· 6 min read
By the end of this guide, you will have a 5-agent content pipeline that takes a topic brief, researches it, produces a structured draft, edits it for clarity and tone, optimizes it for search engines, and delivers a publish-ready article. The pipeline handles the entire content lifecycle from ideation to final copy, producing work in 15-20 minutes that would take a human content team several hours.
The team uses a Sequential Pipeline pattern with five agents: a Research Scout that gathers background material and competitive context, a Content Strategist that builds an outline and angles the piece, a Draft Writer that produces the full article, an Editor that refines prose and enforces style guidelines, and an SEO Specialist that handles metadata, internal linking, and keyword placement. Each agent hands its output to the next, with structured checkpoints between stages.
You need Claude Code (Anthropic's CLI tool) or the Claude Agent SDK for programmatic pipelines. You also need a content brief -- a topic, target audience, primary keyword, and desired word count. Vague briefs produce vague articles. "Write about AI" gives your agents nothing to work with. "Write a 1,500-word guide for marketing directors on using AI agent teams to automate competitive analysis, targeting the keyword 'AI competitive analysis automation'" gives every agent in the pipeline a clear objective.
If you plan to run this pipeline on a recurring schedule, the Agent SDK is the better choice -- it lets you trigger runs programmatically and integrate with your CMS.
Mission: Given a topic brief, gather background material including competitor content on the same topic, relevant statistics, expert perspectives, and gaps in existing coverage that represent opportunities for differentiation.
The Research Scout does not write -- it collects raw material. Its output is a structured research brief containing: 5-8 key facts or statistics, 3-5 competitor articles with summaries of their angles, 2-3 underexplored angles that competitors have missed, and a list of potential sources to reference.
Prompt guidance: Instruct this agent to prioritize recency and specificity. A statistic from 2024 is more useful than one from 2021. A case study from a named company is more useful than a generic claim. Ask it to flag where data is thin so downstream agents know where to be cautious with claims.
Mission: Using the research brief, build a detailed content outline that defines the article's angle, structure, key arguments, and how it will differentiate from existing coverage.
The Strategist decides what the article is actually about. It picks the angle that best serves the target audience, sequences the sections for logical flow, and assigns a purpose to each section (introduce concept, provide evidence, address objection, call to action). It also identifies where specific research findings should appear in the article.
Prompt guidance: Give the Strategist your brand's editorial guidelines if you have them. Specify the reading level, whether the tone should be authoritative or conversational, and any structural requirements (must include a how-to section, must address common objections, must end with next steps).
Mission: Produce a complete first draft following the Content Strategist's outline, incorporating research from the Scout, and hitting the target word count.
The Draft Writer focuses on getting ideas on the page with proper structure and flow. It follows the outline section by section, weaves in data points and examples from the research brief, and maintains consistent voice throughout. It does not self-edit -- that wastes its context window on a task better handled by a dedicated agent.
Prompt guidance: Specify the exact voice and tone. "Write like a knowledgeable colleague explaining something over coffee -- confident but not academic, specific but not jargon-heavy." Include examples of sentences in your preferred style if possible.
Mission: Refine the draft for clarity, concision, tone consistency, logical flow, and factual accuracy. Cut filler. Tighten prose. Flag any claims that lack supporting evidence from the research brief.
The Editor is adversarial by design. Its job is to make the draft shorter and sharper. It removes hedging language ("it could potentially maybe help"), eliminates redundancy between sections, strengthens transitions, and ensures every paragraph earns its place.
Prompt guidance: Give the Editor specific style rules: maximum sentence length, banned phrases ("leverage," "utilize," "in today's fast-paced world"), required formatting conventions (H2 for major sections, H3 for subsections, bullet lists for 3+ items).
Mission: Optimize the edited article for search performance without degrading readability. Handle title tag, meta description, header hierarchy, keyword placement, and internal linking suggestions.
The SEO Specialist works on the finished article, not the draft. This sequencing is intentional -- optimizing a rough draft is wasted effort since the Editor will rewrite significant portions. The Specialist ensures the primary keyword appears in the title, first paragraph, at least two H2 headers, and the meta description. It suggests 3-5 internal links and recommends alt text for any images.
Prompt guidance: Provide the primary keyword, 2-3 secondary keywords, and your site's URL structure so the agent can suggest valid internal links. Specify your meta description character limit (typically 150-160).
Content production is inherently sequential -- you cannot edit a draft that has not been written, and you cannot write a draft without research. Define the handoff protocol between each stage:
Each handoff should include explicit quality criteria. The Editor, for example, should not pass a piece forward if it exceeds the target word count by more than 10% or if it contains unsupported claims.
Each prompt needs four components for this pipeline to work reliably:
Identity and constraints. "You are a content editor specializing in B2B technology content. You do not add new information -- you only refine what exists."
Input specification. "You will receive: (1) a draft article in markdown, (2) the original content brief, (3) a style guide."
Task instructions. "Read the draft against the brief. Cut any section that does not serve the target audience's stated need. Tighten sentences to under 25 words where possible. Remove all instances of passive voice unless the active alternative is awkward."
Output specification. "Return: (1) the edited article in markdown, (2) a changelog listing every significant edit and the reason for it, (3) a quality score from 1-10 with justification."
Start with a single article to calibrate the pipeline. Common first-run issues:
The Research Scout returns too much generic information and not enough specific data. Fix by adding constraints: "Return exactly 5 statistics with publication dates newer than 2025" and "Identify 3 specific competitor articles by URL or title."
The Draft Writer ignores parts of the outline. Fix by numbering outline sections and requiring the Writer to confirm each section is addressed with a checklist appended to its output.
The Editor and Writer have conflicting style preferences. Fix by giving both agents the identical style guide document so their conception of "good writing" is aligned.
A finished article run through this pipeline includes:
The article should read as though a human writer produced it with editorial support -- coherent argument, specific evidence, clean prose, and thoughtful structure.
Batch processing. Once your pipeline is stable, feed it a queue of topic briefs. The Agent SDK lets you run multiple pipeline instances in parallel, producing 5-10 articles per batch.
Add a Fact Checker. Insert a verification agent between the Editor and SEO Specialist. This agent cross-references every factual claim against the original research brief and flags anything that was embellished or distorted during writing and editing.
Tone variants. Run the same research and outline through multiple Draft Writers with different tone instructions to produce versions for different channels -- a formal version for your blog, a conversational version for LinkedIn, a concise version for email newsletters.
Feedback loops. After publishing, feed performance data (time on page, bounce rate, search ranking) back into the Content Strategist's prompt as examples of what works. Over time, the Strategist learns which angles and structures perform best for your audience.