CFO Asks What Company Received for $2.3 Million in Tokens; Engineering Responds With 400-Page Report Generated by Tokens
The report's executive summary, written by the model, describes the spend as "well within expectations," which the CFO noted were the model's.
BOSTON — The chief financial officer of enterprise software company Halden asked the engineering organization Monday to account for $2.3 million in annual model inference spend, and received by Wednesday a 400-page report on AI adoption and impact that the engineering team confirmed was produced, in large part, by the models in question.
"It is thorough," said CFO Marguerite Lindgren. "It has charts. It has an executive summary. The executive summary says the spend is 'well within expectations.' I asked whose expectations. Nobody was sure."
The report cites a 41% increase in pull requests, a 22% reduction in time to first commit for new hires, and 1.4 million assistant sessions, a figure that includes the sessions used to write the report. It also includes a section titled "Limitations of This Analysis," which the CFO described as "the best part" and "suspiciously self-aware."
Engineering leadership defended the document. "She asked what we got for the money," said VP of engineering Tomás Reyes. "We got the ability to answer that question in two days. Last year it would have taken a quarter and a contractor."
Lindgren said she has requested a one-page version. Engineering said it would take "about a minute" and asked whether the page should be optimized "for reading or for approval."
The finance team has since begun tagging inference spend by department, a change that revealed the largest single consumer of tokens to be an internal service named "default," which, according to the cost dashboard, does not report to anyone.
At press time, the one-page version had arrived, and was three pages.