I’m following what AI costs, where it genuinely helps, how it changes finance operations, and which questions become more interesting once it meets real work.
Model price is only one part of the economics. The surrounding workflow includes people, retries, review, integrations, failures, and whatever the technology causes downstream.
Retries, review, latency, failure, and correction can turn an inexpensive model into an expensive operating system.
Read the field noteUsage, ownership, allocation, savings claims, and the distance between a provider bill and the real economics of the work.
An AI vendor invoice tells Finance what the company owes. It rarely explains what the company bought, who consumed it, or whether the spending created value.
Read the field noteCredentials are potential boundaries for ownership, budgeting, allocation, and accountability.
Read the field noteA spending alert identifies that something changed. The useful work is determining why, what it means, and what should happen next.
Read the field noteProcesses rarely become simpler just because a tool was added. These notes follow exceptions, hidden rules, human decisions, and operational dependencies.
Finance automation should be evaluated by what happens when the normal process fails.
Read the field noteSpreadsheets often carry hidden rules about authority, precedence, permission, transitions, and change.
Read the field noteA formula can calculate the expected result while the process around it remains difficult to understand or trust.
Read the field noteAI changes who proposes, who reviews, who decides, and who remains responsible when the result reaches another system.
The useful question is where human judgment belongs and what information a reviewer needs.
Read the field noteA chart of accounts explains where activity belongs. A chain of command explains who may decide.
Read the field noteUseful prototypes can become operational dependencies before anyone notices that the boundary changed.
Read the field noteA small look at the system that releases an approved Close to Code article and preserves what happened.
Read the system note