intelligence should not be treated as a tool subscription you buy but as capital you allocate. a few ways to optimise for it: 1/ budget it by the job, not the tool. ask what a specific task is worth if a model does it well, and then what it costs to run. you'll then be able to fund the ones that pay back. 2/ match the model to what's at stake. if you don't put your best people on low-stakes work, think about intelligence the same way: a cheap model for the high-volume tasks and an expensive one only where a mistake is costly. most teams overpay by running a premium model across everything. 3/ don't over-commit all at once, as intelligence typically gets cheaper every few months. a task that isn't worth automating today often is 2 quarters later. underwrite in short cycles rather than locking in one big flat contract. @stripe also put it well in their openrouter letter: you need to reason for the cost and return of every unit of intelligence the way you would for capital.