Amazon Caught Its Employees Faking AI Usage to Hit Internal Targets

Welcome back to SavvyMonk, your one-stop for AI and tech news that actually matters.

Amazon employees are gaming their own company's AI usage dashboards. And it turns out Amazon is not the only one with this problem.

Let's get into it.

If you’re optimizing for growth, you need ecomm tactics that actually work. Not mushy strategies.

Go-to-Millions is the ecommerce growth newsletter from Ari Murray, packed with tactical insights, smart creative, and marketing that drives revenue.

Every issue is built for operators: clear, punchy, and grounded in what’s working, from product strategy to paid media to conversion lifts.

Subscribe for free and get your next growth unlock delivered weekly.

Sign up now

Big Tech has a measurement problem, and it starts from the top. Over the past year, companies like Amazon, Meta, and Microsoft have built internal dashboards and leaderboards to track how much their employees use AI tools. The logic was straightforward. More usage means more productivity gains, and leaders want proof those gains are happening.

Amazon set a target requiring more than 80% of its developers to use AI tools every single week. It tracked consumption on internal leaderboards and created an environment where the pressure to use AI was constant.

One employee said there was so much pressure to use these tools. Another said the tracking created "perverse incentives."

Amazon officially told workers that AI usage statistics would not factor into performance evaluations. Multiple employees said they did not believe that.

How It Works

The behavior that emerged has now been given its own name in tech circles: tokenmaxxing. The idea is simple. If you are being judged on how many AI tokens you consume, you find ways to run up the number regardless of whether the work needed doing.

At Amazon, some employees turned to MeshClaw, an in-house agent platform capable of initiating code deployments, triaging emails, and interacting with Slack. Rather than using it for genuine tasks, workers used it to maximize token consumption scores on the leaderboard.

Meta and Microsoft had similar dynamics surface last month. Meta's internal AI usage leaderboard lasted only days after it was publicly reported before the company pulled it. Amazon has since restricted visibility of team-wide usage statistics, though the underlying pressure remains.

Why It Matters

This is not just a quirky story about office politics. The numbers that executives report to investors about internal AI adoption are part of how the industry justifies spending at a scale that would have seemed absurd five years ago.

Combined 2026 capital expenditure from Amazon, Microsoft, Alphabet, and Meta is tracking between $700 billion and $725 billion, up 77% from last year's record $410 billion.

Every hyperscaler has told investors that inference capacity is being absorbed as fast as it can be deployed. Internal developer consumption sits alongside paying external customers in the data that informs capacity planning, GPU procurement, and power infrastructure commitments.

If a meaningful share of that internal consumption is performative rather than productive, the demand signal is distorted. And the industry is placing GPU orders and power contracts years in advance, priced against those signals.

The Jensen Huang Problem

Nvidia CEO Jensen Huang has made per-engineer token consumption a kind of moral benchmark.

He said at a public event that he would be "deeply alarmed" if a $500,000-a-year engineer was not consuming at least $250,000 worth of tokens annually. He compared not using AI tools to designing chips with paper and pencil.

That framing lands very differently when you know engineers are being pressured to hit token targets regardless of genuine need. Every inflated token is real GPU time. Nvidia's long-term growth depends on that GPU time translating into actual productive work that compounds and persists, not into developers running an agent platform in the background to move their leaderboard score.

What Changes Next

Angie Jones, formerly VP of engineering for AI tools at Block, has said she expects the industry to pivot toward measuring efficient token usage rather than raw volume. That framing is more honest about what actually matters, but getting there requires companies to admit that their current metrics are broken, which is a harder conversation when the stock is up.

Meta pulled its leaderboard within days of exposure. Amazon restricted visibility of team-wide data. Both moves suggest companies know the incentive structures were flawed. The harder question is whether the consumption intensity those structures created was already baked into the demand forecasts that justified hundreds of billions in infrastructure spending.

The Bottom Line

Tokenmaxxing is what happens when you measure the wrong thing and tie people's careers to the number. The practice is widespread enough now to have its own vocabulary, its own leaderboard culture, and its own paper trail. The broader concern is not that employees gamed a metric. It is that those gamed metrics may have shaped the AI infrastructure investment decisions that the entire industry is now committed to at a cost of $700 billion this year alone.

Category: Workplace Productivity Audit

"You are an operations consultant. I am a [role] at a [company size] company. We track [specific metric] to measure AI adoption across our team. Review this metric for perverse incentives and explain what behaviors it could encourage that look good on paper but do not add real value, then suggest two alternative metrics that better capture genuine productivity gains."

This story about Amazon is one of those that makes you step back and look at the whole picture. These companies have spent the last two years telling investors that AI adoption is accelerating and that demand for compute is insatiable.

Some of that is true. But when the evidence of adoption is being manufactured on internal leaderboards, it raises a fair question about how much of the underlying demand picture reflects real work and how much reflects pressure to look like a believer.

Hit reply, I read every response.

See you in the next one.

— Vivek

P.S. If you know a founder, developer, or product manager who wants straight-talk coverage of AI and tech, forward this their way. They can subscribe at https://savvymonk.beehiiv.com/