build shared skill and surface what works.

They also wrote down rules before starting projects. Atlassian calls these "AI working agreements", a short document where the team decides together what they will use AI for, what they will keep doing themselves, which AI agents they will share, and what basic skills everyone needs. Teams that adopted this practice moved faster and produced better work.

The deeper lesson is not a new one. Hidden assumptions and misaligned mental models have slowed teams down long before AI existed. AI does not create those problems. It just makes them more expensive to ignore.

Common questions

Does this mean AI tools are a waste of money if teams aren't ready?

Not exactly. Individual workers do get genuinely faster, and that has some value. The problem is that speed without coordination rarely translates into the kind of return that shows up on a balance sheet.

What can a manager do starting this week?

Hold a short meeting before the next project and write down, together, which tasks the team will use AI for and which they won't. That single conversation closes more gaps than hours of individual AI training.