Amazon has issued new guidance urging software engineers to use AI tools selectively, warning that measuring productivity by token counts or lines of code could encourage wasteful and unnecessarily complicated work.
In a memo published on Tuesday, the company said it had entered “the era of the Renaissance Developer” – a worker who combines artificial intelligence with professional expertise and judgement to create new products and solutions.
The guidance, written by Terk Terkowitz, vice-president and distinguished engineer at Worldwide Amazon Stores, also tells employees to report when AI is not useful to their work rather than using it for its own sake.
“If your work does not benefit from current AI tooling, that is a finding, not a failing,” Mr Terkowitz wrote. “Document why, share it with your community of peers, bring it to your manager, and work to find solutions.”
He said an engineer who could explain where AI did not yet help was demonstrating the judgement Amazon wanted from its staff.
Amazon warns against ‘tokenmaxxing’
Mr Terkowitz described counting tokens – the units used to process AI-generated material – and measuring lines of code as “deeply flawed” methods of assessing productivity. He said neither necessarily reflected the quality of the work being done.
Turning those measures into targets could increase costs and lead to unnecessary complexity, he warned. “Tokenmaxxing is at odds with frugality,” he wrote, referring to the practice of maximising AI usage to improve performance figures.
Amazon has already faced concerns about employees using AI agents on meaningless tasks to inflate their token counts and rise up an unofficial internal leaderboard known as Kirorank. The company scrapped the leaderboard in May amid concerns over the resulting computing costs.
It was also reported that an unnamed company accidentally spent 500 million dollars on Claude in one month after failing to impose limits on employee usage.
Amazon’s approach does not amount to a retreat from AI adoption. Mr Terkowitz said the company’s existing usage measures still “correlate well with the outputs that matter”.
He encouraged managers to compare the 10th and 90th percentiles of AI users in their teams, examining why some employees use the technology least and whether those using it most are producing more as a result.
“Using these input metrics, with care and judgement, as you think about your career and your organization’s performance is appropriate,” he wrote. “Working together, over time, we will find better metrics and set smarter goals.”
The guidance follows similar concerns elsewhere in the technology industry. Shopify chief executive Toby Lütke has said some employees are producing “slop grenades” – AI-generated emails and code that leave colleagues with more work to review.
