AI Tokenmaxxing: Companies Reward Heavy AI Use – NYT Report

The pressure to demonstrate productivity in the tech sector is taking a peculiar turn, with reports emerging that employees at companies like Meta and OpenAI are now being evaluated, in part, by how quickly they consume computing resources – specifically, large language model (LLM) tokens. This metric, tracked on internal leaderboards, is raising questions about the true measure of value creation in an era increasingly reliant on artificial intelligence.

The Rise of “Tokenmaxxing”

According to a recent column by New York Times’ Kevin Roose, the practice extends beyond simple usage tracking. Managers at Meta and Shopify, in particular, are reportedly “rewarding workers who make heavy use of A.I. Tools and chastening those who don’t.” This has led to a phenomenon Roose terms “tokenmaxxing” – a drive to maximize token consumption, even if the output isn’t necessarily commensurate with the cost. The situation is drawing comparisons to odd metrics of success, like NBA mascots judged by the volume of t-shirts fired from cannons, albeit with the t-shirts being exceptionally expensive Hermès creations. The Awl detailed the cost of such luxury goods back in 2013.

The numbers involved are substantial. Roose reports that one OpenAI engineer reportedly processed 210 billion tokens – equivalent to the entire text of Wikipedia 33 times over. A Swedish software engineer told Roose his company spends more on his Claude Code tokens than his entire salary. This surge in token usage is closely tied to the growing popularity of “claws,” or agentic AI platforms like OpenClaw, which have become a focal point for AI enthusiasts, prompting OpenAI to acquire OpenClaw’s creator in a bid to stay competitive. Gizmodo covered the acquisition of OpenClaw’s creator by OpenAI.

The Cost of Computation

The financial implications are significant. OpenAI president Greg Brockman recently touted that GPT-5.4 is processing 5 trillion tokens per day, generating an annualized run rate of $1 billion in new revenue. Brockman shared this data on Twitter, framing it as a positive indicator of the model’s adoption. However, this highlights the inherent cost of running these powerful LLMs. Tokens aren’t free; they represent computational power, and the more tokens consumed, the higher the expense. The focus on volume, raises questions about whether companies are prioritizing efficient AI usage or simply rewarding expenditure.

Beyond OpenClaw: Claude Code and Mobile Integration

The trend isn’t limited to platforms like OpenClaw. Claude Code, Anthropic’s coding-focused LLM, is evolving to mirror some of OpenClaw’s capabilities. A recent update allows users to interact with Claude Code directly from their phones via platforms like Telegram and Discord, further facilitating constant, on-the-go token consumption. The promotional material for this feature even features a cartoon crustacean – a playful, if somewhat ironic, symbol of LLM token profligacy.

What Does This Mean for Tech Workers?

The shift in evaluation metrics could have a chilling effect on thoughtful AI implementation. If employees are incentivized to simply *use* AI as much as possible, it could discourage them from carefully considering whether AI is the *right* tool for a given task. This could lead to inefficient workflows, wasted resources, and potentially lower-quality outputs. The analogy to soldiers firing bullets indiscriminately, as Roose points out, is apt – volume doesn’t necessarily equate to effectiveness. The pressure to “tokenmax” could also create a stressful work environment, where employees feel compelled to constantly engage with AI tools, even when it’s not beneficial.

A Broader Question of Value

This practice also speaks to a larger issue: how do companies measure the value of AI? Simply tracking token consumption provides a quantifiable metric, but it doesn’t capture the nuances of AI’s impact. Are employees using AI to solve complex problems, automate tedious tasks, or simply to generate more code, regardless of its quality? The focus on tokens risks reducing AI to a cost center, rather than a strategic asset. The New York Times’ Hard Fork column, where Roose regularly contributes, explores these evolving dynamics in the tech industry. Hard Fork provides ongoing coverage of the rapidly changing tech landscape.

The Regulatory Landscape and Future Scrutiny

While there’s no immediate regulatory response to “tokenmaxxing,” the broader scrutiny of AI development and deployment is increasing. Concerns about the environmental impact of large-scale AI training, the potential for bias in AI algorithms, and the ethical implications of AI-driven decision-making are all gaining traction. As AI becomes more pervasive, regulators are likely to demand greater transparency and accountability from companies, potentially including a closer look at how they measure and incentivize AI usage. Gizmodo’s coverage highlights the growing awareness of these issues.

Looking ahead, it’s likely that companies will need to develop more sophisticated metrics for evaluating the value of AI. These metrics should focus on outcomes, such as increased efficiency, improved product quality, and enhanced customer satisfaction, rather than simply on token consumption. The current trend of “tokenmaxxing” may prove to be a short-lived experiment, as companies realize that true AI success lies not in how much AI is used, but in how effectively it’s applied.