Imagine an office rewarding employees for the number of pages they print.
Soon, the printers are working overtime. Paper stacks grow taller, and everyone looks exceptionally busy.
But did anyone check whether those pages were useful?
Welcome to a surprisingly relevant question in the age of AI!
When Tokens Become the Scoreboard
Every time we interact with an AI tool, our text is broken into smaller pieces called tokens. These are the units the system processes to understand requests and generate responses.
The more we interact, the more tokens we may consume.
Now, as organisations increasingly explore ways to measure AI adoption, token usage can become an attractive number to track.
And that’s where tokenmaxxing enters the conversation — the practice of deliberately maximising AI token consumption, sometimes to demonstrate greater AI usage.
At first glance, higher consumption might suggest greater engagement. But does it necessarily indicate greater productivity?
Not quite.
More Tokens, More Progress? Or Just More Activity?
Consider two professionals tackling the same complex problem.
One uses a few AI prompts, spends several hours figuring things out independently, and eventually arrives at a partial solution.
The other engages in extensive brainstorming with AI, questions suggestions, tests alternatives, identifies mistakes and eventually produces a more complete solution in considerably less time.
The second person might have consumed ten times as many tokens.
But was that wasteful? Or was it time well spent?
Now imagine a third person repeatedly generating responses merely to increase their usage count.
Similar token consumption. Very different purpose.
More tokens don’t automatically mean greater productivity. But fewer tokens don’t automatically mean greater efficiency either.
The missing piece is what those interactions actually accomplished.
And Where Does Human Thinking Fit In?
Using AI frequently isn’t necessarily the same as engaging with it meaningfully.
There’s a difference between repeatedly accepting generated answers and actively questioning, refining and testing them.
A lengthy discussion with AI might involve exploring possibilities, challenging assumptions or resolving a problem that initially seemed unsolvable.
Conversely, a short interaction might be all an experienced professional needs to complete a task effectively.
Neither the number of prompts nor the volume of tokens tells us how much genuine thinking happened.
And here’s another interesting possibility: sometimes the most thoughtful decision might be not to use AI at all.
Should that count as poor AI adoption?
When Measurement Changes Behaviour
Organisations understandably want to know whether their investment in AI is helping employees.
Usage figures are convenient. They’re visible, countable and relatively easy to compare.
But when such figures become performance targets, an unintended problem can emerge.
People may begin optimising their behaviour for the measurement rather than the work itself.
Reward excessive usage, and unnecessary activity may flourish.
Discourage high usage indiscriminately, and valuable experimentation or deeper problem-solving might suffer.
Neither extreme guarantees better results.
A fairer discussion would consider what was accomplished, how well it was accomplished, and whether AI genuinely helped compared with reasonable alternatives.
Even those comparisons aren’t straightforward. Different tasks require different levels of effort, and not every valuable outcome can be reduced to a convenient number.
A Little Food for Thought!
If someone uses thousands of tokens to solve a difficult problem in two hours instead of ten, was that excessive consumption or effective collaboration?
If another person achieves the same result with fewer tokens but considerably more effort, which approach was actually more efficient?
And if employees are encouraged to demonstrate AI adoption through usage figures, could the measurement itself begin influencing how they choose to work?
Final Thoughts
Tokenmaxxing raises a bigger question than simply whether people are using too much AI.
It asks whether we’re confusing the amount of AI activity with the value of human–AI collaboration.
Sometimes more interaction is unnecessary. Sometimes it’s exactly what a difficult problem requires.
Perhaps the real challenge isn’t deciding how many tokens are too many.
It’s understanding whether all those tokens helped us achieve something worthwhile.
After all, a printer naturally takes longer to print more pages. But what matters isn’t how many pages come out — it’s what’s printed on them.