Atlassian separates AI work agreements from role-based spending wallets

In an [October 29 TNW report](https://thenextweb.com/news/avani‑prabhakar‑atlassian‑ai‑super‑users‑tnw‑podcast), Atlassian’s Avani Prabhakar describes AI working agreements that define where each team can use AI. The interview took place at HumanX in Amsterdam on September 24. A work agreement answers whether AI belongs in a particular task and where a person retains the decision. A spending allowance answers a different question: how much use is authorized before another approval is needed. Neither answer can safely be inferred from the other. For a team setting up a workflow, the useful record is therefore more specific than a general instruction to use AI. It can identify the task, the information the tool may receive, the decision left to the employee and the person responsible for reviewing a proposed change. Those are prospective design questions, not an Atlassian policy document obtained by OpenTools.

TNW [reports](https://thenextweb.com/news/avani‑prabhakar‑atlassian‑ai‑super‑users‑tnw‑podcast) that employees receive a monthly AI wallet whose allowance depends on their role. Exceeding it requires approval and an explanation of the usage and the workflow it improves. A request for more allocation can explain what changed: a new work queue, a different task or more revision before an output can be accepted. That gives the approver a question to examine rather than treating a larger usage total as its own justification. It does not establish which allocation is right for another organization. The same distinction matters when reading an adoption score. In [our report on ProHance’s AI Adoption Index](https://opentools.ai/news/prohance‑ai‑adoption‑index‑approved‑use‑activity‑output‑evidence), permission to use a tool, activity and accepted work answer separate questions. That is reporting about another company’s measurement proposition, not evidence of an Atlassian deployment or a recommendation to buy the product.

Prabhakar also [argues](https://thenextweb.com/news/avani‑prabhakar‑atlassian‑ai‑super‑users‑tnw‑podcast) for monitoring token use rather than rewarding it, and against making AI usage a performance mandate. These are her reported management views, not evidence that a policy caused better results. A usage record can show that resources were consumed. To understand a workflow, a reviewer also needs to know what work was attempted, which outputs were accepted and where human revision remained necessary. More consumption can accompany more useful work, more rework or a change in the type of task; the total alone does not distinguish those explanations. The interview offers a set of operating choices to examine, rather than a controlled comparison of management policies. OpenTools did not conduct the interview, inspect Atlassian’s internal agreements, audit employee allowances or test an AI workflow. AI‑generated conceptual illustration of task permission, resource allowance and human review. It does not depict Atlassian’s staff, policy documents, office or software. Original artwork by OpenTools Team, informed by [TNW’s interview report](https://thenextweb.com/news/avani‑prabhakar‑atlassian‑ai‑super‑users‑tnw‑podcast).