An editor at a wire service used to spend most of a shift working one piece at a time: tightening a sentence, questioning a claim, deciding what belonged in the lede. The Associated Press's own published generative AI standards, when they arrived in 2023, spent comparatively little space on that kind of guidance and much more on something else entirely — what counts as verified, what an AI tool is not permitted to touch unsupervised, and what triggers a second human read rather than a quick approval.
That imbalance is the tell. When drafts can be generated quickly and cheaply, the scarce editorial resource stops being sentence-level polish and becomes the design of the checkpoints a piece has to pass through before it reaches a reader — a systems design problem closer to product management than to a red pen on a printed page.
Newsrooms that treat this as a technology upgrade, bolting a drafting tool onto an otherwise unchanged editorial process, tend to discover the bottleneck simply relocates: verification, not drafting, becomes the constraint, and editors who used to spend their time improving prose now need to spend it designing and auditing the process that decides what gets checked and how.
The old model of editorial authority — one experienced person reading a finished piece and using judgment to approve or reject it — does not scale to a volume of drafts no individual can carefully read one at a time, which is precisely the operating reality a wire service the size of the AP has to plan for explicitly rather than trust to a single reviewer's attention on any given day.
The red pen is not obsolete, but for a growing share of newsrooms it is no longer where the real editorial work happens first. That work now happens earlier, in the design of the verification pipeline a story moves through — and the newsrooms writing that pipeline down explicitly, the way the AP did, will be easier to trust than the ones still pretending a disclosure line does the same job.