In September a small internal pilot of an AI-generated daily financial briefing was conducted. The objective was to establish whether a large language model could be made to produce a financial morning note of the calibre that a serious reader would actually finish, given a sufficiently constrained prompt and a sufficiently disciplined data context.

The pilot answered in the affirmative, but with significant qualifications. The model was capable of the prose; the discipline had to come from elsewhere, specifically from the verification layer that ensures every figure stated in the briefing has a verifiable source and a correct time-stamping. The pilot established the editorial template that would, several months later, become Drusus Daily.