What's this about?
A team lead at a Silicon Valley AI infra company ran an aggressive experiment: a $1M budget of unlimited AI tokens for a ~20-person team, to see what happens when cost is no object. Baoyu (dotey) amplified the writeup. The conclusion: unlimited tokens is impractical — it can cost more than the humans, people got *more* tired, and gains hit a ceiling because org structures are still built for people, not AI.
What exactly is a "token"?
A token is the unit AI models read and write — roughly a word-piece — and the unit you're billed on. "Unlimited tokens" means removing the usual cost cap on how much the AI can read and generate. The experiment tested whether simply letting a team spend infinitely on AI unlocks proportional productivity. It didn't.
What's the community saying?
original thread (Kay2289123) 783 likes; dotey's summary 330 likes.
- Diminishing returns: the headline finding — throwing infinite tokens at one team hits a marginal-benefit wall fast.
- Org design is the bottleneck: the biggest gains come from going 0→1 on AI adoption, not from one AI-native team burning more; companies are still structured for humans.
- Loss of control: people who outsourced their thinking to AI lost grip on project details and had to *ask the AI* to answer their own colleagues.