Skip to content
Case 05 · Automated content development

The writers’ room where “probably true” gets sent back.

Good writing takes research, structure, a real voice and careful editing, and most teams can’t do all of that at the pace they need to these days. This began as a sketch of a writers’ room that never sleeps: someone to research, outline, draft, critique and fact-check, with writers and editors going back and forth at every step that matters. Today it turns a list of ideas into finished pieces for a few dollars each.

the idea: idea list pick a writer research draft + check a person says yes ready …one piece at a time, in each writer’s own voice
the first sketch↓ what it became
1Tuesday

A messy list becomes clean assignments.

The team’s idea list is a real-world spreadsheet: headers in the wrong row, initials instead of names, half-finished thoughts. The system reads it anyway and proposes clean assignments, each with a note on how sure it is.

Designed forStarting from what teams already have. Today, a person picks ideas from the trend report in The Signals (Case 04). Next, the two run as one system, and the team can still add its own ideas.
wk?? reset wknd?! RB · grp trips $ quiet towns (ask M) messy is fine
From the weekly trends · sureThe one-night reset
From the weekly trends · sureGroup trips, no money fights
Team idea · check ownerOur hosts’ quietest towns
2Tuesday

It picks the right writer.

Every writer on the roster has a voice profile built from their own best work. Rosa writes warm, first-person and practical, and she wrote the last two short-trip guides, so “The one-night reset” goes to her. The editor can always override.

Designed forEvery writer sounding like themselves. Profiles are built from real samples and approved by an editor before they’re used.
Writer

Rosa

Warm, first-person, practical. Opens with a scene.

Wrote the last two short-trip guides

Writer

Jules

Wry, data-first, short sentences.

Covers money and planning

best fit for this piece (the editor can override) great writer, wrong piece
3Tuesday

Research, an outline, and a person says yes.

It researches the topic with real sources, then sketches an outline the way Rosa would build it. Before a single paragraph is written, an editor says yes, or asks for a different angle.

Designed forPeople steering where it matters. Two approval points, the outline and the final piece, and nothing goes out without both.
Outline · The one-night reset1. The Friday feeling [1]2. Why one night is enough [2]3. How to pick the place4. One thing to try this month

Sources: two studies and one survey, all linked

approved a person signs off before any drafting
4Tuesday

“Probably true” isn’t good enough.

The draft says most travellers now prefer one-night trips. No source, so it goes back. Another line turned “some evidence suggests” into “research shows.” That’s a hedge made stronger than the evidence, so it goes back too.

Designed forFacts you can stand behind. Unsourced or overstated claims are sent back automatically, and no editor can wave them through.
Draft · Rosa

Most travellers now prefer one-night trips.

Some surveys suggest short stays leave people more rested. [2]

Research shows a single night away resets your sleep.

source? was “some evidence suggests” sent back
5Wednesday

A lesson sticks, and the piece ships.

The editor gives Rosa a note: open with the scene, not the statistic. The system learns from it, so every piece she writes from here on starts that way. The final piece is approved: 540 words, two revision rounds, about four dollars of machine time.

Designed forGetting better per writer. It learns from feedback, so a note only has to be given once.
Rosa: open with the scene, not the statistic. (one note → now a rule) every piece after this remembers
Final · The one-night reset

Approved

540 words · 2 revision rounds

ready to publish

What it took to make it hold up.

Writing one good piece is craft. Writing every piece in the right voice, with every claim sourced, without a single unsourced claim slipping through, is a system.

The real system: seven specialist AI roles · two human approvals per piece · 400–600 words · at most three revision rounds · about $2–7 per finished piece

A

Accurate

Every claim needs a source, and a hedge stays a hedge. Anything unsourced or overstated goes back for revision automatically.

R

Robust

It works from the messy lists teams already keep, caps revisions at three rounds, and hands anything stuck to a person.

L

Learns as it goes

It learns from feedback, so the same edit isn’t made over and over.

Editors stay in charge.

  • Editors approve every writer profile, every outline and every final piece.
  • Nothing publishes automatically.
  • The cost of every piece is visible, down to each step.

Same idea, other places.

It fits anywhere writing has to be accurate, on-voice and steady, at a pace people alone can’t keep.

  • Newsletters and internal communications
  • Product and help documentation
  • Research summaries and reports
  • Proposals in a firm’s house voice