Summary
CJR assembled a cross-section of newsroom leaders and practitioners to explain where AI is already useful, where it still causes harm, and which tasks deserve the most skepticism. The piece is a strong legacy reference because it captures a transitional moment when news organizations had moved past pure theory and were starting to describe bounded, task-specific AI use in public.
Why It Matters
This is a valuable direct journalism record because it compresses several distinct newsroom use cases into one well-sourced practitioner map.
- it covers coding help, transcript handling, translation, metadata extraction, tip management, and visual-investigation support
- it also documents what many practitioners still reject, especially generative writing that obscures reporting accountability
- it matters because it shows that "using AI in journalism" had already split into narrow workflow approvals rather than broad endorsement
- it strengthens the archive's legacy newsroom-use lane with concrete examples from Reuters, Semafor, CalMatters, WBEZ, and The Washington Post
What the Source Says
The feature says Reuters developers were already using AI-assisted coding heavily enough that one participant estimated about a quarter of his code output involved AI support. It reports that Semafor used AI to help sort or classify reports, while CalMatters used AI to make legislative hearings, commission meetings, and tip-sheet material more searchable and usable. The Washington Post's visual forensics team described using AI for tasks such as identifying armored vehicles, geolocating imagery, or counting people at scale, while still insisting that the editorial judgment and evidentiary burden remain human. Several contributors drew a sharp line between assistive workflow use and automated article generation, making the piece a practical snapshot of where newsrooms were willing to operationalize AI and where they resisted it.