Amazon tells engineers to focus on outcomes, not token usage, slowing AI roll

The need to deliver valuable software and achieve business objectives, rather than simply maximizing AI token usage, is what Amazon has directed its engineers to prioritize, stemming from growing worries about the escalating costs and inherent complexity of artificial intelligence.

In a directive issued to engineering groups on Tuesday, Amazon publicly unveiled this strategic pivot, detailing the evolving responsibilities of its software development engineers amid the rise of AI. Terk Terkowitz, a distinguished engineer and VP at Amazon, penned the note, asserting that "Our most prolific engineers have demonstrated that neither the developer role nor their output is defined by the use of AI." He further elaborated, stating, "We're now in the era of the Renaissance Developer, blending AI tools with human judgment and domain expertise to build entirely new things."

Essentially, Amazon desires its developers to incorporate AI, but the memo clarifies that merely expending tokens does not equate to producing meaningful work. Instead of framing the developer's function by the extent of AI utilization, the communication emphasizes what remains constant: the necessity of creating functional software that serves both the business and its clientele.

The memo further posits that determining what constitutes effective AI-assisted engineering practices in a way that endures the test of time presents a significant challenge. Consequently, the company opted to concentrate on unchanging elements as the SDE roles have transformed, noting that "the outcomes that matter aren’t changing, even if the inputs, tools, and techniques are changing faster than ever."

For software developers presently, this translates into a clear message: AI input metrics, even if Amazon has observed a link between tokenmaxxing teams and greater output or delivery, are not what the company views as a proper gauge of efficient use of developer time and resources. The memo firmly contends, "We are all aware of how deeply flawed these metrics are, primarily because they are disconnected from what matters." Moreover, it cautions that "Using these metrics could, if used as a target instead of a measure, drive negative outcomes, increasing costs, and increasing complexity," referencing the issues associated with tokenmaxxing expenditures and code becoming overly elaborate, which conflicts with Amazon's mission to "invent and simplify."

An Amazon representative confirmed that the memo’s aim is to discourage the practice of tokenmaxxing, referencing comments from AWS CEO Matt Garman in June where he urged developers to measure results rather than token consumption when appropriately employing AI tools for software creation. Furthermore, Amazon expects its developers to consider how AI can enhance "delivery for customers" and business units, moving beyond just finding ways to demonstrate its utility.

The document specifically called out certain engineering teams that have "become less connected to their customers and business," contrasting them with teams that maintain stronger ties to their products and business functions, as these latter groups "are able to take the most advantage of AI and agentic software development." In essence, the message encourages more teams to adopt this more connected posture. The memo concluded by stating, "If your work does not benefit from current AI tooling, that is a finding, not a failing," adding that "An engineer who can explain where AI does not yet help their work is demonstrating the judgement we are asking for."

Amazon's viewpoint closely mirrors guidance provided by Microsoft to its internal development staff in August, particularly in its emphasis on customer and business returns over input maximization. Microsoft EVP Jay Parikh informed developers, "Tokenmaxxing is not what we are optimizing for. I want all of us focused on maximizing outcomes that move the needle for our customers and our business."

This signals a potential shift away from the habit of applying AI indiscriminately to everything. When two corporations of this magnitude begin to question whether the consumption of AI tokens serves as a valid metric for measuring the productivity of their thousands of software engineers, it is plausible that the wider industry will soon adopt a more measured approach to token use. Alternatively, we might soon observe an increase in enterprises embracing the philosophy of utilizing "good enough" open-weight models capable of handling numerous tasks at a substantially lower expense than pricier alternatives—a stance Garman advocated for in June when discussing the selection of cost-effective models based on the specific task.

Why it matters

This signals a significant shift in how large tech companies view AI productivity, moving away from input metrics like token consumption. Developers should watch for this trend influencing how engineering teams measure success and justify AI tool adoption moving forward.