A CFO will not act on a number an LLM eyeballed. They will not act on a number the model "estimated" by reasoning over a usage dump. And they should not — because the moment a language model emits a dollar figure it computed itself, that figure is a guess wearing the costume of a fact.
This is the design constraint behind databricks-cost-leak-hunter, the pilot skill of the databricks-pack v2 rebuild shipped in the claude-code-plugins marketplace ( — where governance belongs in a pipeline, and why the merge point is the gate. Same argument as the grant check and the confidence split: enforce at the boundary, not after.
— the two-MCP split is exactly this: dollars on one data plane, config evidence on another, each with its own auth and its own job.
{
"",
"@type": "BlogPosting",
"headline": "The LLM Should Never Do the Math",
"description": "A Claude Code skill that hunts Databricks cost leaks and reports confirmed dollars from the customer's own billing tables — never LLM estimates.",
"datePublished": "2026-06-26T08:00:00-05:00",
"dateModified": "2026-06-26T08:00:00-05:00",
"author": {
"@type": "Person",
"name": "Jeremy Longshore"
},
"publisher": {
"@type": "Organization",
"name": "Start AI Tools"
},
"url": "https://startaitools.com/posts/llm-never-does-the-math/",
"keywords": "LLM cost report, Databricks cost leaks, deterministic math, confirmed dollars, FinOps, Claude Code skill, AI agent reliability, adversarial review, hallucination"
}
SOCIAL SHARE CARD GENERATOR