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
Agent architecture moving from personal to shared systems
Personal AI agents are now considered a baseline, not a destination. The next layer involves shared, specialized agent systems that enable gains from trade, event-driven triggers, and collaborative human-agent workspaces rather than single-user tools.
- MCP Events could replace cron and manual prompts as agent wake-up mechanisms.
- Sandboxed VMs outperform narrow browser tools as full agent infrastructure.
- Whoever controls supply chains may ultimately control personal agent markets.
Shared AI agent systems called the next leap beyond personal agents · @nbaschez MCP Events could enable true event-driven triggers for AI agents · @nbaschez Event-driven triggers free agents from cron jobs and manual prompts · @nbaschez Shared agent factories unlock specialization beyond personal AI tools · @nbaschez Sandboxed VMs beat narrow browser tools as full agent infrastructure · @AniC_dev AI agent market winners may depend on who controls supply chains · @royrubin05 Shared browser desktops let agents take on whole projects · '@VesenceAI' AI agents need owned supply so no marketplace can hold them hostage · @royrubin05 New agent-native desktop lets humans and AI share browser workspaces · @VesenceAI Personal AI agents could unite Google and Facebook in one product · @signulll
AI floods drug pipelines, shifting value to physical testing
AI is generating compounds faster than physical infrastructure can validate them, making contract research organizations the near-term picks-and-shovels winners. The bottleneck is moving from discovery to wet-lab throughput, not from ideation to molecule.
- AI-designed drugs are already reaching human trials.
- The key open question is whether better discovery yields fewer expensive clinical failures.
- Infrastructure, compute, and energy constraints draw parallel investor attention.
CROs seen as near-term winners as AI drug discovery scales up · @bubbleboi AI floods drug pipelines, shifting value to physical lab testing · @bubbleboi AI, biotech, and infrastructure top investor megatrend list · @AviFelman Investors are grouping AI with biotech, energy, compute and defense · '@AviFelman'
Startup funding logic is bifurcating into two extreme camps
The venture conversation is converging on a binary view: capital-light businesses should bootstrap, while moonshot or winner-take-most ventures should raise aggressively. The broad middle ground of moderately funded growth startups is losing its advocates.
- Business type, not founder ideology, should determine the funding approach.
- Anthropic chasing a two-trillion-dollar valuation is making junior VC roles look unattractive.
- Legibility to capital is a practical skill technologists routinely undervalue.
Technologists underestimate the importance of being legible to capital · @rargulati Being understandable to capital is not cosmetic · @rargulati Venture dollars may shift from saturated SaaS toward sci-fi moonshots · @AlanaDLevin Business type, not founder dogma, should drive startup funding choices · @victoriagamlen Startups now fall into two camps: lifestyle or sci-fi moonshot · @mikekarnj Fund raising should match the actual shape of the business · @victoriagamlen Moritz says young VCs should find another career as AI valuations soar · @CEOinterview
AI models are weak moats, adoption curves may be overstated
Models are framed as the fastest-depreciating assets in the stack, making them poor investment theses on their own. Skeptics draw a direct parallel to the 2021 blockchain cycle, where widespread awareness failed to translate into durable, paid adoption.
- Awareness and actual paid use are two very different adoption metrics.
- Agent loop primitives are becoming table stakes, not differentiation.
- Personal AI assistants already cost three thousand to seven thousand dollars per user annually.
AI models called fastest-depreciating assets with few defensible moats · @scottastevenson AI adoption may be overhyped, mirroring blockchain bubble of 2021 · @zeroxkyle AI adoption looks inflated when awareness outpaces actual paid use · @zeroxkyle AI models depreciate fast and make poor startup moats · @scottastevenson AI agent features are becoming table stakes across products · @a16z Shared agent loop patterns are just the new AI table stakes · '@a16z' Personal AI assistants cost up to $7k per user per year · @illscience