How I Built an AI Research Machine That Never Sleeps
JH
Joachim HøgbyFrom Manual Research to Autopilot
I used to start my day by reading newsletters, scrolling YouTube, and manually copying notes into Obsidian. It took hours. Now one Claude Code skill does everything in seconds.
The Problem Everyone Has
Research is time-consuming because it's fragmented. You search on YouTube, find a good video, watch it, take notes, organize the notes, and repeat. Every time you switch tools, you lose your flow.
The Solution: One Skill, Four Tools
I built a Claude Code skill that orchestrates four tools in a pipeline:
- YouTube search – Claude Code searches for relevant content based on a topic I give it
- Transcription – The video is automatically transcribed
- NotebookLM – The transcript is sent to Google's NotebookLM for deep analysis and summarization
- Obsidian – The result is stored structured in my Obsidian vault with proper tags and links
What Happens in Practice
I write: "Research latest developments in agentic AI for enterprise"
Claude Code:
- Finds the 5 most relevant YouTube videos from the last week
- Transcribes all five
- Sends each transcript to NotebookLM for analysis
- Generates a summary with key insights, quotes, and sources
- Saves everything in Obsidian under the right folder with backlinks
The entire process takes under 3 minutes. Manually, it would have taken half a day.
📬 Likte du denne?
AI-nyheter for ledere. Kuratert av en CIO som bygger det selv. Daglig i innboksen.
