Tokenmaxxing in IT: $150,000 a Month on Claude Code and OpenAI Leaderboards
Developers are spending billions of tokens on agentic AI coding tools. One Claude Code user from Anthropic spent $150,000 in a month. An OpenAI engineer processed 210 billion tokens in a week—equivalent to 33 full copies of Wikipedia. This phenomenon, known as tokenmaxxing, reflects a competition in LLM usage volume to demonstrate productivity.
Agentic systems autonomously generate subagents, process tasks for hours, and consume thousands of tokens per iteration. According to Mechanize co-founder Ege Erdil, a continuously running agent spends 700 million tokens per week. Erdil himself uses 1–10 billion tokens weekly.
Factors Driving Tokenmaxxing
Companies are fueling the trend with internal metrics. OpenAI introduced token leaderboards. Shopify factors in AI usage when evaluating employee performance. Token budgets are becoming a standard perk, like corporate benefits.
Loophole example: a startup founder got access to $70,000 in Claude credits through a $20/month Figma subscription—before limits were introduced. Such practices accelerate resource consumption.
For AI vendors, this is lucrative: Anthropic doubled its revenue forecast in two months thanks to agents. OpenAI reported a tripling of active Codex users and fivefold growth in total token volume since the start of the year.
- Scale of consumption: 210 billion tokens/week for top users.
- Autonomy: Agents run 24/7 without intervention.
- Success metrics: Token leaderboards instead of code evaluation.
- Economics: Tokens as the currency of productivity in the tech industry.
- Market growth: Tripling of users and 5x traffic at OpenAI.
Quality and Metrics Problems
Tokenmaxxing focuses on volume, ignoring output quality. Gergely Orosz from his engineering newsletter notes: in big companies, skipping AI becomes a career risk, regardless of results.
Kevin Roose from the NYT spends 4–5 hours daily on Claude Code, using just a few million tokens—“beginner numbers” compared to leaders' billions. Venture investor Nikunj Kothari coined the term "token anxiety": conversations now revolve around the number of agents launched, not the product.
Continuous agent operation requires optimizing prompts and chains of thought to minimize useless cycles. Senior developers highlight risks: overfitting to noisy LLM data undermines pipeline reliability.
What Matters
- Tokenmaxxing boosts AI company revenue but masks code quality issues.
- Agentic tools consume 700 million–10 billion tokens/week per user.
- OpenAI leaderboards and Shopify metrics tie careers to token volume.
- Risks: career pressure without focus on results.
- The trend is shifting culture: from "what are you building" to "how many agents have you launched".
— Editorial Team
No comments yet.