Atlassian's "Inside Atlassian" newsletter has been on an AI kick lately, and one line from a recent edition stuck with us: nearly half of knowledge workers say their AI knows about as much about their company as a brand new hire, at best. Feed it real context about how your team actually works and it produces up to 48% more output you can use without revision.
That framing is doing a lot of work. A new hire isn't a friend or a foe, they're just new. Keen, occasionally overconfident, and only as useful as the context you've given them. Sound familiar? It's basically how most of us are using AI right now, minus the induction.
The AI Play/Pause experiment
Atlassian ran a nice piece of research to test how deep that reliance actually goes. 123 Atlassians did an "AI Play/Pause" day: one workday using AI as much as possible, one workday with none at all.
On the AI day, priority work felt 44% easier and stress was 22% lower. On the no-AI day, 82% of people felt the urge to reach for it anyway, and 23% did without even thinking. That's not really a productivity story, it's a habit story. And afterwards, 89% of teams said they came out with a clearer shared sense of when they should and shouldn't be using it.
That last bit is the interesting part. Not "should we use AI", but "do we actually agree on when we should." Most teams have never had that conversation out loud, they've just each landed on their own answer separately.
The fragmentation tax
The other Atlassian stat worth sitting with is from their 2026 State of Teams report: 89% of executives say AI has sped up individual work, but only 6% can point to specific, organisation-wide examples of AI ROI. They call the gap the "fragmentation tax", the coordination cost of everyone getting individually faster in slightly different directions. Estimated at $161 billion a year across the Fortune 500.
We've seen a version of this ourselves. A few weeks back, on one of our Thursday dev-time sessions, we got talking about an all-hands where leadership had set out a vision for engineers to be "multi-agent" within the year, running several agents at once rather than writing code line by line. The room split roughly in two. Some people had already reframed themselves as problem solvers directing tools rather than typists, and were asking things like "how much are you going to spend on me per month, a thousand, two thousand?" Others were much closer to "I didn't get into this job to hand my code to a machine." Same team, same tools, genuinely different working models sitting side by side. That's the fragmentation tax in miniature, individual speed with no shared agreement underneath it.
Where AI actually fits in a people-first framework
Here's the thing though. PETALS was built around five feeling factors, Productivity, Enjoyment, Teamwork, Learning and Serenity, and every one of them is about people, not output. So where does AI fit into that?
Not at the top. AI isn't a sixth petal, and it's not a priority to optimise for its own sake. The moment "are we using enough AI" becomes the question a team health check is trying to answer, something's gone sideways. The actual question is always the people one, do you feel supported, are you learning, is the work sustainable, does the team trust each other. AI is, at best, one of the things that can help or hurt those answers, not a goal in itself.
Where it does earn a place is as an assistant. A very capable, occasionally overeager one that needs the same onboarding you'd give any new hire, context about how your team works, and clear agreement on when it's appropriate to lean on it and when it isn't. Treated that way, it's genuinely useful. Treated as a shortcut around the harder, human parts of teamwork, it just moves the friction somewhere less visible.
Atlassian's own "AI Pulse" newsletter made a similar point the day after their context research landed: "getting faster with AI doesn't actually make your work better." Speed isn't quality, that comes down to the person directing the work, their subject-matter expertise, judgment and taste, and their ability to spot when something doesn't add up. Which is really the whole argument for treating AI as an assistant rather than a priority, if the human doing the judging isn't in the loop, faster just means you get to a worse answer sooner.
A few real-world shapes this takes:
Friend: someone uses AI to get a rough first draft or scaffold out of their head faster, then does the actual thinking themselves. It clears the boring bit so the human bit gets more attention, not less.
Foe: someone reaches for AI as a substitute for a conversation they should be having with a teammate, asking it to write the awkward Slack message or draft feedback they haven't worked out how to say themselves. It papers over a Teamwork or Serenity problem instead of surfacing it.
Enthusiastic junior: AI produces something confident, fluent, and subtly wrong because it didn't have the context a real teammate would have picked up from six months in the room. Fast doesn't mean right, and it takes someone with actual domain expertise to spot the difference. Useful, but needs the same review you'd give any new starter, not a rubber stamp.
Why the A in PETALS isn't AI
We get asked this a fair bit, usually half-jokingly: shouldn't the A in PETALS stand for AI? It's a tempting bit of wordplay, but no, and it's a deliberate no rather than an oversight.
Every existing petal is a feeling. Productivity, Enjoyment, Teamwork, Learning, Serenity, they're all about how someone experiences their work. AI isn't a feeling, it's a tool. Bolting it on as a petal would mean measuring tool adoption alongside human wellbeing, and those aren't the same kind of thing. If you genuinely want to track AI usage, there's a whole category of DX tooling built specifically for that, and they'll do a far better job of it than a team health check ever could.
There's a sharper reason too. AI is one of the more divisive topics in a lot of teams right now, some people are all in, some are actively resentful of it, and most are somewhere uneasy in between. PETALS only works if people feel safe enough to be honest about how the job actually feels, and a dedicated AI score risks turning that into a referendum, or worse, a number that gets used to justify a decision someone's already made about who is or isn't "using AI enough." That's exactly the kind of thing that shuts honesty down rather than opening it up.
So AI stays where it belongs in this framework, a thing that can show up inside Teamwork, Serenity or any of the others depending on how it's being used, not a category with its own scoreboard.
Where this lands for PETALS
If you're running a PETALS check-in with your team, AI is worth a mention, but as a contributing factor, not a category of its own. Worth asking directly: are we agreed on when we use it and when we don't, the way Atlassian's Play/Pause teams ended up doing? Is it quietly propping up a Teamwork or Serenity score that would look worse without it? Is anyone using it to avoid a conversation rather than have one?
None of those are AI questions really, they're team health questions that happen to have AI sitting inside them right now. Which is exactly why a people-first framework doesn't need a bolt-on AI pillar to make sense of it, the same five feeling factors already cover it, you just have to remember to ask.
This is the kind of thing our newsletter subscribers hear about first. If you'd rather get product updates and insights like this straight in your inbox instead of waiting for the blog, subscribe below.