AI Working Agreements
AI Working Agreements is a one-hour team session for co-creating explicit norms on when, why and how the team uses AI. It is a play from the Atlassian Team Playbook: the team discusses what people need to feel confident and safe using AI, which standards apply to responsible use, and where AI should and should not help, then writes down what it agrees. A short confidence survey before and after the session shows whether the conversation reduced uncertainty.
Facilitation script
- 1
Share the goals of the session and run the baseline survey on confidence in how, when and why to use AI.
5 min - 2
Discuss opportunities and cautions, and collect current and wished-for uses of AI on the whiteboard.
10 min - 3
Area one: what people need to feel confident and safe using AI. Silent notes, then discussion.
12 min - 4
Area two: standards for responsible and effective use, such as checking output, disclosure and handling of data.
13 min - 5
Area three: where AI can help the team and where it should not. Draft the agreements, including knowledge sharing and how they are updated.
15 min - 6
Assign refinement tasks and repeat the survey to compare with the baseline.
5 min - 7
Agree where the agreements are documented, who posts the summary and when the first review takes place.
5 min
Tips
Start from real situations, such as a customer email drafted with AI or code generated by an assistant, and ask what the team expects in that case.
Check company policy beforehand so the team does not agree on something it is not allowed to do.
Give sceptics and non-users as much space as enthusiasts; the agreements must work for both.
Keep the list short enough to remember, with five to eight agreements as a good target.
Common pitfalls
Writing vague principles such as 'use AI responsibly'. Nobody can tell whether they are being followed and behaviour does not change
Agreeing norms that contradict company policy on data or tools, which puts team members at risk when they follow them
Letting the heaviest AI users write the agreements. Cautious colleagues stay silent in the session and ignore the result
Skipping the before-and-after survey, so the team cannot tell whether the session changed anything
Treating the agreements as finished. Tools and policies change within months and an unreviewed document is soon ignored
Variations
Atlassian suggests an asynchronous variant in which the team adds input to the template before a shorter live session, and a start-small variant that pilots the play with one project team before rolling it out more widely. Teams that already have general working agreements can run this as an add-on chapter instead of a separate document. The step timings in the source add up to slightly more than the stated hour, so plan 65 to 70 minutes or shorten the co-creation block.
Where it fits
When to use it
Team members use AI tools to very different degrees and nobody is sure what is acceptable
People use AI quietly because they do not know whether it is welcome
New AI tools have been rolled out and the team needs shared expectations on disclosure, review and data
A problem has occurred, such as unchecked AI output reaching a customer, and the team wants to prevent a repeat
A new team is forming and AI use is part of how it will work
When not to use it
The organisation has not yet decided which tools and data are permitted. That is a policy decision for leadership, legal and security, and a team cannot agree around it
The team needs general collaboration norms on meetings, communication and decisions. Run Working Agreements or a Team Charter first
Nobody on the team has used AI for work. Agreements will be guesses, so run an AI Innovation Day first and write the norms afterwards
A manager intends to announce rules already decided. Present them as rules; calling a briefing co-creation costs trust
The underlying issue is fear of job loss. Address that directly with leadership before discussing usage norms
Related methods
Frequently asked questions
What is AI Working Agreements?▾
AI Working Agreements is a one-hour play from the Atlassian Team Playbook in which a team co-creates explicit norms for its use of AI. The team discusses confidence and safety, standards for responsible use, and where AI should and should not help, and documents what it agrees. A short survey before and after measures the change in confidence.
How is it different from regular working agreements?▾
The mechanism is the same: the team writes its own norms and commits to them. Regular working agreements cover collaboration in general, such as meetings, communication and decisions. AI Working Agreements focus on one topic and add a baseline and follow-up survey.
What do typical AI working agreements cover?▾
Common topics are which tasks AI may be used for, what data may be entered into which tools, who checks AI output before it is used, when AI use is disclosed, and how the team shares useful prompts and lessons. Each agreement should describe a concrete behaviour. The content has to stay within company policy.
How many people can take part and how long does it take?▾
The play is designed for 2 to 10 people and a stated 60 minutes, plus the five minutes of preparation that Atlassian lists. Setting up the whiteboard with examples from your own team's work, as described in the first step, takes somewhat longer in practice. The individual step timings add up to a little over an hour, so 65 to 70 minutes is a safer booking. Larger departments should run it team by team.
How often should the agreements be reviewed?▾
Atlassian suggests a quarterly review. AI tools, company policies and the team's own experience change fast enough that agreements go stale within a few months. Review earlier when a new tool is introduced or when an agreement has clearly been broken.
Plan your next workshop with AI
Workshop Weaver helps you combine methods like AI Working Agreements into a complete, timed agenda in minutes.
Try it freeMethod descriptions on Workshop Weaver are original content written by our team, based on established facilitation practices. This method was inspired by work from Atlassian Team Playbook.