What AI actually costs
A working method for finding out what your organization's AI usage costs, which savings are real on your own distribution of work rather than on a benchmark, and what to measure so the answer survives a finance conversation.
Why AI bills rose while per-token prices fell, and why cost per completed task is the only unit a finance conversation will accept. The six moves worth making first, in order, starting with the one that returns the most for the least effort. Where the money goes in agentic workflows, which is what enters the context window rather than which model reads it. An honest account of three widely cited savings, routing, semantic caching, and context compression, each of which fails quietly rather than throwing an error. And a fourteen-question worksheet for a single workflow, which you can fill in this afternoon.
Because a guide that withholds its own answer is an advertisement wearing a guide's clothes. If the method is useful and you run it yourself, that is a good outcome. If you would rather we ran it with you, that is what the Studio rungs are for.