U.S. e-commerce giant Amazon issued a memorandum on Tuesday, revising its guidelines for software engineers. While encouraging employees to use AI, it also emphasized that such tools should be used with caution.
The memo states that Amazon has entered the era of "Renaissance developers" – those developers who are able to combine AI with their own professional knowledge and judgment, thereby "creating something entirely new." This also means that when AI is not helpful for work, employees should proactively inform management and explain why.
"If your work cannot benefit from the existing AI tools, this is considered a 'discovery', not a 'dereliction of duty'," wrote Terk Terkowitz, Vice President of Amazon's Global Retail Business (WorldWide Amazon Stores) and Distinguished Engineer. "Please document the reasons, share them with your colleagues, provide feedback to your supervisor, and work together to find solutions. If an engineer can articulate in what aspects AI is not yet able to assist with their work, that is precisely the kind of judgment we expect."
The core argument of this memorandum is that using artificial intelligence without discernment may ultimately backfire and harm the company's interests. Terkowitz points out that measuring productivity by counting the number of Token or lines of code has "serious flaws," as these indicators may not necessarily reflect the quality of work. He writes that setting these indicators as "goals" could lead to increased costs—pursuing the maximization of Token goes against the principle of "savings"—and add unnecessary complexity, which contradicts Amazon's philosophy of "innovation and simplification" ( invent and simplify ).
“They are out of touch with what is truly important,” he wrote.
Amazon once reaped the consequences of blindly encouraging the use of artificial intelligence. It was reported that some employees, in order to increase their own usage of token and thereby improve their ranking on an unofficial internal AI leaderboard called “Kirorank”, had the AI intelligent agents perform meaningless tasks. Due to concerns about unnecessary computational costs—which can often accumulate rapidly—Amazon discontinued this leaderboard in May. Additionally, it was reported that a company spent $500 million unexpectedly on the Claude tool within a month because it did not impose any restrictions on employee usage.
Nevertheless, Amazon has far from stopped using artificial intelligence, and has not even completely abandoned its AI adoption metrics, as Terkowitz writes, stating that these metrics are still "closely related to important outcomes." Terkowitz encourages managers to compare the 10% of teams that use artificial intelligence with the 90% that do not, to understand the reasons for the lower usage among the former, and to determine whether the output of the latter has increased accordingly.
"When considering one's own career development and the performance of the organization one is part of, it is appropriate to refer to these input indicators prudently and wisely," he wrote. "With joint efforts, over time, we will be able to find more ideal indicators and set more scientific goals."
