TL;DR

Atlassian has introduced “AI wallets” for staff in its research and development team, giving each employee between $500 and $2,000 of monthly AI spend across four products including Claude Code. According to an internal memo seen by Guardian Australia, employees get notifications as they near the limit and usage pauses when the money runs out. They can request more, and the company has reportedly not refused any request so far.

The move runs against a tech-sector trend dubbed “tokenmaxxing”, where firms push staff to use as much AI as possible — some reportedly running leaderboards for the heaviest users. The costs mount quickly: OpenAI charges $5 per million tokens for GPT-5.6 Sol, Anthropic $10 per million for Claude Fable and Mythos. Uber reportedly exhausted its AI budget in four months, and Amazon has told employees to stop using AI for its own sake.

Atlassian, which cited AI among the reasons for cutting 1,600 staff, frames the wallets as an increase rather than a restriction. A spokesperson said the company is becoming an “AI-first company” and that “Atlassian provides a significant budget for our builders to leverage multiple AI tools”, with budgets set by role.

Gartner’s Arun Chandrasekaran said the driver is agents. “You suddenly have these systems that are all trying to do independent tasks that are spawning smaller agents, that are creating their own prompts and initiating requests for the model.” Model prices have fallen for three years, he noted, while agent-generated token volume has climbed sharply.

Looking Forward

The supporting survey data is Australian — a PureProfile poll of 500 senior staff found 80% worried that heavy usage was being mistaken for productivity, and 32% had paused, cancelled or scaled back AI deployments on cost — so the specific percentages do not transfer to Britain. The mechanism does. Any UK firm running agentic workflows faces the same structure: consumption that scales with autonomy rather than headcount, on a per-token bill nobody sees until it arrives. That is a governance problem before it is a budget one, and it is more acute for smaller firms, where a single misconfigured agent loop is a material cost rather than a rounding error. Only 9% of Australian organisations have any consumption limit on agents, per Elastic’s Jeremy Pell; there is little reason to think UK adoption is further ahead. Metering is the unglamorous control that makes the rest of an AI policy enforceable.