Build a Content Pipeline with AI Agents: Research, Write, Illustrate

Content creation has a bottleneck, and it is not writing. Most teams can write fast enough. The bottleneck is everything around the writing: researching the topic, finding data to support claims, creating visuals, and formatting for distribution. Each step involves a different tool, a different interface, and often a different person.

An AI agent with access to the right tools can handle the full pipeline. Not just the writing, but the research that informs it, the images that illustrate it, and the distribution that gets it in front of people. One conversation, one pipeline, multiple outputs.

What the Pipeline Looks Like

A typical content pipeline has four stages:

  1. Research the topic using web search and news
  2. Write the draft using research as source material
  3. Illustrate the piece with generated images or found visuals
  4. Distribute the finished piece via email or other channels

Most people use different tools for each stage. Google for research, a text editor for writing, Midjourney or DALL-E for images, and Mailchimp or Resend for distribution. The agent approach collapses these into a single session.

Step 1: Research

Good content starts with good research. The agent needs current, factual information to produce writing that is useful rather than generic.

“Research the current state of remote work in 2026. Pull recent news articles, Google Trends data for ‘remote work’ and ‘return to office,’ and any relevant statistics from web search results.”

The agent calls web search and news tools to gather source material. It returns a structured research brief: key statistics, recent trends, notable company policies, and expert opinions. This brief becomes the foundation for the draft.

Research through an agent has an advantage over manual research: the agent can search multiple queries in parallel and synthesize results across sources. A human might search three queries. An agent can search ten and merge the findings.

What Good Agent Research Looks Like

The research output should include:

  • Key statistics with sources (e.g., “37% of US workers are fully remote as of Q1 2026, per Bureau of Labor Statistics”)
  • Recent developments from news coverage (company announcements, policy changes)
  • Trend direction from Google Trends (growing, declining, stable)
  • Contrasting viewpoints to make the content balanced

This is not the agent making things up. Every data point comes from an actual tool call that hit an actual data source.

Step 2: Write

With research in hand, the agent drafts the content. The quality of this step depends entirely on the quality of the research.

“Using the research you just gathered, write a 1000-word blog post about the state of remote work in 2026. Use the statistics and sources from the research. Write in a direct, factual style. Include section headers.”

The agent produces a draft grounded in the research it just collected. Because the source material is fresh (from live web and news searches), the content reflects current reality rather than stale training data.

You can iterate on the draft in the same conversation:

“Make the intro more specific. Lead with the BLS statistic instead of a general statement.”

“Add a section about hybrid work models. Search for recent data on hybrid work if you need more material.”

That second prompt triggers another round of research, which the agent folds into the revised draft. The research and writing steps become iterative, not sequential.

Step 3: Illustrate

Text-only content underperforms content with visuals. Blog posts with images get more engagement. Social posts with images get more clicks. Newsletters with images get more opens.

The agent can generate images for the content it just wrote:

“Generate a header image for this blog post. It should show a modern home office setup with a laptop, dual monitors, and a city view through a window. Clean, professional style.”

The agent calls an image generation tool and returns the image. You can request multiple variations:

“Generate three image options. One realistic, one illustration style, one minimal/abstract.”

For data-heavy content, the agent can also create charts or infographics by describing the visual to the image generation tool, though purpose-built charting tools will produce more precise results for complex data.

Matching Images to Content

The key advantage here is context. The agent wrote the content, so it knows what the piece is about, what the key themes are, and what kind of visual would reinforce the message. A standalone image generation tool requires you to write the prompt from scratch. The agent already has the context.

Step 4: Distribute

The final step is getting the content out. For email distribution, the agent can send the finished piece directly:

“Send this blog post as an email newsletter to our mailing list at [email protected]. Use the header image we generated. Subject line: ‘Remote Work in 2026: What the Data Shows.’”

The agent calls the email tool, formats the content for email, attaches the image, and sends it. For teams that publish via CMS, the agent can format the content as markdown with frontmatter, ready to paste into your publishing workflow.

You can also ask the agent to create derivative content for different channels:

“Write a LinkedIn post summarizing this article in 200 words. Write three tweet-length summaries highlighting different sections.”

Same research, same writing, multiple formats. One session.

Tools Involved

This pipeline uses tools from AgentPatch:

  • google-search for web research and finding source material
  • google-news for recent coverage and current events
  • google-trends for trend data and demand signals
  • generate-image for creating header images, illustrations, and visuals
  • send-email for distributing the finished content via email

Each tool is called through a single API. No separate accounts, no switching between platforms.

Setup

Connect AgentPatch to your AI agent:

The AgentPatch CLI is designed for AI agents to use via shell access. Install it, and your agent can discover and invoke any tool on the marketplace.

Install (zero dependencies, Python 3.10+):

pip install agentpatch

Set your API key:

export AGENTPATCH_API_KEY=your_api_key

Example commands your agent will use:

ap search "web search"
ap run google-search --input '{"query": "test"}'

Get your API key from the AgentPatch dashboard.

Claude Code

Install the AgentPatch skill — it teaches Claude Code when to use AgentPatch and how to use the CLI:

/plugin marketplace add fullthom/agentpatch-claude-skill
/plugin install agentpatch@agentpatch

MCP Server (Alternative)

If you prefer raw MCP tool access instead of the skill:

claude mcp add -s user --transport http agentpatch https://agentpatch.ai/mcp \
  --header "Authorization: Bearer YOUR_API_KEY"

Replace YOUR_API_KEY with your actual key from the AgentPatch dashboard.

OpenClaw

Install the AgentPatch skill from ClawHub — it teaches OpenClaw when to use AgentPatch and how to use the CLI:

clawhub install agentpatch

MCP Server (Alternative)

If you prefer raw MCP tool access instead of the skill, add AgentPatch to ~/.openclaw/openclaw.json:

{
  "mcp": {
    "servers": {
      "agentpatch": {
        "transport": "streamable-http",
        "url": "https://agentpatch.ai/mcp",
        "headers": {
          "Authorization": "Bearer YOUR_API_KEY"
        }
      }
    }
  }
}

Replace YOUR_API_KEY with your actual key from the AgentPatch dashboard. Restart OpenClaw and it discovers all AgentPatch tools automatically.

Wrapping Up

The content pipeline bottleneck is coordination, not capability. Research, writing, illustration, and distribution are all solvable problems. The friction is moving between tools, copying information from one interface to another, and keeping context consistent across stages. An AI agent with access to search, image generation, and email tools collapses those stages into a single conversation. Visit agentpatch.ai to connect the tools and build your content pipeline.

← Back to blog