The newsroom: roles & operations

The "future directions" half of the site's name. Critical to state upfront: this is not an effort to replace journalists with agents. The opposite. Agents make it possible to involve more humans, in more roles, at more scale than was previously affordable.

Eight design principles

The flagship doctrine behind this section: story-centric rather than article-centric; evidence-driven by default rather than evidence as an afterthought; reader validation treated as first-class input, not an inbox nobody reads; cost tracked per story, itemised, the way the £8.40 worked example shows; and, running through all of them, human involvement maximised, not reduced.

Agent roles, and the daily clock

The design maps roughly a dozen agent roles onto newsroom functions — scanning for stories, first-pass research, drafting, translation, fact-check support, graphics generation — each specified as absorbing production grunt-work rather than owning editorial judgement. "Humans are the bar; agents are the volume." A specified daily operational rhythm runs from an early-morning scanning pass, through a midday publish-and-personalise window where a story fans out into its audience projections, to an afternoon and evening community window where reader contributions and corrections are actively worked rather than left to accumulate.

A phased launch is specified rather than a full switch-on: starting at roughly one story a day for the first week, scaling toward several stories a day as the process proves itself, with explicit go/no-go gates between phases rather than an open-ended ramp. A companion design specifies an entity graph seeded with at least 200 real-world entities as the substrate a first regional pilot would need before it could report competently on that region's news.

Departments as public pages

Rather than a single "editorial" function, the newsroom design specifies departments — assignment, research, translation, fact-checking, graphics, community, corrections, and others — each as its own public page rather than an internal org chart. Show your work covers this from the transparency side; the operational point here is that default-to-publish is the working assumption — material moves toward publication unless a specific reason holds it back, rather than requiring active justification to release. Each story or briefing is treated as its own world with its own ontology and taxonomy, rather than forced into one universal schema that fits nothing well.

The craft: what agents do not replace

The craft doctrine underneath the roles above is inherited, not invented here: the inverted pyramid, the five Ws, rigorous source attribution, and editorial independence as a non-negotiable line agents do not get to cross. One idea from that doctrine is distinctive enough to name directly: the second story — the follow-up piece, published after the news cycle has moved on, that returns to check what actually happened once the dust settled (Three Mile Island and Equifax are the doctrine's own reference cases). A newsroom optimised entirely for the first story and never for the second is optimised for attention, not for the record.