Atlassian has rolled out usage-based limits on its Rovo AI features after watching its monthly AI expenditure balloon from roughly $5 million in August 2025 to over $15 million by May 2026.

That’s a 3x increase in less than a year.

The tokenmaxxing problem

Atlassian isn’t an isolated case. The company is part of a broader reckoning sweeping through the tech sector around a phenomenon dubbed “tokenmaxxing.” Some companies reportedly built internal leaderboards ranking employees by how much AI they consumed in their daily work.

The logic was straightforward enough: more AI usage equals more productivity. The reality was considerably less flattering. Companies discovered that employees were maximizing token consumption without a corresponding bump in actual output. Amazon scrapped its KiroRank leaderboard after catching employees gaming the system. Adobe ended unlimited Claude access on June 30, 2026. Citi went so far as to disable premium AI models entirely on June 24, only re-enabling them a week later on July 1.

Meta scaled back AI access too, citing inflated costs and underwhelming productivity returns.

How the caps work

Atlassian’s new system measures usage through Rovo credits, previously called AI credits, across its Jira and Confluence platforms. The tiered monthly allowances break down like this: Standard plan users get 25 credits per month. Enterprise users receive 150 credits. Higher tiers, like the Teamwork Collection, offer between 250 and 700 credits per user per month.

Scott Wu, CEO of AI company Cognition, has been pushing the industry toward a different framework entirely. His argument: stop measuring how many tokens your employees burn through and start measuring what they actually produce.

Why crypto should pay attention

The tokenmaxxing saga carries direct implications for the crypto and Web3 ecosystem, even though Atlassian itself isn’t a blockchain company.

If major enterprises are capping AI usage rather than expanding it, the total addressable market for AI compute tokens may not grow as aggressively as bulls have projected.

The tokenmaxxing backlash is a useful case study in incentive design. Gamifying consumption without aligning incentives to productive outcomes is essentially the same design flaw that plagued many DeFi yield farming strategies in 2020 and 2021.

The shift toward accountability metrics in AI spending could benefit decentralized compute networks that offer transparent, verifiable usage data. If enterprises demand better visibility into what they’re actually getting for their AI dollars, on-chain compute providers that can prove work completion have a potential competitive edge over opaque centralized alternatives.