Box CEO Aaron Levie wouldn’t be surprised if his company’s engineers set up a tokenmaxxing leaderboard.
“To be honest, they might exist,” he told Business Insider. “We don’t yet know if it’s widespread throughout the company.”
As such, Mr. Levy would not object. He said this trend of engineers competing to spend as many AI tokens as possible is a “fun and novel thing” that moves in the right direction. This will push AI agents to their limits and provide the greatest productivity gains, he said.
Token rankings are an emerging topic in the Big Tech industry. One Meta employee created a “Claudeonomics” leaderboard with titles such as “Token Legend” before it was shut down, The Information reported. According to the New York Times, OpenAI also has a token leaderboard. The token is a measure of computing that determines how the use of AI is priced. Large language models split words into numeric inputs, treating each token as approximately 3/4 of a word.
At cloud storage company Box, Levy tracks token spending, including at the employee level, but “we don’t celebrate token maximization in the same way,” he said. “However, we are focused on increasing the speed of product development and expanding the scope of our product roadmap as our primary goals.”
He has another way to identify which engineers are AI power users. Box has a Slack channel where people share best practices for AI coding, and Levie said this already generally correlates to who uses agents the most.
CEOs want to use agents and use them frequently. (Box sells enterprise AI agents.) Levie said engineers will soon not be the only ones hiring agents. This technology will impact fields such as marketing, finance, and law.
“We need to start thinking seriously about where we can leverage this new form of rich intelligence,” he says. “Tokenmaxxing is an extreme method.”
Allocating these agents (and their associated tokens) is another matter. According to Levie, 90% of the economy cannot do tokenmaxx like Meta or VC-led startups.
“This is a new frontier of what companies need to think about,” he said.
Levie had heard of several different strategies for token allocation. He described an anonymous company that was making a “Shark Tank”-style pitch about budget calculations. (Levie declined to name the company because it was a Box customer.) Employees need to seek funding, test whether it works, and then scale it up, he said.
Other examples Levie has heard of are giving more powerful models to the most productive areas, such as Opus 4.6 and GPT 5.4. “Through the organization, more efficient and cheaper models are available from the stack,” he said.
Companies are “heat-seeking missiles” for increasing productivity, he added. They find the areas that need the tokens the most.
But no, Levy does not intend to create its own corporate leaderboards and offer prizes.
“There will be a lot of interesting results based on that idea,” he said. “There are going to be people spending their budgets on things that are completely ridiculous.”