---
title: Repurposing Content with AI Agents
description: A hands-on tutorial teaching patterns for working with AI coding assistants — research, delegation, parallelism, and semantic enrichment. The email pipeline is the example; the patterns are the point.
---

# Repurposing Content with AI Agents

This course teaches you how to build a content processing pipeline with AI agents. You'll convert 218 email newsletters into structured, searchable markdown with rich metadata. But the interesting part isn't _what_ you build — it's _how_ you build it.

Instead of writing scripts from scratch, you teach AI agents to research solutions, write other agents, and process data at scale. The result is a pipeline that extracts content, downloads images, and applies rich taxonomy metadata — built through conversation rather than upfront specification.

## The Four Phases

### [Phase 1: Setting Up](/content-repurpose/01-research/)

Install the tools and create a research agent that can learn about technologies for you.

The key insight: instead of manually writing agent instructions, you describe what you want and let the AI write the agent definition. This "prompts to build prompts" pattern means you work at a higher level of abstraction.

### [Phase 2: Cleaning and Organizing the Data](/content-repurpose/02-extraction/)

Ask the research agent how to extract EML files to markdown. It evaluates options, considers your environment, and documents its findings. Then build the converter based on that research. The agent discovers that plain text email content is already markdown-formatted — a simplification you wouldn't have found by jumping straight to code.

### [Phase 3: Decorating the Data with Subagents](/content-repurpose/03-classification/)

Analyze the converted content to build a taxonomy schema. Create a classification agent that applies that schema to each file. Run it in parallel batches: "classify the next 50 emails" spawns 50 concurrent agents. What would take hours manually completes in minutes.

### [Phase 4: Cleanup and Validation](/content-repurpose/04-validation/)

Validation scripts catch inconsistencies across 200+ files. Fix edge cases, refine the schema, iterate until the pipeline produces clean output.

## The Patterns (Not Just the Pipeline)

This course is about **patterns** that apply far beyond email processing:

| Pattern                    | What You Learn                                                                                                                                     | Real-World Application                                                                              |
| -------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------- | --------------------------------------------------------------------------------------------------- |
| **Context-Aware Research** | Ask the LLM how to solve problems within _your_ project. It sees your existing stack and gives tailored recommendations.                           | "How should we setup Python?" becomes "extend your existing mise.toml" rather than "install pyenv." |
| **Agents Writing Agents**  | Describe what you want in plain language, let AI write the agent instructions. You stay at the "what" level while it handles the "how."            | Build specialized subagents without writing prompt engineering from scratch.                        |
| **Compounding Knowledge**  | Research agents inform specialized agents. A tech-research-advisor produces reports that become context for building an email-taxonomy-classifier. | Each layer builds on the last — cumulative intelligence.                                            |
| **Parallel Execution**     | Process batches with concurrent agents. Classifying 50 emails spawns 50 parallel workers.                                                          | Scale agent workloads linearly by adding more concurrent workers.                                   |
| **Semantic Enrichment**    | LLMs extract meaning that would be tedious manually: topics, mentioned people, companies, AI models, audience level, content depth.                | Any unstructured content becomes queryable and cross-linked.                                        |

## What You'll End Up With

```
content/
├── uncategorized/   # 218 markdown files (raw extraction)
├── decorated/       # 210 classified files (with taxonomy)
└── images/          # 710 downloaded images

reports/             # AI research findings
.claude/agents/      # Custom agent definitions
magazine/            # Generated magazine website
```

Each decorated file has rich frontmatter:

```yaml
---
title: "Golden Age for Indie Devs and Engineers"
date: 2024-08-26
series: fod
episode: 64
content_type: digest
primary_topic: industry
tags: [indie-dev, ai-tools, entrepreneurship]
people_mentioned:
  - name: "Andrej Karpathy"
    role: "AI Researcher"
companies_mentioned: [OpenAI, Anthropic]
audience: technical
depth: overview
---
```

## Start Here

- **[About This Course](/content-repurpose/about/)** — Prerequisites, tools, and what you need to get started
- **[Phase 1: Research & Setup](/content-repurpose/01-research/)** — Install tools, create your first research agent
- **[Phase 2: Extraction](/content-repurpose/02-extraction/)** — Convert EML emails to clean markdown
- **[Phase 3: Classification](/content-repurpose/03-classification/)** — Build a taxonomy and classify at scale
- **[Phase 4: Validation](/content-repurpose/04-validation/)** — Scripts, edge cases, and iteration
- **[Project Ideas](/content-repurpose/projects/)** — Ways to apply these patterns to your own work