Four sources. Everything else is too expensive in opportunity cost.

The Four Sources

1. X (Twitter) — Following the actual practitioners, researchers, and founders who are closest to what's happening. Not journalists writing about them. The primary sources.

2. Conversations with top practitioners — The people doing the most interesting work in a given field. Not conferences, not podcasts with editors — direct conversation.

3. Leading AI models — Using frontier AI to synthesize, research, and think through problems. Treating the models themselves as information sources, not just tools.

4. Old books — Not recent books. Old books. The ones that have survived decades of filtering. If a book is still being read 50 years after it was written, it probably contains something durable.

What He's Explicitly Not Doing

The implicit argument in the list: anything not on it carries more noise than signal.

That means most news coverage, most podcasts, most newsletters, most industry reports, and most social content outside of X.

The opportunity cost argument is real. Every hour spent on low-signal information is an hour not spent on high-signal information. The challenge is that low-signal information often feels productive — you're staying current, staying informed, staying connected.

But current ≠ informed. A lot of "staying current" is consuming processed versions of what the primary sources already said, with added distortion.

The AI Models Point Is Underrated

Most discussions of this list focus on X and the books. The AI models point is more interesting.

Using frontier models as research and thinking partners — not just query-response machines — is a significantly different posture toward AI than most people have adopted. It treats the model as a collaborator in sense-making, not a search engine with better phrasing.

For solo founders, this is the most actionable item on the list. You probably can't call Marc Andreessen. You can use the same AI models he uses.

The list is simple. Simple doesn't mean easy. Most people will keep reading the things that feel informative but aren't. The ones who enforce the constraints Andreessen describes will compound differently.