Token Champions and the AI Cargo Cult

Someone posted recently about their company running a leaderboard tracking AI token usage. Top 25 users got called “Champions.” People who barely used it were labelled “Sleepers.” The champions had collectively burned through $850,000 since January first.

It took me a minute to fully absorb that.

The kicker is that the Sleeper who posted about it seemed genuinely worried about being at the bottom of the list. Not because they were less productive. Just because they weren’t spending enough of the company’s money on an AI that may or may not have been doing anything useful.

I’ve spent a fair chunk of my career in DevOps. I’ve watched a lot of technology trends roll through organisations like a tide, depositing hype on the shore and then quietly pulling back. But there’s something different about this one. The sheer confidence with which non-technical leadership has decided they understand what AI can do, and more importantly, what it should be used for, is genuinely something to behold.

There’s a thread of comments from people in IT describing situations where management pushed AI solutions for problems that already had perfectly good solutions. Monitoring scripts. Process automation. Error handling. Stuff that’s been solved, reliably and cheaply, since roughly the Clinton administration. One person pointed out that a daemon is a concept from early Unix. You set it, it watches things, it screams at you when something breaks. It does not hallucinate. It does not accidentally decide that a crashed service is actually fine. It does not bill you per query.

The response from management, apparently, is still: “But can we use AI for it?”

I sat with that for a while because I wanted to be fair about it. Maybe some of this is just the natural lag between leadership and technical teams on any new technology. That’s always existed. And frankly, some AI tools are genuinely useful. I’ve used them. Not everything being pushed is cargo cult nonsense.

But there’s a specific flavour of stupid happening right now that feels different. Companies rolling out AI tools company-wide, mandating usage, building leaderboards, and then being surprised when the bill arrives. One person described their workplace hitting its monthly token limit on the second day of the month. Another described migrating four thousand VMs to AWS to cut costs, only to watch the bill nearly double, and now the same leadership is enthusiastic about AI.

There’s a version of this I’ve seen before in enterprise IT. Someone senior comes back from a conference, or reads a Gartner report, or has lunch with a vendor, and suddenly there’s a new thing that will solve everything. Usually it’s expensive, usually it requires significant effort to implement, and usually the people who have to actually implement it can already see three reasons it won’t work as described. The new thing eventually gets replaced by the next new thing, and the cycle continues.

What’s different this time is the scale, the speed, and the fact that the cost is genuinely enormous. We’re not talking about an expensive software licence that sits unused. We’re talking about consumption-based billing that scales with how enthusiastically your entire workforce engages with a chatbot. Measuring that engagement with a leaderboard is not a strategy. It’s a way of ensuring your bill grows as fast as possible while generating no useful data about whether any of it is working.

One comment I read made the point well: you can’t judge a knowledge worker by their outputs, only their outcomes. Lines of code, story points, commits, and now token burns. All of them are proxy metrics that people will optimise for, because that’s what people do when you measure the wrong thing. Several people in the thread admitted they were gaming the token leaderboard just to stay off the radar. Asking AI to do nothing useful, just burning credits to avoid being labelled a Sleeper.

The environmental side of this doesn’t sit easily with me either. Every one of those wasted queries has a real cost that doesn’t appear on the corporate invoice. Water for cooling. Power for the data centres. I’m genuinely interested in what AI can do, and I think some of it is remarkable. But the gap between “this is a useful tool applied thoughtfully” and “burn tokens to win the leaderboard” is not a small gap. It’s the whole thing.

I don’t think the technology is going away. I’m not arguing it should. What I’m watching, though, is an industry-wide experiment in what happens when you deploy powerful and expensive technology without any coherent idea of what success looks like, and then measure usage instead of value.

The bill is already arriving at a lot of companies. The conversations that follow will be interesting.