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
Document-native retrieval threatens to replace vector search
PageIndex, a new retrieval method, discards vector databases, embeddings, chunking, and similarity search in favor of reasoning over full documents, challenging the foundational assumptions the RAG industry is built on.
- The shift moves retrieval from statistical similarity matching to document-level reasoning.
Personal AI agents become viable through architecture, not capability
Zero-idle infrastructure and orchestration layers are cutting personal AI costs by orders of magnitude, while hybrid human routing fills reliability gaps where models still fail, making personal agents economically viable without waiting for model improvements.
- Orchestrators that ingest Slack and GitHub context coordinate specialized agent teams.
- VM-per-agent architectures are being replaced by virtual filesystems on cloud workers.
- Hybrid human routing beats pure-AI agents on real-world task completion rates.
How to build a personal AI software factory with agent orchestration · @ianlapham New personal AI startup deploys humans for tasks AI cannot complete · @shivanipod New approach cuts AI personal assistant costs by 1,000x · @dctanner Zero-idle architecture makes personal AI agents economically viable · @dctanner Some personal AI products will win by routing tasks to humans · @shivanipod
Outsourcing cognition trades efficiency for understanding
A single general-purpose AI agent handles more tasks but distributes cognitive cost equally across everything, leaving no way to choose what to understand. Domain-specific agents let you calibrate the tradeoff domain by domain.
- Writing policies that require full authorship protect thinking, not just output quality.
- Domain agents make the understanding sacrifice explicit and adjustable.
One chief-of-staff AI agent means losing your own understanding · @alexatallah Clay rolls out AI writing policy requiring full author ownership · @vxanand AI writing policies should protect thinking, not just polish · @vxanand Domain agents let you calibrate how much understanding to outsource · @alexatallah
AI capabilities far outpace actual mass adoption
A large gap persists between what AI can do and how most people use it, and the historical pattern of category-defining companies starting niche suggests this gap is where the next generation of platforms will be built.
- Investors overweight grand TAM narratives over specific niche entry points.
- Amazon, Google, and others started as narrow tools before defining new categories.