Atlassian is expanding usage-based charges for artificial intelligence and automation, adding a new layer to the familiar software subscription. The move reflects a wider problem for the industry: an employee count no longer shows how much work a company can run through its software once AI agents enter the picture.
The maker of Jira and Confluence said on September 1 that allowance limits and billing for most of its new usage meters will take effect on December 3, 2026. The meters cover Rovo AI credits, steps completed within automated workflows and some customer requests resolved by an AI agent without human escalation.
Atlassian is not abandoning subscriptions. Its announcement says most customers on paid cloud plans already receive an included allowance. Customers that need more can upgrade, buy credits in advance or pay for extra usage.
The result is a hybrid bill. A company still pays for access to the software, often according to the number of employees licensed to use it, but some automated work is measured separately.
Why the employee seat no longer tells the whole story
Software-as-a-service companies have traditionally sold “seats,” meaning licenses assigned to individual users. If a business needs 500 employees to use an application, it buys roughly 500 seats and can usually predict its annual cost.
An AI agent breaks the one-person, one-license relationship. In business software, an agent is a system that can carry out a sequence of tasks, such as answering a service request, updating a record or routing work, with limited human involvement.
One employee may direct an agent that performs a handful of tasks each month. Another may run thousands. Both occupy one seat, yet the computing resources they consume and the amount of work produced can differ sharply.
Microsoft described the same shift during its fiscal 2026 third-quarter earnings call. Chief Executive Satya Nadella said business applications were moving from a traditional seat model to “seats plus consumption.” Microsoft also said nearly 60% of its service customers were purchasing usage-based credits.
Salesforce offers another example. Its Agentforce pricing includes credits consumed when an agent completes actions, alongside conversation-based and per-user options. The menu of models shows that software companies have not settled on one standard unit for AI work.
Each pricing model moves uncertainty somewhere else
A seat-based subscription gives the customer a predictable bill. It also leaves the software provider exposed if a small number of licensed users generate unusually heavy computing demand, unless the contract imposes separate limits.
Usage-based pricing reverses part of that equation. The provider earns more as the customer runs more work through the system, while the customer has to forecast a bill that can move with activity.
Outcome-based pricing goes a step further. Atlassian says its Customer Service Management AI agent is billed for a successful resolution. In broad terms, the supplier takes on more performance risk because an attempt that fails or requires human escalation does not produce the same billable outcome.
That model can put the charge closer to a business result, but the contract still needs a precise definition of success. A closed service request, for example, may not always mean that the customer received a satisfactory answer.
Atlassian’s controls show why budgeting will become part of managing AI agents. According to its usage guide, administrators receive alerts at 80% and 100% of an allowance or a limit they have set. Extra usage is enabled by default and billed afterward, although an administrator can cap or disable it.
Lower waste does not always mean a lower bill
Usage charging could reduce one familiar source of software waste. As we recently reported, companies often keep paying for licenses that employees barely use or no longer need.
A meter can make the charge more closely reflect activity. A dormant AI workflow should consume little or nothing beyond any base subscription. A widely used agent, however, may generate a larger and less predictable bill than a fixed seat would have done.
Finance and technology teams will consequently need to examine more than user counts and login data. They will have to know what creates a billable event, how included allowances are pooled, whether unused credits expire, how retries and errors are treated, and what happens when a spending limit is reached.
The unit being purchased is also becoming part of the product decision. A per-action price rewards shorter workflows. A per-conversation charge favors predictable customer-service volumes. Payment per successful resolution places more of the commercial pressure on whether the agent finishes the job.
From December, Atlassian customers using the affected capabilities will start seeing this change in their bills. For the wider software market, the employee seat is unlikely to disappear soon, but it is no longer enough to measure work being done by people and agents together.
