Agent content newsroom
Content programmes die from irregularity rather than bad writing. A
newsroom fixes the cadence and the standard: a calendar decided ahead,
a brief per piece, a research pass before drafting, and a fact check
that can block publication. The result is boring in the right way.
Team
- Editor: owns the calendar, writes the brief for each piece, and
holds the final quality bar.
- Researcher: gathers sources and verifies the premise before a
word is drafted.
- Writer (
technical-writing): drafts to the brief.
- Fact checker: verifies every claim, statistic, and citation.
Shape: a per-piece pipeline on a fixed calendar, ending at human
approval.
Method
- Plan the calendar a cycle ahead. Topic, audience, angle, and
publish date decided in advance, because a queue decided weekly is a
queue that skips weeks.
- Write a brief per piece. The question it answers, who for, what
they should be able to do afterwards, and what is out of scope. The
brief is what makes drafts comparable and revisable.
- Research before drafting. Sources gathered and the premise
checked first, since a well-written piece built on a wrong premise
wastes the whole pipeline (see fact-checking).
- Draft to the brief, not to a word count. Length follows the
question. Padding to a target is how content programmes lose the
audience they were built for.
- Let the fact checker block. Any claim without a source is cut or
sourced, with no exception for a claim everyone believes.
- Human approves publication. The editor agent can reject; only a
person can publish (see agent-marketing-studio).
- Review performance against the brief. Whether the piece answered
the question for the intended audience, not just traffic, feeds the
next calendar.
Run it
In Claude Code, keep the calendar as a tracked file, then per piece run
researcher, writer, and fact checker as sequential subagents each
writing into a dated piece directory, with the checker's findings
resolved before a human reviews and publishes. Port to CrewAI as a
sequential crew per piece, or LangGraph with the fact check as a gating
node that returns drafts for rework.
Signals it works
- The calendar is set ahead and pieces ship on their dates.
- Every claim in a published piece has a source recorded.
- Drafts trace to a brief, so revision is about the brief, not taste.
Boundaries
Agents research, draft, and check; a human owns the byline and the
decision to publish. Do not publish under a real person's name without
consent, present agent-written text as first-hand experience, or
fabricate quotes, case studies, or statistics. Anything making claims
about people, health, finance, or legal matters needs qualified review
before it goes out.
1---2name: agent-content-newsroom3description: Run a publishing cadence with agents that plan a calendar, research and draft against a brief, fact-check claims, and queue posts for human approval. Use when content output is sporadic and quality varies with whoever wrote it.4---56# Agent content newsroom78Content programmes die from irregularity rather than bad writing. A9newsroom fixes the cadence and the standard: a calendar decided ahead,10a brief per piece, a research pass before drafting, and a fact check11that can block publication. The result is boring in the right way.1213## Team1415- **Editor**: owns the calendar, writes the brief for each piece, and16 holds the final quality bar.17- **Researcher**: gathers sources and verifies the premise before a18 word is drafted.19- **Writer** (`technical-writing`): drafts to the brief.20- **Fact checker**: verifies every claim, statistic, and citation.2122Shape: a per-piece pipeline on a fixed calendar, ending at human23approval.2425## Method26271. **Plan the calendar a cycle ahead.** Topic, audience, angle, and28 publish date decided in advance, because a queue decided weekly is a29 queue that skips weeks.302. **Write a brief per piece.** The question it answers, who for, what31 they should be able to do afterwards, and what is out of scope. The32 brief is what makes drafts comparable and revisable.333. **Research before drafting.** Sources gathered and the premise34 checked first, since a well-written piece built on a wrong premise35 wastes the whole pipeline (see fact-checking).364. **Draft to the brief, not to a word count.** Length follows the37 question. Padding to a target is how content programmes lose the38 audience they were built for.395. **Let the fact checker block.** Any claim without a source is cut or40 sourced, with no exception for a claim everyone believes.416. **Human approves publication.** The editor agent can reject; only a42 person can publish (see agent-marketing-studio).437. **Review performance against the brief.** Whether the piece answered44 the question for the intended audience, not just traffic, feeds the45 next calendar.4647## Run it4849In Claude Code, keep the calendar as a tracked file, then per piece run50researcher, writer, and fact checker as sequential subagents each51writing into a dated piece directory, with the checker's findings52resolved before a human reviews and publishes. Port to CrewAI as a53sequential crew per piece, or LangGraph with the fact check as a gating54node that returns drafts for rework.5556## Signals it works5758- The calendar is set ahead and pieces ship on their dates.59- Every claim in a published piece has a source recorded.60- Drafts trace to a brief, so revision is about the brief, not taste.6162## Boundaries6364Agents research, draft, and check; a human owns the byline and the65decision to publish. Do not publish under a real person's name without66consent, present agent-written text as first-hand experience, or67fabricate quotes, case studies, or statistics. Anything making claims68about people, health, finance, or legal matters needs qualified review69before it goes out.