November 22, 2026

Amazon Issues New Guidelines to Software Engineers Warning Against Unnecessary AI Usage

In a significant shift that reflects a growing sense of pragmatism within the technology sector, Amazon has officially instructed its software engineering workforce to move away from the performative use of artificial intelligence. The retail and cloud computing giant has joined a mounting list of high-profile corporations urging staff to abandon "tokenmaxxing"—a practice where employees inflate their use of AI tools to appear more productive or technologically savvy—and instead focus on utilizing the technology only when it provides genuine, measurable value.

The directive was delivered via an internal memo, which the company subsequently chose to publish, signaling a transparent change in how it evaluates the performance of its developers in the age of generative AI. The message to the company’s Software Development Engineers (SDEs) and Software Development Managers (SDMs) was unequivocal: AI is a tool for efficiency, not a metric for its own sake. In a line that has resonated across the developer community, Amazon told its staff: "If your work does not benefit from current AI tooling, that is a finding, not a failing."

This internal pivot comes as the initial wave of AI euphoria begins to collide with the economic realities of running large-scale language models. For much of the past two years, the tech industry has been characterized by a "move fast and break things" approach to AI integration. However, as the costs of compute power and token usage have skyrocketed, major enterprises are now recalibrating their strategies to prioritize return on investment over sheer adoption volume.

Not every task needs AI

The memo, distributed in September, was spearheaded by Terk Terkowitz, Vice President and Distinguished Engineer at Worldwide Amazon Stores. Terkowitz and other engineering leaders within the organization conducted a comprehensive review of the current software development landscape. Their conclusion was a sobering reminder of the fundamentals of the craft: despite the undeniable impact of AI-assisted coding, the core pillars of what makes a successful software engineer have remained largely unchanged.

According to Terkowitz, Amazon’s most effective and "prolific" engineers are not defined by how many AI prompts they generate or how many tokens they consume. Instead, their value is derived from their ability to solve complex problems, design robust architectures, and deliver products that meet customer needs.

"Our most prolific engineers have demonstrated that neither the developer role nor their output is defined by the use of AI," Terkowitz stated in the memo. He introduced the concept of the "Renaissance Developer"—a professional who successfully blends the speed of AI tools with the indispensable qualities of human judgment and deep domain expertise. This new archetype is not someone who offloads their thinking to a machine, but someone who uses technology to build "entirely new things" that were previously impossible or too time-consuming to achieve.

This shift in rhetoric suggests that Amazon is attempting to course-correct after reports emerged earlier this year regarding internal culture. Some employees had reportedly been using AI for redundant or unnecessary tasks specifically to climb internal rankings on an AI adoption leaderboard. Such gamification of technology adoption often leads to "tokenmaxxing," a term that has become a pejorative in the industry for the habit of ramping up AI usage regardless of whether the task at hand actually requires it.

The trend of tokenmaxxing is not unique to Amazon. Across the tech landscape in 2026, several major players have faced a "bill shock" resulting from unchecked AI consumption. Uber, for instance, was recently forced to implement a strict cap on AI usage after the company reportedly exhausted its entire annual AI budget in just four months. Similarly, Accenture issued orders to its staff in late July to cease using generative AI for basic, low-value tasks as costs spiraled out of control. Even Microsoft, a primary backer of OpenAI, sent a memo to its own engineers over the summer emphasizing that "tokenmaxxing is not what we are optimizing for," and urging a focus on customer and business value.

AI needs to add value

Despite the warning against unnecessary usage, Amazon is not backing away from its commitment to artificial intelligence. The memo was careful to emphasize that AI remains a foundational element of the company’s future engineering strategy. A spokesperson for Amazon clarified to CNBC that there is no hard cap on AI usage at the company, unlike the restrictive measures seen at some of its peers. The goal is not to limit the technology, but to refine its application.

"We are certain that whatever future best practices are, AI will be at the center of software engineering practice," the memo noted. It encouraged engineers who have not yet integrated AI into their daily workflows to do so as soon as possible, acknowledging that the technology is essential for staying competitive in a rapidly evolving market.

However, the company also used the memo to offer a rare admission of failure regarding its previous attempts to track AI productivity. Amazon leaders acknowledged that their initial efforts to measure the impact of AI were "deeply flawed." These efforts relied on "input metrics"—such as the volume of tokens used or the frequency of AI tool adoption—and "throughput metrics," such as the raw number of code changes or reviews.

The company realized that these metrics were fundamentally disconnected from meaningful business outcomes. By focusing on volume rather than value, the metrics inadvertently incentivized the very "tokenmaxxing" behavior the company now seeks to eliminate. "We are all aware of how deeply flawed these metrics are, primarily because they are disconnected from what matters," the memo admitted, noting that such targets can "drive negative outcomes," including unnecessary complexity and bloated operational costs.

This complexity is particularly at odds with one of Amazon’s core leadership principles: "Invent and Simplify." The company’s "Renaissance Developer" initiative is an attempt to realign AI usage with this mantra. If an engineer uses AI to generate five hundred lines of code for a problem that could be solved with ten lines of manual, elegant logic, that engineer is failing to "simplify," regardless of how much AI they utilized.

Moving forward, Amazon is directing its managers to take a more nuanced approach to oversight. Rather than pushing for 100% adoption across the board, SDMs are being told to analyze both the top and bottom deciles of their teams. By looking at the bottom 10% of AI users, managers are tasked with understanding the specific barriers to adoption or, conversely, identifying areas where AI truly offers no benefit.

Amazon is essentially asking its engineers to become critics of the technology as much as they are users of it. The memo explicitly states that an engineer who can articulate exactly why AI is not helpful for a specific project is demonstrating the high level of professional judgment the company expects.

"Deploying AI in your organizations requires carefully balancing between exploring new and better ways to do things and investing in doubling down on the ones you already have," the memo concluded.

This evolution in Amazon’s stance reflects a broader maturation of the AI industry. As the "novelty" phase of generative AI concludes, the world’s largest technology firms are beginning to demand the same level of discipline and accountability for AI projects that they require for any other capital expenditure. For Amazon’s thousands of software engineers, the message is clear: the era of using AI to look busy is over; the era of using AI to be better has begun.