Tech workers are on the edge: the race for maximum AI consumption
Artificial intelligence is changing not just how programmers work, but also how they are evaluated, compared, and even how they see themselves within their companies.
And the thermometer for this shift has a very specific name: tokens.
If you haven’t heard of tokenmaxxing yet, brace yourself, because this concept is taking over conversations in the hallways — virtual or otherwise — of the biggest tech companies in the world.
The idea is simple on the surface: the more tokens you consume using AI tools, the more productive you appear to be. But is that really how it works in practice? The answer, like almost everything in tech, isn’t as straightforward as we’d like it to be. 😅
What started as a way to measure usage of tools like ChatGPT, Claude, and Gemini has turned into an all-out corporate competition, complete with internal leaderboards, token budgets as a job perk, and performance reviews tied to how many tokens each engineer manages to burn through per week. Some people are spending over $150,000 a month on AI-powered coding tools. One engineer processed the equivalent of 33 times the entire content of Wikipedia in a single week. And companies are using all of this as criteria to decide who moves up — or who gets left behind. It’s a race that blends real productivity, career anxiety, and, let’s be honest, a good amount of corporate theater. 🎭
What are tokens and why they became a currency of power
Before jumping into the game, it’s worth understanding the pieces. Inside language models — the famous large language models — tokens are the basic units of text processing. Think of them as chunks of words, sometimes a syllable, sometimes a whole word, sometimes a symbol. Every time you send a message to ChatGPT or ask Claude to review a block of code, you’re consuming tokens — both the ones you send and the ones the model sends back. And that consumption has a cost, whether financial or computational.
For a long time, tokens were just a technical metric. Engineering teams monitored them to control API spending and make sure infrastructure budgets didn’t spiral out of control. But when companies started handing out individual token quotas to their engineers — as a workplace benefit, almost like a digital meal plan — the dynamic shifted completely. Suddenly, the volume of tokens consumed started to signal, at least in the collective perception, how much that professional was really using artificial intelligence in their daily work. Whoever consumed more was apparently more integrated into the new era of technology.
This movement didn’t come out of nowhere. Companies like Meta, Shopify, and OpenAI have publicly stated that they expect their employees to use AI tools intensively, and that this usage will be evaluated in performance reviews. Some managers have started rewarding engineers who make heavy use of AI tools and cautioning those who don’t adopt the technology. In this context, tokens stopped being an infrastructure metric and became a professional performance indicator — with all the pressures, distortions, and misaligned incentives that can bring. 🤔
The numbers that impress — and terrify
To get a sense of what’s happening, just look at the numbers that have come out recently. An engineer at OpenAI processed 210 billion tokens in a single week — enough text to fill the entire Wikipedia 33 times over. At Anthropic, a single user of Claude Code, the company’s AI coding system, racked up a bill of over $150,000 in one month.
Receive the best innovation content in your email.
All the news, tips, trends, and resources you're looking for, delivered to your inbox.
By subscribing to the newsletter, you agree to receive communications from Método Viral. We are committed to always protecting and respecting your privacy.
Until recently, power users consumed thousands of tokens per day on tools like ChatGPT, Claude, or Gemini. For reference, a student writing an essay with AI assistance, including multiple rounds of revisions, might use around 10,000 tokens — roughly equivalent to 7,500 words. Using millions of tokens would require consecutive hours in front of a computer doing nothing but typing. And using billions was practically impossible.
The landscape changed dramatically with the arrival of so-called agentic coding tools. These systems can work autonomously for hours on end, reviewing and editing large codebases and writing entire programs from a single prompt. Each agent can spawn sub-agents to handle different parts of a task, generating thousands of tokens at every step of the process. Some AI systems, like the popular open-source assistant OpenClaw, are designed to run 24/7, consuming tokens while their human users sleep.
Ege Erdil, co-founder of Mechanize, an AI startup, estimated that a single agent running full-time can consume 700 million tokens per week. He calculated his own personal consumption at between 1 billion and 10 billion tokens per week. In his view, that’s not even that much when you have agents running continuously.
Max Linder, a software engineer in Stockholm, put it bluntly: he probably spends more on Claude than his own salary. In his case, the employer picks up the token tab. 💸
Tokenmaxxing: real productivity or performance art?
The term tokenmaxxing was born out of exactly this tension. It describes the practice — conscious or not — of maximizing token consumption to appear more productive within corporate evaluation systems. And here’s the core problem: consuming a lot of tokens isn’t necessarily the same as delivering more value. An engineer can use Claude to generate pages and pages of code that never makes it to production. Another might use the same model to fix a critical bug in fifteen minutes with a single well-crafted question. The token volume is completely different. The impact on the business? The second case might be infinitely more valuable.
This disconnect between consumption volume and quality of output is the most dangerous blind spot in the token race. When companies use consumption dashboards to evaluate who’s being productive with artificial intelligence, they risk rewarding people who know how to game the metrics — and not necessarily those generating the best results with the technology. It’s the kind of incentive that, in game theory, is called Goodhart’s Law: when a measure becomes a target, it ceases to be a good measure. The numbers go up, but what they actually measure starts to drift from reality.
Several tech professionals interviewed by The New York Times expressed concern that colleagues are consuming billions of tokens — which can cost thousands of dollars a day — essentially to accumulate bragging rights. Even within the AI labs themselves, where employees have unlimited access to company tools, the idea that all this consumption is genuinely productive seems pretty unlikely. An OpenAI employee, who asked not to be identified because they weren’t authorized to discuss their colleagues’ AI coding habits, put it simply: this doesn’t seem sustainable.
That said, it would be unfair to ignore that there is a reasonable correlation between heavy AI use and real productivity gains. Data published by GitHub about Copilot shows that developers who integrate AI coding tools into their workflow can complete tasks significantly faster in certain contexts. The problem isn’t using AI intensively — the problem is turning usage volume into a ranking of human value within a company. That’s where things start to go off the rails. 😬
The money behind the race
All this runaway token consumption translates into significant revenue for the companies selling them. Anthropic more than doubled its revenue projections in just two months this year, largely because of the accelerated growth of its agentic coding tools. OpenAI recently revealed that Codex, its agentic coding tool, tripled its weekly active users since the beginning of the year, and that total Codex usage, measured in tokens, increased fivefold. Last year, Google reported that its AI models were processing more than 1.3 quadrillion tokens per month.
AI companies aren’t just watching this phenomenon passively — they’re encouraging it. The biggest token consumers receive trophies, rewards, and public recognition. It’s a model that feeds competition and keeps users engaged, spending more and more. For tech executives, if a programmer wants to run a swarm of 10 AI agents executing parallel tasks across separate windows, they’re happy to foot the bill.
On the individual subscriber side, paid plans for Claude and ChatGPT offer a fixed amount of tokens for a monthly fee. Some tokens are cached, meaning the system already has them in memory and doesn’t need to generate them from scratch, and companies charge more for output tokens than input tokens. Anyone who needs more tokens can buy them separately or upgrade to a more expensive plan.
But the savviest users have learned to work around the system, stacking multiple subscriptions or taking advantage of promotional offers. One startup founder revealed that he discovered a Figma AI tool that allowed him to use the equivalent of $70,000 in Claude tokens through an account that cost just $20 a month. He used that loophole to build six software projects simultaneously. A Figma spokesperson clarified that this usage occurred before AI credit limits were enforced, which went into effect recently. 👀
How companies are structuring this race
The picture taking shape across the tech industry is curious and, depending on your perspective, a little unsettling. Companies like Meta and OpenAI have created internal leaderboards — visible dashboards showing which engineers consume the most tokens per week. The logic, according to the managers who champion this approach, is to build a culture of accelerated artificial intelligence adoption. If everyone can see who’s using the most, social pressure and the natural competitiveness of tech teams push the numbers upward. It’s gamification applied to the workplace — and it works, at least for driving up dashboard numbers.
At Meta, managers have started factoring AI usage into performance reviews, rewarding engineers who use the tools intensively. At Shopify, the company has stated that token usage is just one of several metrics used to measure performance, also evaluating how AI improves and amplifies the work being done.
Beyond the rankings, another practice gaining momentum is the distribution of token budgets as a corporate benefit. Instead of giving unrestricted access to tools, some companies set individual monthly quotas. Generous token budgets are becoming a workplace perk for programmers, comparable to dental coverage or free lunch. Those who exceed their quota get approval to expand the limit, signaling that they’re plugged into the intelligent workflow. Those who fall well below it might get a conversation with their manager about why they’re not taking advantage of available tools. This kind of structure creates constant pressure to consume more, regardless of whether the task at hand actually benefits from AI usage.
Gergely Orosz, author of a popular newsletter among software engineers, defended the practice of evaluating professionals through AI leaderboards, calling it an extremely cheap way to learn about new and interesting ways of working. He acknowledged that previous metrics used to assess programmer productivity — like how many lines of code they wrote or how many changes they committed — weren’t perfect either. But his most pointed observation was something else entirely: inside major tech companies, not using AI at a rapid pace is becoming a career risk, regardless of the quality of the output.
It’s worth noting that not all companies are going down this path. There’s a strong contingent of technical leaders advocating for a more qualitative approach: instead of measuring tokens, measure deliverables. Instead of ranking consumption, evaluate impact. Organizations with more traditional engineering cultures have resisted turning AI usage into an individual performance metric, arguing that it distorts incentives and can actually degrade the quality of generated code — after all, code written in a rush, even with AI assistance, still needs to be reviewed, tested, and maintained by humans. ⚙️
Token anxiety and the fear of falling behind
Nikunj Kothari, a venture capitalist in San Francisco, wrote in a recent post about what he called token anxiety. He described a scenario where the tech ecosystem has become obsessed with AI productivity, not human productivity. Kothari said he swapped Netflix for Claude Code and noted that dinner conversations in Silicon Valley have shifted. The old default question used to be what are you building. Now, the question is how many agents do you have running.
This shift in mindset reveals something deeper than a simple market trend. There’s a genuine fear among tech professionals that we’re on the threshold of a radical transformation in the job market for programmers. If we really are on the brink of a white-collar jobs revolution, maybe token anxiety is rational. Nobody wants to be the last programmer writing code manually, without teams of AI agents working around the clock on their behalf.
For the employers footing the bill for all these anxiety-driven tokens, the expense might seem like a reasonable investment to stay ahead of the curve. And for the engineers, the race represents both a genuine opportunity to expand their capabilities and an unprecedented pressure to adapt or become obsolete. 😓
The real impact on coding and the future of engineers
Behind all this talk about metrics and rankings, there’s a very concrete transformation happening in how coding work gets done. Tools like GitHub Copilot, Cursor, Claude Code, and the latest code agents from Anthropic and OpenAI can already write complete functions, refactor entire modules, identify bugs, and even propose system architectures. The engineer who learns to work well with these tools — who knows how to ask the right questions, review the output with a critical eye, and integrate generated code coherently with the rest of the system — can genuinely deliver a lot more in less time.
Most of the tokenmaxxers featured in recent reports were engineers or hobbyist programmers building and maintaining large, complex software using coding agents running in parallel. They generally reported that AI coding tools were making them more productive. However, some also framed their AI usage as a strategic move — a way to signal to colleagues and bosses that they’re keeping up with the times, as the era of purely human coding appears to be winding down.
That’s the upside of the equation, and it’s real. The productivity these tools unlock isn’t marketing fiction. You can see, in concrete situations, engineers solving in hours problems that would have previously taken days. The ability to use artificial intelligence to speed up repetitive coding tasks — like writing unit tests, documenting APIs, or converting code from one language to another — frees up mental bandwidth for what truly requires human creativity: understanding the business problem, making architectural decisions, and making sure the system as a whole makes sense. When token usage reflects that kind of work, the high numbers are absolutely justified.
The big question: are we building something good?
The question hovering over this entire race is unavoidable: are any of these tokenmaxxers actually producing something good? Or are they just spinning their wheels, generating useless code and wasting valuable processing power trying to look busy?
Leaderboards don’t measure quality. They measure volume. And that gap is the Achilles heel of the entire token-based evaluation system. Maybe today’s AI addicts will become tomorrow’s 100x engineers. Or maybe all of this is nothing more than productivity theater — a gleaming tower of tokens, built by the competitive and the fearful, that will crumble the moment we truly understand what constitutes useful work.
The challenge for companies is to build evaluation systems that can distinguish smart usage from performative usage. That means combining quantitative metrics — like token volume — with qualitative assessments of deliverables, peer reviews, and product impact. It also means fostering a culture where engineers feel safe using AI honestly, without needing to inflate numbers to seem more relevant.
Either way, one thing seems certain: we’re going to need a lot more data centers. At the end of the day, the healthiest competition isn’t the race for who burns the most tokens — it’s the race for who ships better, faster, and more reliable software. And that race, thankfully, artificial intelligence still can’t win on its own. 🚀
