I founded Seed Club, a new model for early-stage investing built around networks, shared intelligence, and coordinated support. 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 April 27, 2026
Agentic infrastructure is becoming the primary value layer
Applications are thinning as the infrastructure managing state, memory, context, and coordination becomes the load-bearing substrate of software. The harness, filesystem, and identity primitives around agents are beginning to matter more than the models themselves.
- Harnesses should evolve like org charts, not be architected once and built upon
- Memory is converging on git-backed files and grep, not proprietary knowledge graphs
- Agent identity may be the generational primitive this computing era has been missing
AI work shifts to a continuous agent runtime that humans and agents share · @danshipper When agents generate workflows the app goes thin and the database becomes runtime · @anshublog Autonomous development splits into reactive, exploratory, and user-spec modes · Peter Zakin The agent harness is an org chart you keep adjusting, not backend built once · Yohei Nakajima AutoGen and agent marketplaces are dead, context engineering compounds · Ronin Agentic infra makes people coalesce into a new kind of organization · Matt Huang Mesa gives agents a durable POSIX filesystem with branches, diffs, and rollback · Oliver Agent Identity is the new cell, ushering in Sovereign Computing · Josh Crumb Future computing splits into ambient intent, agentic execution, and ephemeral review · signüll AI memory converges on git files and grep, knowledge graphs losing · andrew pignanelli
Crypto becomes default plumbing for agent transactions
The dollar cannot subdivide below one cent and requires permissions across jurisdictions, making crypto the structurally necessary settlement layer for autonomous agent transactions. Stripe's recent crypto integration positions this as invisible infrastructure delivered under a trusted consumer brand.
- Buybacks may be the only credible path to token value accrual beyond speculation
- Platforms that own the customer relationship can now recapture interchange via stablecoins
Buybacks are the only credible path to token value accrual · Jeff Dorman Codex adds messaging and payments, pushing coding tools into financial rails · Brian Flynn Stripe ships crypto as agent-native plumbing under a trusted brand · Dan Romero (@dwr) Platforms reclaim bank revenue by issuing their own stablecoins, cards and FX · Lightspark Universal Basic Equity in crypto L1s replaces UBI in the AGI economy · Raoul Pal
AI eats incumbent moats faster than SaaS did
AI compresses defensibility timelines because the strongest traditional moats are absent from most AI products. Token-reselling and apathy-dependent business models face the steepest structural exposure.
- Cool interfaces and system prompts are far weaker moats than SaaS incumbents had
- Fintech profits built on customer inertia are structurally vulnerable to autonomous agents
- Law AI vendors built on token resale face disintermediation as labs go direct
Harvey and Legora are token resellers exposed if labs go direct · @willchen500 AI-mageddon hits harder than SaaS because the moats are weaker · @scottbelsky AI splits firms into hyper-optimized giants and nanocorps · Katie Chiou Fintech profits run on customer apathy, and AI agents will route around all of it · Anish Acharya
Pathological determination beats brilliance at founder selection
Pathological determination, present in under 1% of founders across thousands of evaluations, is the rarest and most predictive success trait. Most investors systematically overweight raw talent while underweighting obsession, which is anti-fragile where brilliance is not.
- Relentlessness without conviction just maintains someone else's order
- Teams weigh as heavily as product ideas in early-stage evaluation
- Box-checking VC frameworks would screen out many eventually category-defining founders
Across 5000 founder meetings, relentless determination was the rarest success trait · @adamshuaib Pathological determination beats brilliance and shows up in under 1% of founders · Adam Shuaib World-changers pair relentlessness with conviction, and either alone is a curse · Joel John Half of Speedrun's startup evaluation weighs the team, not the product · Jon Lai Running out of money rarely kills startups, it just hides the real cause · arian ghashghai Venture's box-checking scale game would have rejected a young Bezos or Musk · Dan Gray