Ask three people on one of your teams whether they may use AI for the same task. If their answers differ, check which rule each person is following.

Picture conflicting speed-limit signs at the same junction. The problem is knowing which instruction applies. AI governance means deciding how AI may be used, who is accountable and how those decisions are checked. Clear rules are one part of it. Atlassian’s survey gives you a reason to investigate confusion, but it does not show that having no rules is better.

What Atlassian found

Atlassian’s report, The State of Developer Experience in 2026, draws on 1,155 developers and 405 engineering leaders. It asked whether their organisation has a clearly communicated, actively implemented AI strategy across the delivery lifecycle: from planning work to running it. Only 17% of developers said yes, and 23% of leaders.

It compared the share of developers in each group reporting noticeably less friction in their work than before AI. Think of friction as the obstacles and interruptions that make work harder:

  • 41% where the strategy is clear and actively implemented
  • 29% where there is no formal strategy, or they are not sure
  • 24% where a strategy is in progress
  • 22% where it exists but is inconsistent or poorly communicated

The last group reported the least relief. But the comparison is about AI strategy, which is broader than a policy on tool use. The second group also combines respondents with no formal strategy and those who do not know whether one exists.

A related finding concerns cost management. Among developers with a clear, active strategy, 50% said their organisation has clear visibility into AI-tool costs and actively manages them, against 11% where there was no formal strategy or they were unsure.

How far to trust it

This is a vendor survey of self-reported experience, not measured delivery. It covers different organisation sizes, industries and geographies, but gives no group sizes or uncertainty estimates for these comparisons. The methodology does not state whether the findings underwent independent peer review.

Atlassian says its associations, including this one, “should not be interpreted as causal proof”. Organisations that communicate clearly about AI may have other practices that explain the differences. These figures do not establish that a muddled strategy causes friction, or that abandoning rules would improve delivery.

What AI governance should make clear

The useful question for a leader is whether people can apply the decisions already made. Check your existing AI coding rules against these questions. AI coding guidelines should help people find the answers:

  1. Which work is allowed? Name permitted uses and restrictions, including which tools are approved for which tasks.
  2. What must a person check? Say who owns the decision to release AI-assisted work. The article on what code review is for when agents draft the code explains why that ownership matters.
  3. Which data may be used? Point to the approved restrictions and who can clarify them.
  4. Who decides when the answer is unclear? Give people an accountable owner and a way to ask.
  5. How do changes reach the teams? Make the current version easy to find and explain what changed.

Use a common baseline across teams, with explicit differences where the task or risk requires them. Different rules for different work can make sense. Contradictory answers about the same work need investigation. Avoid turning clarity into a target for more AI use; how to report AI productivity honestly explains that measurement problem.

Try it with one team this week

Ask three people separately whether AI is allowed for the same practical task, what data they may use and who checks the result. Compare their answers with the current approved rules. Imagine a team where two of three people are unsure whether customer data may go into an AI tool. That is a question for the accountable policy owner, not permission to choose the most convenient answer.

Have that owner resolve any contradiction. Summarise the agreed rules on one page, with links to the authoritative documents, and discuss it with the team. The summary must reflect existing permissions; it cannot grant new ones.

Before the week ends, ask the people to apply the summary to a similar task. Can they explain what is allowed, what needs checking and where an unresolved question goes? Agreement alone is insufficient if everyone repeats the wrong rule. Check their answers with the owner before extending the exercise across your teams. This tests understanding; it does not yet demonstrate faster delivery.

For a wider view of where AI is helping or straining delivery across your teams, the free AI Delivery Health Check gives you a score, the area to focus on first and one thing to try this week.