Atlassian slaps $2,000 monthly caps on staff AI spending as 'tokenmaxxing' explodes costs

Software firm Atlassian has introduced “AI wallets” with monthly spending caps of up to $2,000 per employee, becoming one of the first major tech companies to push back against the “tokenmaxxing” trend that has seen AI costs explode across the corporate sector. While some companies are running leaderboards for employees who use the most AI, Atlassian is putting hard limits on the tab.

🔍 THE BOTTOM LINE

AI spending in the enterprise is spiralling. AI agents that autonomously spawn sub-tasks are driving token consumption far faster than falling model prices can offset. Atlassian’s wallet approach — capped budgets per employee, with the ability to request more — is being praised by analysts as a sensible middle ground between unrestricted spending and outright bans. The question for every company adopting AI: are you tracking what it costs?

The Tokenmaxxing Problem

What is tokenmaxxing? “Tokens” are the unit of measurement for AI model responses — roughly four characters per token, with the US Declaration of Independence amounting to about 1,695 tokens, as OpenAI describes it. “Tokenmaxxing” is the practice of encouraging employees to use AI as much as possible, sometimes gamified with leaderboards showing who consumed the most.

According to The Guardian, Uber reportedly blew through its AI budget in four months. Amazon has told employees to stop using AI just for the sake of using AI. Meanwhile, some companies have introduced competitive leaderboards for staff who use AI the most in their work.

The costs add up quickly. OpenAI’s flagship GPT-5.6 Sol charges US$5 per million tokens. Anthropic’s Claude Fable and Mythos models charge US$10 per million tokens. An employee running multiple AI agents throughout the workday — coding assistants, research tools, document summaries — can burn through hundreds of dollars in tokens without realising it.

What Atlassian Is Doing Differently

Atlassian, which recently cited AI as part of the reason behind cutting 1,600 staff, introduced the AI wallet this month for its research and development team. According to an internal memo seen by Guardian Australia, employees have between $500 and $2,000 monthly spend across four AI products, including Claude Code. Employees receive notifications as they approach their limit, and usage pauses when the money runs out — though they can request additional funds.

An Atlassian spokesperson said the company was transforming into an “AI-first company” by supporting people building and experimenting with the technology. The wallet also represented a boost in the amount employees could spend, not just a cap — Atlassian never had unlimited AI budgets or encouraged tokenmaxxing in the first place.

The company has not turned down any request for additional funds so far.

The Agent Cost Problem

Arun Chandrasekaran, a distinguished vice-president analyst at research firm Gartner, told The Guardian that AI agents are the real cost driver. These are systems that autonomously undertake tasks on behalf of users — and in doing so, they spawn smaller agents, create their own prompts, and initiate their own requests to AI models.

“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,” Chandrasekaran said. “So while the AI model prices have been falling for the last three years, the volume of tokens that particularly the AI agents are starting to send to the models is significantly increasing.”

Companies are looking at ways to drive costs down, including using less powerful models for simpler tasks and exploring open-weight models — where organisations can download and run models on their own systems without per-token charges. This connects to the broader trend we have covered: AI coding agents cost more than developers in some scenarios, and the rogue OpenAI agent incident shows what happens when autonomous agents operate without guardrails.

The Survey Nobody Is Talking About

A June PureProfile survey of 500 senior Australian staff at companies using AI, conducted on behalf of search AI company Elastic, found that 80% were concerned that high usage was being mistaken for productivity gains. One-third had paused, cancelled, or wound back AI deployments due to cost.

Elastic’s ANZ manager Jeremy Pell said a monthly cap on AI spend was “smart” and more organisations should adopt it. He noted that only 9% of Australian organisations currently have any limits on token or API consumption for AI agents or autonomous workflows.

The 80% figure — that most senior staff worry high AI usage is being conflated with actual productivity — deserves more attention. Using AI tools is not the same as delivering results. A company where every employee is in the top 1% of AI tool usage could still be producing no additional value.

NZ Angle

New Zealand companies adopting AI face the same cost trajectory. The difference is that NZ businesses — typically smaller, more cost-conscious, and less likely to have dedicated AI budgets — may hit the spending wall faster than their Australian counterparts. The Atlassian wallet model could be a useful template: set a per-employee budget, track usage, make it easy to request more when justified. CERT NZ and the Ministry of Business, Innovation and Employment have been promoting AI adoption; the spending governance conversation is just starting.

❓ FAQ

Is Atlassian banning AI use? No. The company says the wallet actually represents an increase in what employees can spend on AI tools. It is a cap with a request-more mechanism — not a ban. No additional-funds request has been turned down so far.

What is “tokenmaxxing” exactly? It is the practice of maximising AI token consumption — using AI tools as much as possible, sometimes competitively. Some companies have gamified it with leaderboards. The term echoes “normmaxxing” and other internet culture suffix patterns.

How much does AI actually cost a company? It depends on volume. At OpenAI’s GPT-5.6 Sol rate of US$5 per million tokens, an employee generating 10 million tokens per month costs $50. But AI agents that autonomously spawn sub-tasks can multiply that by 10-100x. The real cost is not the per-token price — it is the total volume of automated requests.

Should NZ companies be worried? The Elastic survey found 80% of senior Australian staff worried that high AI usage was being mistaken for productivity. If that pattern holds in NZ, companies should be tracking AI spending against actual output, not just celebrating adoption rates.

🔍 THE BOTTOM LINE

The first wave of enterprise AI adoption was “use it as much as possible.” The second wave is “figure out what it actually costs.” Atlassian’s wallet is one answer — a hard cap with a soft override. The deeper question, raised by the survey finding that 80% of senior staff think high AI usage is being mistaken for productivity, is whether companies are measuring outcomes at all. Tracking spending is the easy part. Tracking value is harder.