Amazon finally admitted it.
All this AI, all the time? Maybe not the smartest play.
They pulled the plug on a beta test site last week. Called it Kirorank. An AI leaderboard, basically. It had only been around for a few weeks before getting the axe. Employees used it to brag about how much they were using AI. Or rather, how many tokens they were burning.
Money talks, though. Or costs at least.
Financial Times reports Amazon had two massive headaches. One is obvious. Token prices are skyrocketing as companies shovel data at models like there’s no tomorrow. The other is more insidious. Tokenmaxxing.
You know what it sounds like it is.
Employees made the AI do tiny, menial tasks. Why? To rack up token usage. To climb the Kirorank ladder. It wasn’t about work. It was about the game. Amazon was pouring cash into digital voids that generated nothing but a high score on a spreadsheet.
“One of the internal dashboards… was never intended to promote the use AI for usage’s sake.”
That was Amazon’s line to CNET.
A spokesperson tried to walk it back. Called the dashboard “beta,” “unofficial,” and now “deprecated.” They want us to focus on operational efficiency. Innovation. As if throwing AI at every blinking cursor was a strategic move from the start.
Spoiler: It wasn’t.
Dave Treadwell, Amazon’s SVP, had already sent out a leaked memo back in May. He asked folks to stop using AI “just for the sake of it.” The message was clear. Measuring token count was meant for cost analysis. Not for judging developer worthiness. But gamification? It has a funny way of twisting metrics.
Amazon isn’t alone in the cleanup crew.
Meta shut down their own employee AI leaderboard in April. People were grinding for “Token Legend” status. Competing like it was the World Cup. Meanwhile, Uber is staring at a financial black hole. COO Andrew Macdonald told the Rapid Response team they could barely justify new AI costs. Their CTO admitted in a viral clip that Uber blew its entire 2026 budget in three months. Just one quarter. Gone.
Is it a trend?
Probably. Microsoft canceled Claude Code licenses in May. The Wall Street Journal says Salesforce and DoorDash are doing the same—going from “throw it at everything” to rationing carefully. Returns were lackluster. Bills were not.
“As companies get better at sorting applications… demand will only increase.”
Still. Generative AI usage is at an all-time high.
Google said Gemini jumped from 480 trillion a month last year to 3.2 quintillion by May 2026 [Note: Text said 2025 then 2026, keeping source facts]. Why? Agentic AI. Coding assistants. Tools that run constantly. They eat tokens like dry cereal. Chatbots are thirsty. Agents are drinking the whole lake.
Jackie Rees Ulmer, a business dean at Ohio University, thinks the slowdown is real but minor.
It won’t burst the bubble. Not yet. Companies are learning to filter the noise. Separating value from vanity metrics. It’s a messy process. A costly lesson in maturity.
Will McGough of Prime Capital says they’re still figuring it out.
Nobody has the map. We’re driving blind through an explosion of options. Ulmer advises students to keep learning. But double down on the human stuff. Critical thinking. Communication. The things AI can’t quite replicate without a massive electricity bill.
Maybe we should have started there.
