Tokenmaxxing is like Looksmaxxing: Gamifying the Wrong Metric
AI Daily Brief
Source of Bullshit
In today's edition of the newsletter The AI Daily Brief, there was a defense of tokenmaxxing AI usage: the gamification by leaderboard and other incentives for companies to encourage AI use by recognizing (and sometimes bestowing prizes on) users burning through as many tokens as possible.
The defense goes on to poo-poo the stories that have come out about tokenmaxxing as not understanding the "true value" that the practice brings: encouraging AI usage (hopefully novel, productive AI usage):
The logical leap being made is [...] that most of the demand is people using AI for silly, non-consequential purposes rather than real, valuable work. That leap requires stacking three logical fallacies on top of each other. First, selection bias [...] Second, hasty generalization [...] Third, category error: using gaming behavior as evidence about the quality of the technology, when the only thing it's evidence of is the incentive structure.
What this apologia fails to appreciate is that even if we take all of the above as given, the entire idea that "token consumption" is any kind of a meaningful metric in development is incorrect. Token consumption by itself is a meaningless big-number-good, biggest-number-best view on the state of the art in development. Just like any other free-floating metric, token consumption is valueless, with the sole exception of perhaps water consumption and the impending heat death of the universe.
This attitude shares a lot in common with the current Internet subculture/trend that is looksmaxxing (LM). Looksmaxxing, if you're not familiar, is the where men do everything to maximize their physical attractiveness. Diet and exercise are the starting point, but smashing your cheekbones with a meat tenderizer, DIY cosmetic surgery and disordered eating are all in the playbook.
Like tokenmaxxing, LM takes a single metric (a person's physical attractiveness), and seeks to optimize it as a panacea for a range of other metrics that are far more important to success in romantic relationships. The idea is that so long as you shine like a supernova in the looks department, a lack of mental wellness, emotional intelligence, spiritual fulfillment or even a shared hobby, can be foregone. Let's face it, you can look like Brad Pitt on a good day but if you're as interesting as watching paint dry because you've spent your morning starving yourself and pummeling your face with a mallet, you might not achieve anything meaningful. (YMMV.)
Full disclosure, I have a Claude Max account. I use Claude Code and Cowork extensively, both for client work and my own development. I just checked my usage with npx ccusage@latest monthly, and here's what I found:
180M tokens consumed since January 0f 2026. By tokenmaxxing standards, that's completely anemic. Very beta, very demure. Here's the part where I tell you that I have (increasingly) used Claude to handle routine debugging, feature planning, architecture and even full-scale PoC creation. I'm easily on track to double that 98M in May, without breaking a sweat.
If I'm being completely honest, unless I start running agents autonomously to do absolutely trivial tasks into Claude's hands, I have no idea how to consume more tokens than I currently am. I also don't use Claude frivilously. I make sure that tasks are as atomic and self-contained as possible. I may be overly verbose with my instructions to Claude (more tokens!), but that also limits the scope of what Claude ingests per prompt (less tokens overall!) - and I always use Claude in plan mode first and foremost.
My bang-per-token-buck is ENORMOUS, if you ask me, and I'm barely ticking the odometer over in terms of what my subscription allows. I have yet to hit my daily limit, even in the "nerfed token spend" of Opus 4.7. I might be naïve, but I don't think so.
In the end, what matters is results, not the process you used to get there. I think it's telling that companies like Meta are pitching token consumption as a metric of any value, while the experience of their products (Facebook, Instagram, etc) continue to degrade into AI-slop nightmares. Pointing to a few billion burnt tokens as a metric of "we're doing something about the problem" is an easy way to deflect from the fact that absolutely NO ONE can make the argument that consumption in any way improved user experience.
If these companies were serious about the promise of what AI can bring to their platforms, then they'd hold X-Prize-style competitions:
- Reduce or eliminate the proliferation of content shown to promote eating disorders in adolescent users utilizing the existing public-facing Instagram API. Prize: $5M
- Ensure that all accounts purporting to be human-run are free from automated activity, inflated engagement or purchased influence. Prize: $10M
Those are the sorts of metrics that can actually push these ecosystems forward, or at least pull them out of the enshittification spiral we're witnessing.