Legora CTO calls tokenmaxxing “really stupid,” says demo days beat dashboards

Legora's CTO Jacob Lauritzen did not mince words: he called tokenmaxxing “a really stupid way to do anything,” describing it as people burning AI tokens just to perform well on internal usage dashboards. Lauritzen, speaking on the “20VC” podcast episode released on Saturday, argued there are better ways to gauge actual AI adoption and impact than tracking raw token consumption.

In his view, the fix is straightforward. Instead of rewarding token volume, reward building and results: he pointed to “hack days or demos where employees can show others what they're building and the efficiency gains they have achieved.” That difference matters because tokenmaxxing can turn AI from a tool into a scoreboard. When the scoreboard is token usage, the incentive becomes throughput at any cost, not effectiveness.

So what is tokenmaxxing, in practical terms? Lauritzen tied it to the common playbook where companies hand out lots of access to AI tools like Claude, Codex, and Cursor, then monitor how many tokens employees use. The intention is usually to measure whether people are actually using AI. But Lauritzen said the outcome looks like tokenmaxing: “people just burn tokens just to look good.” In other words, dashboards meant to drive adoption can backfire by making usage itself the goal.

Legora's CTO also framed the business math behind the problem. He said fast-growing companies like Legora have “a really high opportunity cost” if they spend a ton of tokens to learn something that only yields modest efficiency improvements. His example: “Is it worth us spending a ton of tokens to learn if it maybe gives us 20% efficiency for us? Yes, we have a really high opportunity cost.” The point is not that AI value is imaginary. It is that token burn can be a costly proxy for value, especially when budgets are finite and compute costs scale with activity.

That timing is important because this debate is moving from theory to policy. The source ties Lauritzen's comments to a broader industry pivot from tokenmaxxing toward token capping. The idea of token capping is to limit monthly spending per AI tool, so teams cannot run up costs just by playing the system. This is not happening in a vacuum. Finance departments increasingly worry about the bill attached to experimentation and internal enthusiasm for generative AI.

Several real-world moves underscore how seriously companies are taking the cost problem. Last week, Uber limited all employees to $1,500 in monthly token spend per AI tool after the ride-hailing company blew through its AI spend budget earlier this year. Last month, the Financial Times reported that Amazon shuttered an internal dashboard that tracked AI use after some staff performed tasks to climb the leaderboard. An Amazon spokesperson told Business Insider that the unofficial dashboard “was never intended to promote the use of AI for usage's sake.” In short: dashboards can create incentives that contradict what leaders thought they were measuring.

Even the “how to think about tokens” conversation is shifting toward more disciplined procurement and model selection. At a Bloomberg conference last week, Andrew Feldman, CEO of Cerebras Systems, said the idea of unlimited tokens was “boneheaded from the get-go.” He argued you do not need a “Ferrari” to go to the grocery store, and pointed to using lower-cost open source models. His analogy landed on shopping behavior, saying, “What we're learning is how to shop at Costco.” Whether you agree with his framing or not, it reflects a growing executive consensus: unlimited experimentation does not stay “just experiments” for long.

For executives, Lauritzen's core claim is a leadership problem as much as a product one. If you measure what you can count, teams will optimize for it. Token usage is countable. Efficiency gains, quality improvements, cycle-time reductions, and defect avoidance are harder to measure. So companies often default to the easiest metric, then act surprised when employees game it. Lauritzen is basically telling boards and CFOs: stop using token burn as your north star.

The strategic stake extends beyond Legora. If your company rolled out an AI access program with leaderboards, the question is not whether people will use the tools. They will. The question is whether your measurement system is driving value or noise. As Uber caps token spend, Amazon turns off usage tracking, and Cerebras' CEO criticizes “unlimited tokens,” the direction of travel is clear. The companies that manage AI responsibly will align incentives with outcomes, not consumption, and they will treat token spend like a budget discipline issue, not a culture buzzword.