How a simple team ritual drove a 34% jump in AI alignment

The story of AI at work is often told in big, sweeping terms: platforms, transformations, enterprise rollouts. Those things matter. But day to day, what actually shapes behavior is what teams talk about regularly.

So instead of launching yet another AI program, Atlassian’s Teamwork Lab gave an AI upgrade to a ritual we already had: retrospectives. And because it’s our job to study how teams work, we ran this as an experiment with 33 teams across Atlassian, so we could capture and measure real results.

Read on to learn how we incorporated AI learning and behavior change into our retros and how you can adapt the format to use with your own team.

The repeatable play: Use your next retro to operationalize AI enablement

You don’t need a new meeting or an AI “initiative.” You can use the retro you already have. Here’s the step‑by‑step format we tested.

Step 1: Anchor in real work, not abstract AI

Retros are already grounded in real work, so use that to your advantage. Start by surfacing specific moments from the last sprint where AI was used, skipped, or could have helped, with questions like:

• Where did we use AI recently, and it worked well?
• Where could AI have helped, but we didn’t use it?
• What use cases, prompts, or patterns do we want to keep?
• What feels risky, off‑limits, or unclear for AI right now?

Encourage people to reference specific workflows. No use case is too small or insignificant. Did they use AI to draft or proofread an email or summarize a 20-page document into a few salient talking points?

“This line of conversation surfaces practical and reusable patterns that are grounded in your team’s workflows,” says Teamwork Lab researcher Ben Ostrowski. “And it reveals gaps and fears that might be quietly (but pervasively) slowing adoption and eroding confidence.”

In our AI Retros experiment, this opening reflection is where most of the “aha” moments came from: 82% of participants said they learned a new AI use case during the exercise — often something simple and immediately reusable, like using AI to turn messy meeting notes into a customer‑ready summary or to draft the first version of a performance review, then edit for nuance.

Step 2: Make AI learning social, not solo

Because AI is most often explored independently, it’s all too easy for folks to give up when they don’t get the results they were hoping for.

But as past Atlassian research has shown, when teammates (and especially team leaders) openly share their misfires as well as their wins, it makes it safer and more inviting for less‑confident colleagues to experiment, learn, and grow together.

During your AI retro, make a point to encourage people to share their frustrations or failed attempts with AI, not just what worked. As a facilitator, frame each story as something the whole team can learn from as opposed to a personal success or failure:

• “That prompt you used for summarizing worked well for you—how could others reuse it for customer calls?”
• “You hit a wall when you asked the tool for X. What could we try differently in that prompt next time?”

In the AI Retros experiment, this is where the conversation took off. Participants realized they were in a similar spot with AI, and used the time to vent, swap tactics, and problem‑solve together.

Step 3: Build a living AI use-case doc

By this point, you’ve surfaced what’s working and what could use some improvement. Now, it’s time to make those memories last. Capture learnings in a simple, shared document your team can update over time.

In the retro itself, co‑edit a doc with two core sections:

1. “AI helps us with…”
   • List real, concrete use cases:
     • Drafting and refining X
     • Summarizing Y
     • Generating options for Z
   • Include links or example prompts where helpful
2. “We avoid AI for…”
   • Sensitive data or customer specifics
   • High‑risk decisions that require human judgment
   • Areas where quality or nuance is critical and AI is still unreliable

In our AI retros experiment, teams used this doc as a kind of lightweight AI playbook. One team incorporated their “AI helps us with…” list into their new‑hire onboarding, so new teammates could see where AI is expected, not just allowed. Another used the “We avoid AI for…” list to unblock a long‑running debate about what data was safe to paste into tools and escalate a few gray areas to their leader for clear guidance.

You don’t need a complex template to make this work. A simple, shared page or doc is enough — as long as it’s easy to find, easy to edit, and treated as a living artifact that’s updated as new use cases emerge. For broader, more scalable impact, pair this document with your AI Working Agreements and bake it into onboarding so every new hire learns “how we use AI here” from day one.

Step 4: End with one tiny, specific AI commitment

Reflection is valuable. But change happens when teams commit to a clear next step.

Before you close the retro, ask each person (or the team as a whole) to make one small, specific AI commitment for the next sprint.

The goal isn’t perfection. It’s forward motion. Over time, these micro‑commitments stack into real behavior change.

The takeaway: Embed AI enablement into your everyday rituals

“We didn’t reinvent the wheel here. We basically just added some AI-powered LED lights to the spokes,” said Teamwork Lab researcher Sara Gottlieb-Cohen. “The best part about AI retros is that you don’t need to invent a new meeting or initiative.”

This is all without adding a single net-new recurring meeting. If it works well, you don’t even need “special” AI retros going forward. You can simply add an AI question (or two) to every retro to keep the conversation going.

Over time, AI reflection stops being an “initiative” and becomes part of how your team regularly inspects and adapts its way of working.