Harvey Increases Its AI Usage by 12 Times in Five Months

Exponential Growth in Token Usage at Harvey

Winston Weinberg, the CEO of Harvey, recently revealed a dramatic increase in the use of AI tokens by his company. In January, Harvey was using about 1 trillion tokens per month. However, in just five months, this usage has skyrocketed to between 12 and 13 trillion tokens monthly. This information was shared during a statement made to Business Insider, highlighting an exponential growth that reflects the massive adoption of artificial intelligence technologies by the startup.

Harvey, which specializes in AI applied to the legal field, has seen a twelvefold increase in its token usage in just a few months. Weinberg warned against the excessive use of artificial intelligence for every task, calling it too costly. He compared this situation to the hourly billing of law firms, where every minute must be justified in terms of costs and benefits. This explosion in token usage underscores the growing importance of AI in legal processes, but also the financial challenges it poses.

The Financial Implications of Token Usage

According to Weinberg, Harvey is on the verge of reaching a monthly usage of 12 to 13 trillion tokens, an estimate confirmed by a company spokesperson for the month of May. "The usage is going crazy," he commented, illustrating the scale of this growth. Tokens, which represent the units of text processed by AI models, have become a crucial measure for evaluating the usage and billing of AI services. The more complex a task is or the more powerful the model used, the higher the number of tokens required.

This increase in token usage comes as many tech companies reassess their AI spending. Business Insider reported that some startups are looking to cut costs by "tokenmaxxing," a strategy aimed at optimizing token usage to avoid exorbitant AI bills. This trend is illustrated by concrete examples such as Coinbase, which redirects certain queries to less expensive models, or Uber, whose executives are questioning the profitability of AI coding tools.

The Challenge of Justifying AI Expenses

Weinberg emphasized that the next challenge for companies will be determining the appropriate level of AI usage for each task. He illustrated this point by explaining that a change of control review might justify intensive AI usage, while a simple document summary would not. This reflection is part of a broader debate on how companies must justify their AI expenditures, similar to how law firms justify billable hours.

During a podcast, Weinberg compared this situation to how law firms detail their invoices, indicating the tasks performed and the hourly rates applied in six-minute intervals. "Why do they do this?" he asked, before answering: "It's to show the return on investment (ROI)." As companies spend more on tokens, they will face increasing pressure to demonstrate the effectiveness of these expenditures. "That's a bit of the main problem I think the whole world is about to encounter," he concluded, "That is, 'I just spent 1 billion dollars on tokens. Where is my ROI?'"