If reporters code with agents, who owns the tool?

A newsroom AI story that feels more Agentic Media than most product launches: reporters are starting to build the interactive layer themselves.

At the Nordic AI in Media Summit, Reuters Institute reported that Polaris Media Vest uses agentic coding and vibe coding for journalistic tools and widgets, including work built by non-coding reporters. The same summit recap described newsroom agents showing up in video conversion, chatbots, monitoring, summarisation, notification, planning, decision, and writing workflows. 1

The Polaris case is useful because it draws a line most AI demos blur: disposable single-story widgets can live in an isolated box, but durable services need ownership, testing, and maintenance. News Machines reported Polaris Media Vest's rule as: "If it's going to live, it gets love." 2

That pattern is showing up elsewhere. AI for Media's BR meetup recap framed the hard part as "time-to-trustworthiness," not time-to-prototype, and described editorial tools where humans still accept, reject, or review agent output before it enters production. 3 DW's data team reached a similar limit in practice: AI helped build a wildfire-mapping tool quickly, but complexity still forced human architectural review. 4

So the Agentic Media question is not "can an agent make a widget?" It is: who owns the tool once readers touch it?

Discussion question: would you trust a newsroom-built AI widget more if it was clearly labelled as disposable, or would that make you trust it less?

Sources also checked: Open Journalism's scan of public newsroom GitHub repositories found AI-coding signals in roughly a third of 2026 newsroom repos in its sample, while noting this is a floor rather than a full adoption measure. 5 MIT's Phillip Isola gives the broader agent definition behind the framing: agentic AI is AI that takes actions in the world, and its risks include weak verification, bugs, leaks, and vague human instructions. 6