Notes from the work
Notes on AI FinOps.
Short practitioner essays about where AI cost meets accepted outcomes, workforce capacity, governance, and the way the work actually runs.
Infrastructure Economics · Practice
The agent billing problem.
Agents work in bursts and idle between turns, but provisioned infrastructure bills the whole hour. Active-CPU billing and pause/resume change the math.
Token Waste · Practice
The leftover prompt is the bill.
Cache misses, leftover verification rituals, and max effort on a flat task show up as tokens. Token price is downstream of those choices.
Capital Allocation · Practice
The peanut butter trap in enterprise AI.
Spreading USD 250,000 across twenty departmental sandboxes produces demos and meeting hours. It does not produce a production queue.
Field Notes · Practice
Where the AI money actually goes.
Cloud hosting was easy: you rented a box and paid the bill. AI spend is completely different. Here is what five pieces of research taught me about wandering prompts, multi-agent loops, and how to stop burning budget.