Software subscriptions have traditionally been easy to picture: a team hires another person, buys another seat and adds another monthly charge. AI complicates that relationship. A small team can potentially run a large volume of automated work without adding employees.
On September 1, Atlassian announced an evolution of its usage-based pricing model for AI capabilities. The company said customers would receive visibility into consumption and spending, with built-in allowances on most paid cloud plans and controls that administrators could adjust. These are descriptions of the company’s announced model, rather than an independent assessment of customer savings.
GlobalFounderTimes analysis: the interesting decision concerns the unit being sold. A seat gives someone access to software. A usage meter charges for an activity. Neither automatically tells a buyer whether a useful outcome was achieved. Founders introducing AI pricing therefore face a design problem as much as a commercial one: make the bill understandable before asking customers to trust it.
A practical test is whether a buyer can estimate the cost of a normal working week. If the answer requires guessing how often an agent will retry a task, the product may need clearer limits, previews or approval rules. Budget controls can become part of the experience rather than an administrative afterthought.
There is a trade-off for suppliers too. Charging only by seats may underprice heavy automated use. Charging for every interaction can make experimentation feel expensive. Allowances offer a possible bridge, but their usefulness depends on how well customers understand what consumes them.
For smaller founders, the lesson is to connect packaging with a recognisable job. Explain the included work, show consumption while it happens and give the customer a way to stop spending. Atlassian’s announcement is a useful case to watch because the pricing question will outlast any particular AI feature. The measure of a good model is not simply higher consumption; it is whether customers can connect that consumption to work they wanted completed.