Amazon has released updated guidance for software engineers in the age of artificial intelligence, urging employees to prioritize business outcomes and customer impact over simply burning tokens for the sake of statistics.
Amazon has released updated guidance for software engineers in the age of artificial intelligence, urging employees to prioritize business outcomes and customer impact over simply burning tokens for the sake of statistics.
This is stated in a memorandum by Amazon vice president and engineer Terk Terkowitz, which was obtained by Business Insider.
The document notes that artificial intelligence is changing the software development landscape faster and more fundamentally than previous technology waves, including cloud computing. The company calls this a “developer renaissance,” where human judgment is combined with machine intelligence.
"This is a time of change for our industry, and a time of change in what it means to be a software engineer. We all need to embrace these changes and take the time to learn, or we risk falling behind," Terkowitz noted.
The updated guidelines were a response to tensions within the company. Previously, Amazon tracked the use of internal AI tools, token consumption, and lines of code to create team rankings. This raised concerns among employees who feared that pressure from management was encouraging waste rather than productivity.
In the memo, Terkowitz explicitly warned against making such metrics the primary goal, as it could lead to negative consequences — including “tokenmaxing” instead of improving code quality. More token consumption increases cloud infrastructure costs, and excessive focus on the number of lines of code creates unnecessary complexity.
At the same time, the company expects AI to remain at the center of development, and department heads will use these tools in production processes themselves. Managers are advised to analyze the work of both the 10% of the most effective teams and the 10% of those who are least actively using new technologies. For engineers whose tasks cannot yet be optimized with the help of AI, they left the opportunity to argue this: explaining the limitations of the tools will be considered a manifestation of professional judgment.
