Amazon urges developers to measure AI by results, not usage
Amazon has told its software engineers to focus on useful software and customer and business results, rather than maximising AI token use. The shift reflects concerns that token-based productivity measures can encourage higher costs and overly complex code, while failing to show whether AI is helping deliver meaningful outcomes.
A memo from Amazon vice-president Terk Terkowitz describes a “Renaissance Developer” who combines AI tools with human judgement and expertise. Amazon says teams should assess how AI improves delivery, and that recognising where current tools do not help is a valid finding. The guidance echoes comments from AWS chief executive Matt Garman and a similar message from Microsoft executive Jay Parikh.
In short
- Amazon says useful outcomes matter more than AI token use.
- Token metrics can raise costs and encourage complex code.
- Engineers should use judgement about where AI helps.
Steelman: both sides, in good faith
The strongest fair case each way — we don't pick a winner.
The case for
Measuring AI by useful software and customer or business outcomes gives teams a clearer picture of whether the tools are helping. Token counts can reward activity that adds cost or complexity without improving the product, while human judgement helps developers choose when AI is useful and when it is not.
The case against
Usage measures such as token counts can be easy to track and may offer a practical early signal of whether teams are adopting AI, especially before its effects on outcomes are clear. Focusing mainly on results could also understate AI’s contribution, since delivery depends on many factors and benefits may take time to appear; combining usage data with outcome measures could help teams assess adoption and impact.
Explained simply
Amazon is one of the world’s largest technology companies, and its cloud computing arm, Amazon Web Services (AWS), supplies computing services to businesses and public bodies. Its software teams build the systems behind Amazon’s shops and services, as well as tools sold to other organisations.
AI tools can help programmers write and review code, but they often charge according to how much text they process, measured in “tokens”. Counting token use or AI-generated code can show activity without showing whether software is better, cheaper or delivered more quickly.
The discussion is part of a wider question facing technology companies: how to tell whether AI is improving work in practice. Amazon’s guidance puts the emphasis on useful results and on developers using their own judgement to decide when AI helps.
AI Americas Business Companies Football Software Sport Technology World
Read the full article at the source →
Originally published by The Register as “Amazon tells tokenmaxxing devs to slow their AI roll”.