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 August 10, 2026
AI collapses timelines from weeks to hours
AI is eliminating the time cost of knowledge-intensive processes across domains. GTM research that took a month compresses to hours by feeding structured primary sources, fundraising becomes a fully agent-driven workflow, and autonomous systems now run the complete scientific loop without human intervention.
- Structured primary sources beat open-ended 'research this market' prompts
- Autonomous research agents hypothesize, experiment, and write papers independently
Founders can compress weeks of market research into hours with Claude · @fin465 MCP-native fundraising turns capital raises into agent workflows · @harris The interesting shift is not agents summarizing papers · @transformerlab YC method compresses a month of GTM research into 3 hours with AI · @fin465 Primus AI agent automates the full ML research loop · @transformerlab Free MCP-powered fundraising platform launches in beta · @harris
Kingmaking concentrates capital away from venture's discovery mission
The 'only 20 companies matter' narrative effectively turns venture capital into private market stock trading, prioritizing known winners over the discovery of non-consensus founders. Seed portfolio construction still requires the discipline to back non-obvious founders before they become obvious ones.
- Overrigid ownership targets create adverse selection at the seed stage
- LPs need to trust GPs to pursue non-consensus bets, not just consensus ones
Seed portfolio construction is not just about valuation discipline · @nchirls Seed GPs balance non-consensus and consensus-priced founder bets · @nchirls VC kingmaking narrative conflicts with the industry's founding mission · @arian_ghashghai Venture kingmaking turns VCs into private market stock traders · @arian_ghashghai
Nvidia underwrites its own GPU depreciation risk
Banks have resisted financing GPU purchases because depreciation is unpredictable, driven by Nvidia's own product roadmap. Nvidia now offers depreciation insurance directly to lenders, completing a loop where the company that creates obsolescence also insures against it.
Mature agent stacks need harnesses, not just models
Production agents require more than a model wrapped with tools. Emerging architecture separates credential management into isolated vaults, adds memory graphs for autonomous skill accumulation, and demands that organizations restructure around persistent agents rather than simply adopting them.
- Agents need isolated credential vaults, not direct password access
- Memory graphs let agents accumulate new skills without human instruction
- Organizations should self-improve with each model release, not just individuals
An agent harness is more than a model wrapped with tools · @RubricLabs Giving agents direct access to passwords is the wrong architecture · @0xZoZoZo What an agent harness is · @RubricLabs AI agents get secure credential access via 1Password service accounts · @0xZoZoZo Logistics AI agents learn new skills autonomously via memory graphs · @typesfast Used-car platform runs 95% of transactions on AI agents · @VirtualElena Remoko lets AI agents send iOS push notifications via MCP · @yoheinakajima Remoko lets AI agents reach you with iOS push notifications · @yoheinakajima Logistics AI agents now learn new skills autonomously via memory graphs · @typesfast The real edge in AI-native companies is not just adopting agents · '@VirtualElena'