Amazon says using more AI does not make engineers more productive, warns against tokenmaxxing

Amazon is asking its software engineers to look beyond artificial intelligence (AI) usage figures and focus on the results they deliver. As companies encourage employees to use AI coding tools more often, tokenmaxxing is drawing scrutiny. The term refers to the practice of maximising AI token consumption, sometimes to climb internal usage leaderboards or show greater AI adoption, without necessarily improving productivity.

Tokens are small units of text and data that AI models process when responding to prompts or generating code. Higher consumption can also mean higher costs. Amazon’s latest guidance highlights why using AI more often should not automatically be treated as evidence of better productivity.

MUST READ | OpenAI launches GPT-6.1 Terra with advanced computer-use capabilities at half the API price

Amazon says AI usage metrics can be misleading

According to CNBC, Amazon shared updated AI usage guidelines with software engineers in an internal memo in September and published them on its website on October 27. The company cautioned that token consumption, AI adoption rates and code changes can provide an incomplete picture of productivity, while potentially increasing costs and adding unnecessary complexity.

Amazon had already shut down an internal leaderboard in May after finding that employees were gaming token consumption to climb the rankings. However, the company has clarified that it has not imposed caps on how many tokens employees can use.

DON'T MISS | Accenture’s $29 Claude cap raises a question for Indian IT: Who will pay for rising AI costs at work?

Instead, Amazon wants to assess whether AI helps engineers deliver meaningful results. Marc Brooker, a vice president and distinguished engineer at Amazon Web Services, told CNBC that the company looks at “feature velocity”, including product launches and successful changes made in production.

The company’s memo also makes its position clear, “if your work does not benefit from current AI tooling, that is a finding, not a failing.” Engineers are not expected to use AI simply to increase their usage numbers.

Rising AI costs put pressure on companies

The debate comes as businesses seek clearer returns on their AI investments. An IBM report cited by Mint found that tokenmaxxing’s popularity began declining after peaking in spring 2026, amid high costs and limited visible improvements.

Uber faced a similar challenge in May, when it said its entire AI budget for 2026 had been spent by the end of April. Its chief operating officer said the company struggled to clearly connect higher token consumption to better customer-facing features and improved developer productivity.

For Amazon, the focus is on whether AI helps engineers solve meaningful problems and deliver better work, rather than how much of the technology they use.