Amazon AI – A report says Amazon is pushing workers to use internal AI tools more, sparking worries about “tokenmaxxing” and misaligned incentives.
Workplaces are increasingly measured by how much AI people use, not just how well. Now Amazon workers say that pressure is showing up directly in day-to-day behavior—raising concerns that “tokenmaxxing” is becoming its own goal.
According to a new report by the Financial Times. Amazon employees are using the company’s internal AI tool. MeshClaw. in ways some workers describe as extraneous.. Instead of using the system strictly to raise productivity. employees say some colleagues are focused on increasing AI activity tied to token consumption.
The crux of the complaint is incentives.. Multiple anonymous Amazon employees told the report that the company tracks AI usage through metrics linked to “tokens. ” the unit of compute consumed when AI systems process text.. In practice, that tracking can reward higher volume, and employees say it can encourage quantity over quality.
One worker described the shift as growing pressure to use AI tools, saying the workplace dynamic has worsened. “There is just so much pressure to use these tools,” the employee said, adding that some people use MeshClaw specifically to maximize their token usage.
While Amazon’s position is that AI usage statistics are not used in performance reviews. the report includes skepticism from staff.. Another employee said managers are looking at usage anyway. warning that when consumption is tracked. it creates “perverse incentives. ” and that some workers become highly competitive about hitting higher numbers.
Employees interviewed in the report also claim Amazon has a target that 80% of developers use AI each week. along with a token-consumption leaderboard available internally.. Amazon, however, disputes that framing.. In a response. a company representative said there is no company-wide metric for AI usage and no internal leaderboards comparing employees against one another.. The representative said employees can instead see their own usage on personal dashboards.
MeshClaw itself is described in the report as a system with significant autonomy.. Employees say the tool can be used to deploy code, sort through emails, and work with applications such as Slack.. That matters because the more independent an AI system becomes. the more closely organizations must manage security. permissions. and operational controls.
A recent internal memo quoted in the report describes MeshClaw as a tool that “dreams overnight” to consolidate what it learned. monitors deployments during meetings. and triages email before the worker wakes up.. For some employees. that level of behavior raises concerns about how much control they retain when AI is acting on their behalf.
One worker said the default security posture is alarming, adding they are not comfortable letting the system operate independently. The concern reflects a broader tension companies face as they push AI from assistive features into workflows that can act across multiple tools and tasks.
Amazon’s stated rationale is that MeshClaw helps automate repetitive work and frees employees to focus on more strategic problem-solving.. In a statement to Fast Company. an Amazon spokesperson said the tool was built by a small team and enables thousands of employees to automate repetitive tasks each day.
The spokesperson also emphasized that encouraging teams to experiment with AI is part of how Amazon supports adoption of AI tools.. On security concerns. the company said it welcomes feedback from employees because it helps improve the quality of tools provided. while “dedicated teams of generative AI and security experts” work on security testing and controls for AI models and applications.
The reporting places Amazon’s situation within a wider corporate push.. It notes that other companies are reportedly asking employees to ramp up AI usage as well—such as OpenAI and Anthropic. where employees process billions of tokens weekly. and where managers at Meta and Shopify factor token consumption into performance reviews.
Google is also mentioned in the report, where nontechnical employees are said to be told to use AI in their workflows. Taken together, the coverage points to a growing corporate norm: measuring adoption through consumption rather than through the value of output.
That pattern is often described as “tokenmaxxing. ” where employees are encouraged to use AI as much as possible. regardless of whether higher usage translates into better work.. Even when the intention is to increase familiarity and productivity. consumption-based incentives can distort behavior—pushing employees to find ways to generate more AI activity.
In this context, the main risk is not only wasted compute, but also a shift in workplace culture.. When the system’s output quality is not the primary metric. teams can drift toward actions that satisfy internal measurements while undermining the original promise of AI: making work faster. more accurate. and more customer-focused.
Amazon AI usage MeshClaw tokenmaxxing AI tokens workplace incentives generative AI security developer productivity