Jacob Lauritzen, CTO of legal AI startup Legora, said that linking the use of tokens for AI tools to employee rankings and performance evaluations can easily lead to "tokenmaxing," which is the deliberate consumption of more tokens in order to appear more proactive on internal leaderboards, rather than improving actual work efficiency.
More emphasis on output than consumption
In his podcast "20VC," he said that this approach leads employees to "burn tokens just to look better" and doesn't actually improve productivity. A more effective approach is to have employees demonstrate how they leverage AI to complete projects and what specific efficiency improvements it brings, through hackathons, internal demos, and other methods.
Lauritzen believes that companies should reward employees who are "more efficient and productive," rather than simply those who "use more AI." In his view, the use of AI itself is not the goal; the key is whether it leads to higher quality work results.
High-growth companies are still willing to pay for efficiency.
However, he also noted that for fast-growing companies like Legora, the opportunity cost of not using AI is equally high. If the additional token expenditure translates into an approximately 20% efficiency improvement, such an investment still makes sense.
Companies are starting to tighten their AI budgets
This statement comes as the way the tech industry manages the use of AI is changing. Previously, some companies encouraged employees to try out AI tools through leaderboards and internal dashboards, but as costs rise, more and more businesses are beginning to worry whether this incentive is counterproductive.
- Uber has set a monthly spending limit of $1,500 for each AI tool.
- The Financial Times reports that Amazon has shut down its internal AI usage leaderboard.
- Cerebras CEO criticizes providing employees with unlimited tokens
At a Bloomberg conference last week, Cerebras Systems CEO Andrew Feldman said that not all tasks need to invoke high-cost models, and enterprises should choose cheaper open-source models based on task complexity to improve token usage efficiency.
Judging from the statements made by Legora, Uber, Amazon, and Cerebras, tech companies are shifting their focus in managing AI from "encouraging as much use as possible" to "pursuing actual output while controlling costs."
