Claude Cowork Automation: From Idea to Working System

From Manual to Automated: Building AI Agent Workflows Step by Step

My exact process for turning manual tasks into automated AI workflows. I build a lead generation system using MCP servers and Claude Code - searching YouTube videos, extracting
caption data, and scoring leads based on localization signals.

The key insight: start manual, validate the process, then progressively automate. Works for any automation you want to build.

What you'll learn:
Building custom MCP tools for YouTube data
Lead scoring based on subtitle/language signals
Generating system prompts from manual workflows
Semi-automation with Claude Cowork
Iterating prompts based on feedback

---
Chapters:

0:00 - My automation process
0:08 - MCP tool: searching YouTube videos
0:40 - Extracting video details & caption tracks
1:01 - Lead scoring logic
1:51 - Generating the lead report
2:56 - Writing a system prompt from manual work
3:47 - Dynamic vs static inputs
4:30 - Semi-automated workflow in Cowork
5:45 - Iterating on prompt issues
7:17 - Next steps: full automation

🛣️ Sharing my learnings 👇
https://stefan.computer/bird
https://stefan.computer/watch

🔗 What I'm cooking 👨‍🍳👇
https://gutsy.so/
https://www.unyte.de/