US Tokenmaxxing Costs Rise Without Productivity Spike Amid AI Hype

• A corporate fad of 'tokenmaxxing' has hit its limits as workplaces see costs rise without productivity spike.
• Companies are reevaluating their approach to AI, with many realizing they have been defaulting to using high-capacity models for tasks that don't require them.
• Rising costs and concerns about data protection and intellectual property risks are leading companies to prioritize ROI when investing in AI.
• Lawyers and compliance officers should be aware of the potential for over-reliance on expensive AI tools, which can lead to unnecessary data exposure and intellectual property risks.

It's very easy to create something you don't need with AI.

A corporate fad of 'tokenmaxxing' on artificial intelligence technology has hit its limits as workplaces throwing AI at everything are seeing the costs rise without a similar spike in productivity. The trend, which started as tech industry-fueled hype over squeezing as much AI-generated work as possible out of products like OpenAI's ChatGPT and Anthropic's Claude, has shifted to a backlash. Executives were previously promoting high token consumption as a signal of high-performing employees, but now many are reevaluating their approach due to rising costs.

The stereotypical 'tokenmaxxer' was seen as someone who would stay up late orchestrating an army of 24-hour AI agents performing work on their behalf. However, experts like Vincent Gusdorf, head of AI analytics at Moody's Ratings, warn that this approach can lead to unnecessary data exposure and intellectual property risks.

As bills started to pile in, people realized that those new tools are quite expensive and need to be used wisely. Companies are now taking a closer look at returns on their AI investments, with some realizing they have been defaulting to using high-capacity models like Anthropic's Claude Opus 4.6 for tasks that don't require them.

The trend of tokenmaxxing has raised concerns about data protection and intellectual property risks. Microsoft CEO Satya Nadella warned in a recent blog post that customers of leading AI models are paying twice for AI, first in spending on tokens and second by feeding all their proprietary data to them. This has led to a reevaluation of the use of expensive AI tools and a search for more cost-effective solutions.

Palantir CEO Alex Karp went further, stating that something had gone 'completely wrong' with the way companies are using AI. He said he was channeling the voice of American businesses privately 'livid' about paying so much for tokens that create no value. This sentiment is echoed by Bain & Company management consultant Jue Wang, who notes that many big businesses have been taking a closer look at returns on their AI investments.

The rising costs of token-based AI pricing are also leading to concerns about the ROI of AI investments. As Gusdorf noted, 'It's very easy to create something you don't need with AI.' This highlights the need for companies to use AI wisely and not default to using high-capacity models for tasks that don't require them.

The shift away from tokenmaxxing is significant because it highlights the need for companies to be more mindful of their AI investments. As costs continue to rise, companies must reevaluate their approach and ensure they are getting value from their AI tools.

Lawyers and compliance officers should be aware of the potential for over-reliance on expensive AI tools, which can lead to unnecessary data exposure and intellectual property risks. Companies must take a more disciplined approach to using AI, focusing on tasks that require high-capacity models and avoiding defaulting to using them for everything.

The trend also underscores the need for companies to prioritize ROI when investing in AI. As Wang noted, 'Not everything needs a Claude Opus 4.6.' Companies must be more selective in their use of AI tools and focus on getting value from their investments.

Lawyers and compliance officers should be aware of the potential for over-reliance on expensive AI tools, which can lead to unnecessary data exposure and intellectual property risks.

Source: Original reporting via The New York Times