Tokenmaxxing Explains Why Your CEO Thinks AI Is Working and Your Team Doesn’t

Amazon recently shut down an internal AI leaderboard, nicknamed KiroRank, after employees started feeding trivial, made-up tasks into the company’s AI coding tools just to climb the rankings. As Fast Company reported, one worker admitted the pressure was so intense that people were maximizing their token usage for the sake of the score, not the work. It’s a small story with a big lesson: measuring AI success by tokens burned tells a company almost nothing about whether the work actually got better.

That gap between what executives track and what teams actually experience is exactly what Intch founder Yakov Filippenko wrote about in a letter to the Financial Times, responding to the paper’s report on young professionals souring on AI. His argument: leadership isn’t measuring the wrong AI, it’s measuring the wrong thing entirely.

“Remote work remains the truer predictor of declining entry-level hiring, not AI,” Filippenko wrote. “By outsourcing the messy curiosity of the Wikipedia rabbit hole — looking for one thing and researching three more — companies are quietly killing the critical thinking that develops through discovery, manufacturing a fragile new class of hire: fast on day one, but completely marooned by Day 30 because their brains were never permitted to train.”

The real predictor isn’t AI, it’s remote work

Filippenko’s letter pushes back on the idea that AI is what’s keeping junior hires out of the workforce. Companies pulled back on training junior talent well before generative AI became a boardroom talking point, largely because remote work dismantled the desk-side shadowing that used to build professional intuition. AI became a convenient explanation for a problem that started with how, and where, people work, not which tools they use.

Why tokenmaxxing backfires

Tokenmaxxing is what happens when a company rewards AI usage instead of AI outcomes. Amazon’s leaderboard was meant to encourage adoption, but it ended up rewarding volume over judgment, and employees optimized for the metric instead of the mission. The company confirmed to the Financial Times that token statistics would not factor into performance reviews, but by then the damage to trust was already done. Filippenko points to the same pattern in hiring: a headline number, tracked for its own sake, quietly replaces the harder work of actually developing people.

What should companies measure instead?

Filippenko’s proposal is to introduce large language model proficiency audits, modeled on the computer literacy courses companies rolled out in the 1990s. Instead of counting how many tokens an employee burns, audits would test whether someone can actually use AI tools to solve a real problem, delegate the routine parts, and still know when to double-check the output. It is a slower metric to build than a leaderboard, but it measures the thing that actually matters: whether people are getting better at the work, not just faster at logging activity.

The management gap behind the AI panic

Filippenko draws a sharp parallel: return-to-office mandates protect shareholders’ commercial real estate portfolios, and the rush toward automation protects the people selling the tokens. Both let leaders sidestep a harder problem, which is that most managers still haven’t learned how to delegate to a generation that won’t tolerate being micromanaged. We’ve written before about what it actually takes to manage Gen Z at work, and the same lesson applies here: the tools change every year, but the management gap doesn’t close on its own.

Whether a business is tracking token counts or headcount, the underlying question is the same: are you building people who can think, or hires who can only follow prompts? For companies building lean teams through part-time and fractional hiring, Intch connects businesses with experienced professionals who don’t need a leaderboard to prove they know how to do the work.