Tokenmaxxing is not a productivity strategy

At an AI-themed meetup in Hong Kong last month, a software developer proudly shared that he usually hits the token limit of his two Claude Max accounts every day with his coding efforts.

Tokens are a measure of AI computing capacity – the more you have, the more you can do with Claude or ChatGPT. Frontier model providers like Anthropic and OpenAI enforce daily caps to manage demand and access for inference, but users can pay for more tokens.

This system has enabled a new form of flexing among AI builders called “tokenmaxxing.”

The term, which started circulating earlier this year in tech and AI-native circles, describes a practice of maximizing AI usage by continuously running as many prompts, agents, and automated workflows as possible. You get serious street cred if you complain about not having enough tokens, like the software developer above.

His comment made me feel momentarily anxious as my team doesn’t max out their daily Claude token limits. What are we missing out on?

See also: Fixing doctor burnout with AI is harder than it looks in SEA

I had to remind myself that for my company, token burn is the wrong thing to track. We’re an AI consultancy, not a software house – token burn is meaningless if there are no visible outcomes. All I’d be watching is “number go up,” not whether the work is better or whether we’re doing things we couldn’t do before.

Communications teams are a useful case study, but the measurement problem is the same across functions. For companies that use AI tools, track vanity metrics like token burn if you must, but make sure you’re measuring what really matters.

The metric gap

I mostly work with communications teams of large enterprises across Asia. Like most corporate teams, they have yet to figure out what a meaningful metric for AI adoption looks like.

A content writer at an insurance company told me recently that no AI-related KPIs or benchmarks had ever been raised with him. I also learned that at an international investment bank, the most sophisticated metric in place was a manager dashboard showing prompt counts per employee – who’s engaging and who isn’t. That’s useful as a first signal, but not much beyond that.

These teams are not behind; in fact, they are probably in the majority. According to a 2025 survey by McKinsey, 62% of companies are either experimenting with or starting their AI programs. It’s highly likely that these firms have not solidified AI measurability.

For all organizations, it’s important to understand how AI productivity should be measured before someone sets the wrong targets.

What should be tracked

Beyond the burn