My reading window is one hour, after the toddler is finally asleep. That hour is too short to waste — so I taught my AI what to ignore. The last thing I want is to doom-click through that hour. (If scrolling can be doom, so can clicking, right?)
THE DAILY SCAN
I built a daily scan: Claude runs the web searches, Gemini turns YouTube podcasts into markdown summaries. Instead of opening a dozen tabs to figure out what to read, watch or listen to, I get one page, filtered, ready every evening.
THE INTERESTING PART IS WHAT IT IGNORES
It runs across six themes: AI labs, cloud hyperscalers, IT-services peers, dev tools and agent frameworks, enterprise adoption, and a practitioner pulse (Simon Willison, Karpathy, Hamel, Latent Space, etc.). Inside each theme, every story is graded by origin: primary first (the paper, the release notes, the filing), then reporting on it, then opinion about it. Only the last 48 hours count.
THEN THE FILTER DOES THE REAL WORK
The exclusion list is longer than the inclusion list:
- Political opinion, influencer drama, stock speculation.
- Vendor PR dressed as news.
- "X vs Y in 2026" listicles with no new data.
- Anything from sources I've rejected more than twice in the last 30 days.
It learns from its own misses. What lands in my Obsidian inbox is 20 to 30 items, at least 40% from primary sources. I approve or reject each one with feedback. Every rejection feeds the learnings, so the AI gets better at predicting what I won't read or watch.
The AI does the gathering. I stay current and still get some sleep.
What would you teach your AI to ignore?