The idea that individual AI proficiency is the key to organizational success has quietly driven most adoption strategies, and it's exactly backwards. Atlassian's research, led by Dr. Molly Sands, points to a more uncomfortable truth: your company isn't failing because people won't use AI, but because you're optimizing for solo speed while ignoring how work actually flows between people. The gap is stark. Nearly nine in ten executives told Atlassian that individuals are moving faster, yet only six percent can point to real ROI. Meanwhile, fourteen percent of teams have cracked it. That's not a technology problem. That's a coordination problem.
What separates those high-performing teams isn't better prompts or newer tools. It's what Atlassian calls a context graph, shared digital records of goals, decisions, and organizational knowledge that give AI the situational awareness it needs to be useful. Without that, you're just asking individuals to type faster into the void. The teams seeing real returns also redesigned entire workflows rather than accelerating isolated tasks. That distinction matters more than any model update. Speeding up people who are pointed in slightly different directions doesn't compound gains; it multiplies collisions. As Sands put it, they crash into each other. The fix isn't more AI literacy training. It's rethinking how work is structured and shared from the start.
The most practical insight here is the idea of AI working agreements. Before a project begins, teams decide not just what they'll use AI for, but what they'll deliberately avoid. They commit to shared agents and common skills. Atlassian found that this simple practice led to more AI usage, faster progress, and better decisions. It's a concrete, low-cost intervention any leader can steal tomorrow. And it addresses a hidden cost of individual experimentation: every worker building their own prompts and assumptions creates another layer of unspoken knowledge that never translates into team performance. That's a quiet tax on your organization that no dashboard will show you. Relatedly, our own reporting on Talking to My AI Clone Taught Me to Question the Tech and Verify Your AI's Understanding: A Simple Check for Tax Season suggests that individual trust in AI outputs is fragile and often misplaced. The lesson aligns: personal comfort with AI doesn't equal organizational reliability. You need shared context and verification, not just personal experimentation.
Here's what we'd tell a reader who asked us what to do Monday morning. Stop auditing how many people use AI. Start auditing how teams share context. Impose constraints, break work into the smallest practical units, commit to no hand-written code for a week, write down every decision in a shared record. Most of these moves aren't sustainable forever, but that's the point. They're fast learning loops. The teams that win won't be the ones with the flashiest AI stack. They'll be the ones who treat AI as a forcing function for better teamwork, revealing old management gaps that were always there. Watch for leaders who adopt explicit working agreements, not because they're trendy, but because they force the transparency that makes AI actually pay off. That's the metric that matters.