Amazon is warning software engineers against 'tokenmaxxing', the practice of maximising artificial intelligence (AI) usage without necessarily improving results. The company says treating AI consumption as a performance target can drive up costs and add unnecessary complexity without guaranteeing better work.
In updated guidance for software engineers, Amazon says employees should focus on customer and business impact rather than simply how much AI they use. The company is not backing away from AI. Instead, it wants engineers to use the technology with judgement and assess whether it delivers meaningful results.
Amazon Challenges AI Usage Metrics
Amazon acknowledges that metrics such as AI token consumption, adoption rates and the volume of code changes can help track how widely the technology is being used. However, it warns against treating these figures as standalone measures of performance.
The guideline describes such metrics as 'deeply flawed' because they are 'disconnected from what matters'. Amazon warns that turning them into targets, rather than using them as indicators, can increase spending and encourage unnecessary complexity.
'If your work does not benefit from current AI tooling, that is a finding, not a failing,' the guideline states. It adds that engineers who can explain where AI falls short are demonstrating the judgement the company expects.
However, Amazon does not dismiss usage metrics altogether. Its guidance says they can correlate with meaningful outputs when interpreted with care and judgement. Managers are encouraged to examine whether teams with high AI adoption are delivering corresponding improvements, rather than assuming that more usage automatically means greater productivity.
Amazon Previously Shut Down an AI Leaderboard
The warning follows an earlier episode in which an internal AI leaderboard reportedly encouraged some employees to chase usage figures. In May, Amazon discontinued KiroRank, an employee-created leaderboard that tracked AI token consumption.
According to Business Insider, some employees used AI agents on tasks that did not meaningfully solve problems in an effort to improve their rankings.
Amazon Senior Vice President Dave Treadwell told staff: 'Please don't use AI just for the sake of using AI.'
An Amazon spokesperson said the leaderboard was 'never intended to promote the use of AI for usage's sake'. The episode illustrates how incentives can encourage counterproductive behaviour when employees focus on improving usage rankings rather than solving meaningful problems.
Uber's AI Spending Highlights the Cost Challenge
Amazon is not alone in confronting the financial implications of expanding AI use. Uber reportedly exhausted its annual budget for Anthropic's Claude Code by April, just four months into 2026, after the coding tool spread across its engineering workforce faster than anticipated.
As Forbes reported, the spending highlighted the difficulty of forecasting costs as engineers increasingly rely on AI coding agents.
The budget overrun does not establish that Uber's AI spending failed to improve productivity. Rather, it illustrates how quickly expenditure can exceed forecasts when adoption grows faster than anticipated.
The Challenge Is Measuring AI's Value
The term 'tokenmaxxing' draws on the internet slang 'maxxing', which refers to pursuing an activity to the fullest or optimising an aspect of life. Dictionary.com named 'maxxing' its 2026 Word of the Year.
For businesses adopting AI, widespread use is not, by itself, proof of value. The challenge is determining whether the technology helps employees deliver better work, solve meaningful problems and improve outcomes for customers — rather than simply consume more tokens.
