Ask most enterprise leaders why their AI investment hasn’t paid off, and they’ll point at the technology — the wrong model, the immature tooling, the integration that never quite landed. In our research at Atlassian’s Teamwork Lab, that’s rarely the real problem. The organizations struggling most with AI aren’t the ones with the worst tools. They’re the ones that treated a profound behavioral shift as a software deployment.

The numbers bear this out. Per our 2026 State of Teams report, 85% of knowledge workers report using AI, but only 29% have embedded it into how they work day to day. Handing someone a license changes what’s on their screen. It doesn’t change how they make decisions, coordinate with a team, or trust an output enough to act on it. Those are questions of behavior and culture, and they don’t resolve on a rollout timeline.

So the more useful question isn’t “which AI should we buy?” It’s “how do we get an organization to work differently?” That’s a change management problem, and no one has fully cracked it yet.

Atlassian has spent two years remaking how roughly 12,000 of our own people work with AI, and the early going was messier than any case study admits. Our first instinct was the common one: make the tools available, celebrate the early adopters, and assume usage would spread on its own. It didn’t.

Individuals got faster while teams stayed exactly as coordinated, or as fragmented, as before. Enthusiasts built clever workflows that never traveled past their own desks. Skeptics quietly opted out. We were measuring logins and calling it progress.

The turning point was structural, not technical. We moved AI enablement out of IT and put it under the same leader who owns our people function — a deliberate signal that this was about how humans work, not which software they run. That reframing did more for adoption than any single feature, because it put the change in the hands of people who understand behavior.

The organizations that make real progress drive AI adoption from the top and grow it from the ground at the same time, and the two halves depend on each other.

Top-down provides what individuals can’t manufacture alone: a clear reason the change matters, permission to spend time learning rather than only executing, and honest guardrails around what’s safe to try. When leadership sets a visible direction, experimentation stops feeling like a risk to someone’s job and starts feeling like the job. When our head of AI did a live demo of our internal tooling, it drove an enduring 90% increase in AI usage.

Bottom-up provides what strategy decks can’t: the actual use cases. The workflows worth scaling are almost never the ones leadership imagines in advance. They surface when someone close to the work solves a real problem and shows a colleague.

In two Teamwork Lab experiments, 75% and 82% of participants learned a new AI use case from their own colleagues rather than a prescribed list. Those ideas spread only because leadership had built the conditions, and the mechanisms, to carry them from one team to the next.

Top-down mandates alone produce compliance and dashboards full of shallow usage. Bottom-up alone produces islands of brilliance that never reach the org. The work of change management is building the bridge between them.

The same change program lands very differently depending on the culture it’s dropped into. Top teams are 3.2x more likely to say they’re encouraged to experiment and find ways AI works for them, per our 2026 State of Teams report. And per our 2025 AI Collaboration report, 37% of workers prefer not to let others know when AI is assisting them, mostly out of fear it will diminish their credit or be seen as cheating.

The implication most transformation plans skip: before rolling out AI, be honest about the culture you’re rolling it into. If experimentation isn’t safe today, that’s the first thing to fix — not the technology. The single most effective thing we did was reframe the goal as building a culture of experimentation, normalizing that most experiments won’t work, and that this is the point.

In practice, that can be as small as a micro-learning ritual. Ask your team to spend 15 minutes in a meeting (cameras and mics off) trying something new with AI, then come back and share what they did. Some will have something cool to show; others will have failed. Either way, the ritual rewards taking a risk together.

The companies using AI well are the ones with the right foundations already in place. Knowledge management isn’t a new problem, but AI is finally the carrot getting people to fix it. We once assumed AI would transform everything on its own. In practice, it’s the unglamorous hygiene — establishing the right knowledge, the right processes, the right communication skills — that creates transformation.

The pattern isn’t unique to Atlassian, which is the encouraging part.Amadeus, a giant in travel tech, recently migrated over 20,000 employees to Confluence Cloud Enterprise. The move freed up two full-time engineers from infrastructure maintenance, established Confluence as the company-wide source of truth, and paves the way for RovoAI tools to save employees an estimated 10% of their time.

For leaders sizing up their own AI programs, a few questions cut closer to the truth than any usage dashboard:

• Who owns this? If AI enablement lives entirely in IT, you’re managing a tool rollout, not a change program.
• Is experimentation actually safe here? Not in your values statement — in what happens to someone whose experiment fails.
• Can a good idea travel, and is it rewarded? When one team finds a workflow that works, is there a real path for it to reach the rest of the organization, or does it die at that team’s edge?
• What are you measuring? Usage tells you people logged in. It doesn’t tell you anyone changed how they work. Track speed, efficiency, quality, and innovation instead.

The enterprises that get real returns from AI over the next few years won’t be the ones that bought the most licenses or shipped the most features. They’ll be the ones who recognized that AI transformation is, at its core, a people transformation — and that nothing moves until the humans in the system decide to move it.