New Delhi: Amazon has issued updated guidelines for its software engineers on how to use artificial intelligence (AI) at work, urging employees to focus on customer outcomes, software quality and measurable productivity rather than simply increasing their AI usage.
The updated guidance, outlined in a memo published by Amazon on October 27 reflects the company’s evolving approach to AI-assisted software development. Amazon wants engineers to incorporate AI into their daily work, but it has cautioned against using the technology indiscriminately or treating higher usage as proof of better performance.
The memo was written by Terk Terkowitz, vice-president and distinguished engineer at Worldwide Amazon Stores. It describes an emerging model of the software engineer as a “Renaissance Developer” — someone who combines AI tools with human judgement, technical expertise and a strong understanding of the business.
Amazon wants engineers to focus on outcomes
Amazon’s updated guidance acknowledges that AI is changing software engineering more rapidly and fundamentally than many earlier technology shifts. The company expects AI to become central to future engineering practices and encourages employees to learn how to use the technology effectively.
However, the company says the fundamental purpose of software engineering remains unchanged: delivering useful, high-quality products and services for customers.
The memo argues that a developer’s role and output should not be defined solely by the tools they use. Instead, engineers should understand how AI can help them solve problems, improve delivery and develop products that meet actual customer needs.
Amazon is encouraging employees who have not yet incorporated AI into their workflows to begin learning and experimenting. Those already using AI are being urged to continue improving their methods as the technology evolves.
The guidance also recognises that effective AI practices are still developing. Rather than imposing rigid rules around specific tools, Amazon is emphasising principles that it believes will remain relevant as the technology changes.
What is tokenmaxxing, and why is Amazon concerned?
One of the central issues addressed in the memo is “tokenmaxxing”, a term used to describe excessive AI usage intended to increase consumption figures or make an employee appear more productive.
AI models process information using units called tokens. Depending on the tool and pricing arrangement, higher token consumption can increase computing costs. Using AI extensively does not necessarily mean that the resulting work is more valuable or that a task has been completed more effectively.
Amazon has warned that metrics such as token consumption, AI-tool adoption rates and the volume of code changes or reviews can become misleading when organisations treat them as performance targets rather than indicators.
According to the company, these measures can encourage employees to maximise activity without delivering corresponding improvements in software quality, customer experience or business results. Such behaviour can also increase costs and introduce unnecessary complexity into software projects.
The warning follows an earlier internal experiment involving an AI-usage leaderboard. Amazon reportedly discontinued the leaderboard in May 2026 after concerns that some employees were gaming the system to improve their rankings.
The episode illustrates the difficulty of measuring AI productivity through simple usage statistics. A developer who consumes more tokens or produces more AI-assisted code is not automatically delivering better results.
Productivity will be measured by real delivery
Amazon wants engineering leaders to place greater emphasis on long-term outcomes, including successful features delivered to users and improvements in development productivity.
The memo acknowledges that long-term outcomes can take time to measure and may be difficult to separate from other factors. Input measures such as AI usage and adoption can still offer useful information, but they should be interpreted carefully rather than treated as definitive evidence of performance.
Engineering leaders have been advised to examine both ends of the AI-adoption spectrum.
Teams with relatively low adoption should be assessed to understand whether they face barriers to using AI effectively. Meanwhile, teams with very high usage should be examined to determine whether that activity translates into faster delivery and meaningful improvements.
Amazon’s approach is intended to distinguish productive adoption from activity that generates additional costs without providing equivalent benefits.
The company is therefore not rejecting AI metrics altogether. Instead, it wants managers to connect those indicators with evidence of improved output, stronger delivery and customer value.
Managers are expected to use AI themselves
The new guidance also places responsibility on technical leaders and engineering managers to understand AI through practical experience.
Amazon expects leaders to use AI tools in their own production work and be able to explain where the technology helped, where it did not and why. This hands-on understanding is intended to help managers make better decisions about adopting AI across their teams.
The memo also encourages leaders to make time for themselves and their employees to learn, experiment and identify more effective ways of working.
Amazon believes that teams benefit when managers understand both the capabilities and limitations of AI rather than relying exclusively on adoption figures or reports from others.
The company also wants engineers and managers to maintain a closer connection with customers and the products they build. Understanding the underlying business problem can help developers determine where AI is useful and where conventional engineering approaches remain more appropriate.
Amazon says AI is not necessary for every task
Despite encouraging wider AI adoption, Amazon’s guidance does not suggest that every engineering task must involve AI.
The memo recognises that current AI tools may not improve certain types of work. Employees are encouraged to document cases where AI fails to provide meaningful assistance, share their findings with colleagues and discuss potential solutions with their managers.
This approach places importance on critical thinking and engineering judgement. Developers are expected to evaluate whether a tool actually improves the work rather than use it simply to demonstrate adoption.
It also reflects the distinction Amazon is trying to establish between learning AI and using it without a clear purpose. Employees should continue exploring new capabilities, but they must assess the results against the requirements of their work.
For engineering teams, this could mean using AI for suitable coding, debugging or development tasks while relying on human expertise to verify outputs, assess risks and make decisions that require deeper contextual understanding.
What the new guidelines mean for software engineers
Amazon’s updated rulebook signals a shift towards a more outcome-focused approach to AI adoption. Employees are encouraged to develop AI skills, but the company is also making clear that usage levels alone should not determine how their work is evaluated.
For software engineers, the guidance highlights the importance of combining technical knowledge with AI-assisted workflows, independent judgement and an understanding of customer requirements.
For managers, it underscores the need to evaluate whether AI adoption produces measurable improvements instead of assuming that more usage automatically results in higher productivity.
The broader message is that AI is expected to play an increasingly important role in software development, but effective adoption requires more than access to tools. It depends on using them appropriately, evaluating their output and ensuring that the final work delivers meaningful value.
Amazon’s approach seeks to balance the urgency of learning AI with the need to control costs, maintain quality and preserve sound engineering judgement as the technology continues to evolve.