Back of the Envelope Math for AI Systems

The Estimate Is the Interview

Classic system design interviews already have a named step for this: back-of-the-envelope estimation, rough math done out loud to size a system.

AI system design also needs explicit estimates because token counts, model choices, and serving methods can produce very different costs.

For example: "At an assumed 3 per million input tokens, 10 billion input tokens cost 30K per day. I would evaluate caching and routing against that budget." This is more useful than saying only that the system is expensive.