Generative AI has been a part of our lives for a few years now, and much has changed since its arrival. It has practically replaced older assistants like Siri, Alexa, or Google Assistant, proving much more effective in providing answers and assistance. At work, we’ve moved from manually searching the internet for everything to AI providing the answers, and more recently, agents have begun to be used for task automation. The problem is that large companies reward those who use AI the most, leading to the rise of “tokenmaxxing” and escalating costs, as employees start using AI agents that consume up to 1,000 times more tokens.
Today, we have dozens of AI models available for use, though the main and most recognized ones remain the same. OpenAI with ChatGPT continues to lead in popularity, followed by Google with Gemini and Anthropic with Claude. Additionally, Grok on X is widely used, though limited to its platform, and Perplexity AI is employed for tasks like advanced web searches. On the other hand, we also have AI agents, though these are more reserved for enterprise use due to their higher token consumption and, consequently, greater expense.
Big Tech Employees at Meta, Amazon, and Microsoft Take “Tokenmaxxing” to the Next Level Using AI Agents That Consume 1,000x More Tokens, Skyrocketing Costs
The concern about token consumption to keep monthly bills manageable is typically for individual AI users, indie developers, or small startups. However, at the opposite extreme, major US tech companies actually incentivize and reward employees for using more AI tokens. This has fueled the popularity of ‘tokenmaxxing,’ a method to maximize AI token usage for better standing within these companies. The problem is evident: massive token consumption drives up costs, and it’s now worse because AI agents are being overused.
Agents enable task automation and are highly valuable tools for accelerating work. However, in this environment where tokenmaxxing is the sole focus, neither productivity nor work quality is prioritized; instead, the emphasis is on spending more. AI agents can consume 1,000 times more tokens than a generative text AI model, and if employees use multiple AI agents simultaneously, the expenditure can be astronomical. Months ago, we saw instances where AI could cost more per month than an employee’s salary, and this situation is now set to worsen.
A Point Will Come When Companies Implement Rules and Employee Monitoring to Prevent Unnecessary Expenses Due to AI Usage
An NVIDIA director announced that his team was spending more on AI than they earned, and Jensen Huang established that he would provide tokens equivalent to half of their annual salary, allowing employees to use them or even sell them for economic benefit. It’s clear that companies are blinded by the promise of increased productivity and haven’t even considered a logical and responsible approach to AI usage. Companies like Microsoft, Meta, and Amazon incentivize employees to use AI, and there’s even a form of ‘internal ranking‘ where those who consume the most tokens rank higher.
Without any form of control or review over token usage, this situation will continue to deteriorate. At some point, companies will realize that using artificial intelligence in this manner is not profitable at all and will implement some form of monitoring or regulations to establish certain limits.
