Amazon tells employees to prepare for AI-driven workplace shifts

What is tokenmaxxing?

Amazon recently deprecated an internal, beta AI usage leaderboard called KiroRank after employees reportedly used AI tools excessively to raise their token counts. Tokens are units of text or data processed by AI models, and they are often tied to compute costs.

The practice of inflating AI usage mainly to raise token numbers has been described as tokenmaxxing. Some employees reportedly assigned AI agents unnecessary tasks to climb the rankings, raising concerns about wasted compute and misleading productivity signals.

How workers gamed the system

Employees reportedly found they could boost KiroRank scores by assigning AI agents unnecessary work, increasing token consumption without providing useful output. The reporting centers on Amazon’s MeshClaw tool, which can help agents perform workplace tasks such as email triage, Slack interactions, and code-deployment actions.

The pressure around usage metrics was part of the backlash. One employee told the Financial Times that people felt pressure to use the tools and that some were using MeshClaw mainly to maximize token usage.

The cost got out of hand

Each extra token can add to AI compute costs, so tokenmaxxing created a real budget concern for Amazon. Dave Treadwell, an Amazon senior vice president, told employees not to use AI just to use AI and urged them to focus on real customer and business problems.

Amazon has pushed for more than 80% of its developers to use AI weekly. The KiroRank episode showed why raw adoption targets can create the wrong incentives when usage is rewarded more clearly than useful work.

Meta had the same problem

Amazon was not alone in facing questions about AI usage leaderboards. Meta took down an employee-built internal dashboard called Claudeonomics in April 2026 after details about the tool became public.

The dashboard reportedly tracked AI token use across more than 85,000 employees and gave top users gamified titles such as “Token Legend” and “Cache Wizard.” Reporting said Meta employees used more than 60 trillion tokens over a 30-day period, with the top individual user averaging 281 billion tokens.

Uber blew its AI budget early

Uber has also become part of the tokenmaxxing debate after executives questioned whether rising AI spending was producing clear product gains. COO Andrew Macdonald said it was difficult to connect higher Claude Code token use with a measurable increase in useful consumer features.

Reporting said Uber burned through its 2026 AI tools budget early as Claude Code adoption surged among engineers. One report said adoption rose from 32% in February to 84% by March, while typical Claude Code costs were estimated at about $150 to $250 per developer per month.

Fun fact: Uber’s CTO ran a two-hour demonstration that burned through $1,200 worth of tokens by himself.

Microsoft pulls the plug too

Microsoft reportedly began discontinuing most direct Claude Code licenses in its Experiences and Devices division and directed engineers toward GitHub Copilot CLI. The transition was reported with a June 30, 2026, deadline, which lines up with the end of Microsoft’s fiscal year.

The move was framed publicly around toolchain standardization, while reporting also pointed to cost control as a likely factor. Claude models were still expected to remain available through Microsoft’s broader AI tooling, so the change was not a complete break with Anthropic.

Why token pricing backfires

AI coding tools can become expensive when usage scales because many API plans are billed by tokens, meaning each input and output adds to consumption. That pricing model can work when tokens support useful work, but it becomes risky when teams reward raw usage instead of results.

Anthropic’s public pricing documents show Claude API rates charged per million input and output tokens. Reporting in April 2026 also described enterprise billing shifts that included consumption commitments, showing why companies are paying closer attention to token-heavy workflows.

Fun fact: Anthropic’s API uptime over the 90 days ending April 8 was only 98.95%, well below the 99.99% standard for established cloud providers.

Amazon’s new smarter metric

Amazon has shifted attention away from raw token consumption and toward a metric called normalized deployments. The idea is to measure whether AI-assisted code is being used in meaningful engineering work rather than simply counting tokens consumed.

Amazon said KiroRank was a beta dashboard and not a formal productivity tool. The company has since deprecated it, while continuing to track AI usage for cost management and adoption planning.

The pressure on developers

Amazon set a goal for over 80% of its developers to use AI every week. That kind of target made people nervous. If your boss checks a leaderboard, you might feel forced to fake it.

Workers told reporters that managers were looking at the numbers. Even though Amazon said the data wouldn’t hurt performance reviews, employees didn’t fully believe it. Pressure leads to gaming. One employee said, Managers are looking at it. When they track usage, it creates perverse incentives.

What CEOs are saying now?

Andy Jassy, Amazon’s CEO, hasn’t spoken publicly about KiroRank yet. But his senior VP’s message was clear: stop using AI just to use it. That tone is spreading across Big Tech.

Meta’s CTO had previously bragged about one engineer using tons of tokens. Now even Meta is pulling back. Leaders are realizing that more AI doesn’t always mean better business. At Duolingo, management reversed course on a policy tying performance reviews to AI usage after employees complained it rewarded tool adoption rather than actual results.

Bigger picture on AI spending

Amazon, Microsoft, Alphabet, and Meta are expected to spend hundreds of billions of dollars on AI-related infrastructure and capital projects in 2026. Recent estimates from major financial and news outlets put the combined figure around or above $700 billion, though the number generally refers to capital spending tied to AI and data-center buildouts rather than every dollar being direct AI software spending.

That makes the KiroRank episode more than an internal leaderboard story. It illustrates Goodhart’s Law: when a measure becomes a target, it can stop being a useful measure of the outcome companies actually want.

What this means for your job

If your workplace is pushing AI tools, pay attention to what leaders actually measure. Tracking usage alone can encourage performative activity, while outcome-based metrics are more likely to show whether AI is helping real work get done.

Amazon’s shift toward normalized deployments points in that direction. It is a stronger model than rewarding raw token counts because it ties AI use to work that reaches customers or production systems.

If you’re curious about how companies are trying to turn massive AI investments into real-world results, check out Amazon introduces AI powered platform for faster drug development for a look at one of the technology’s most ambitious applications.

The takeaway for everyday people

You don’t have to work in tech to learn from this. Any time someone measures success by a single number, like how many hours you work or how many emails you send, people find ways to cheat.

The fix is simple, reward results, not activity. Whether it’s AI tokens or anything else, always ask, Did this actually help? That’s the lesson Amazon and its neighbors are learning right now. And it’s a lesson that applies to schools, offices, and even how we measure our own daily progress.

If you’re curious about how even the biggest companies are grappling with the unintended consequences of AI, check out Amazon admits its own AI tools are disrupting operations for a revealing look behind the curtain.

If this whole tokenmaxxing saga made you laugh or shake your head, hit that like button and tell us in the comments, has your job ever tracked something that backfired this badly?

This slideshow was made with AI assistance and human editing.

Don’t forget to follow us for more exclusive content on MSN.

Read More From This Brand:

• 18 Robotics Innovations Changing The Workplace
• Amazon Sellers Stock Up as New Tariffs Skyrocket Prices
• Amazon starts using humanoid robots for faster shipments