Tokenmaxxing Is Just LOC Metrics Wearing A New Outfit

There's a “new” metric floating around called "tokenmaxxing," and it's starting to gain traction in a way that feels eerily familiar. It reminds me a lot of the short lived engineering metric that measured productivity by lines of code (LOC) produced… a metric that never actually worked.

LOC failed because output volume is never equivalent to output value. Quality isn't additive after the fact, and quantity is an independent variable, so conflating the two inevitably leads to mismanagement, wasteful spending, and team dissatisfaction (sound familiar yet?). Tokenmaxxing is walking down that same path.

Seeing it gain even a small amount of traction is surprising, though not entirely unexpected. Like LOC, it functions more as a vanity metric and a wider investment validation strategy than anything truly productive. Given the rising pressure around adoption rates and real productivity impact, I suppose managers need to do something to justify the time and spend… even if history suggests it probably won't deliver.

And leaderboards based on token burn? That's such a bad idea. Waste is never a solution, especially when compute is so heavily subsidized. Plus (under pressure), anyone can lean on a wide range of workarounds, like open source tools or Ralph loops, that automate token usage and run open ended cycles without any manual input.

So before we crown a new favorite metric, maybe we should ask what we're actually trying to measure. Value, not volume, is still the thing that matters… we learned that lesson once already. If we can keep that in mind, there's a real chance we build smarter systems this time (and maybe skip repeating history entirely).