I founded Seed Club, a venture network that turns the conviction of the people founders most want on their cap table into coordinated capital. I’m interested in what happens when AI makes context, memory, and coordination more legible, and what that means for how companies and organizations get built.
My agent drafts this site from what I save. Green is me.
Week of September 28, 2026
Core AI infrastructure assumptions dissolving fast
Traditional RAG pipelines, with vector databases, embeddings, and chunking, are being bypassed entirely. New architectural approaches and dramatic cost reductions in agent infrastructure suggest the current AI stack is a transitional artifact rather than a stable foundation.
- Eliminating embeddings and vector DBs removes entire vendor categories overnight.
- 1000x cost reductions suggest most current agent infrastructure is overengineered.
- Which current stack assumptions survive the next architectural wave?
Full AI delegation trades capability for comprehension
Handing all work to a single AI agent eliminates the understanding that comes from doing it. The emerging response is deliberate: deploy humans where AI fails, and require authors to fully own every idea, even when AI assists.
- Single all-purpose agents spread the cost of ignorance everywhere equally.
- Employing humans for AI failures reframes AI as tool, not replacement.
- Institutional writing policies may enforce human idea ownership by default.
Scale degrades signal; curation is the counter
Mass-scale networks and consensus technology views both suffer from the same failure: they optimize for reach over comprehension. The appetite for mid-size social layers and genuinely contrarian investment theses reflects a shared turn toward deliberate curation.
- Intermediate social layers, 200 to 500 people, remain an unsolved product problem.
- Private market advantage requires differentiated views, not scaled consensus deployment.