Atlassian caps engineer AI spending to control rising token costs
The software company is limiting monthly AI allowances for its remaining engineers, highlighting a broader industry shift from unrestricted experimentation to strict cost management as generative computing expenses surge.
Atlassian has introduced fixed monthly allowances for its research and development staff to pay for artificial intelligence tools. The software maker will issue AI wallets ranging from $500 to $2,000 depending on the employee's role, which will warn users as they approach their limit and halt access when the funds are depleted.
This rationing arrives after the company cut 1,600 jobs earlier this year to rebrand itself as an AI-first business. Executives had previously told support staff over video that they would be largely replaced by the technology, making the sudden need to meter its use for surviving employees a difficult message to reconcile.
An Atlassian spokesperson defended the policy as a way to fund experimentation without letting expenses spiral. “Atlassian provides a significant budget for our builders to leverage multiple AI tools,” the representative stated, framing the wallets as generous guardrails rather than a restriction.
The move reflects a harsh new arithmetic for technology investors and corporate boards. A single text token equals roughly four characters, and leading models charge several dollars per million tokens. When autonomous coding agents iterate independently, they can consume millions of tokens and generate massive, unpredictable bills.
Atlassian is joining a growing list of major technology firms reining in what the industry dubbed tokenmaxxing. Amazon recently disabled an internal leaderboard that encouraged heavy AI use after staff consumed tokens just to game the system.
Meta has taken similar steps, warning 6,000 employees that internal AI spending could reach billions in 2026 and subsequently rolling out strict token controls. The broader corporate focus is shifting from measuring AI adoption by raw usage volume to evaluating the actual cost per business outcome.
For European markets and enterprise software buyers, this signals that the era of cheap, unlimited generative AI is ending. Companies must now treat AI compute as a heavily managed line item, balancing the productivity gains of agentic tools against rapidly escalating operational costs.
While some competitors still offer unlimited AI access as a recruiting perk, the industry consensus is fracturing. The central debate is no longer whether these tools are useful, but whether their output justifies an open-ended financial commitment.
