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 July 13, 2026
Agents act autonomously without prebuilt features
AI agents are resolving tasks by improvising their own tools rather than waiting for explicit feature support, collapsing the line between product capability and model capability. Small specialized models are also outperforming frontier models on narrow tasks, suggesting capability advantages are becoming domain-specific.
- Agents proactively acquire capabilities the product never shipped.
- Narrow fine-tuned models can beat frontier models at 98% accuracy.
- Personal agents may route around algorithmic feeds entirely.
AI agent autonomously made phone calls to check parts inventory · @tsrkeith Capable agents figure out what they need and get it done · @tsrkeith Intern trains small model to read music with 98% accuracy in 10 seconds · @Austen Small specialized models can beat frontier models in narrow tasks · @Austen Personal AI agents may render social media feeds obsolete · @iamgingertrash Personal agents may render algorithmic social feeds obsolete · '@iamgingertrash'
AI restructures firm economics in favor of small operators
AI tooling is creating a structural advantage for solo and small-team operators who can move faster than large incumbents constrained by existing org design. Research on AI-native firms confirms they are organized differently, not just leaner.
- Large firms talk up AI transformation while protecting their own headcount advantages.
- Broad VC networks outperform deep ones, a gap agents can help close.
- Open question: which org structures will actually capture the n=1 edge?
AI agents help small VC firms most but big firms set the narrative · @credistick AI gives solo GPs a structural edge before large firms can adapt · '@credistick' Broad networks outperform deep ones for VC returns · @credistick VCs gain more from broad networks than from a few deep ties · '@credistick' New paper finds AI-native firms are structured differently than expected · @orgRem AI-native firms are built around systems, not headcount · orgRem
Anthropic loses workflow ROI ground to competitors
A sharp shift in user sentiment signals that ROI on Anthropic's models in production workflows is underperforming expectations, with meaningful switching toward Codex. The break appears sudden, concentrated in roughly one business week.
- The signal is workflow ROI, not model benchmark preference.
- Intelligence getting cheaper does not mean all cost differences collapse.
- Cheap intelligence still has meaningful pricing tiers that affect build decisions.
Users are abandoning Anthropic and switching to Codex over poor ROI · @staysaasy The market signal here is not just model preference but workflow ROI · @staysaasy AI intelligence will be cheaper but never free or infinite · @perrymetzger AI makes intelligence far cheaper but cheaper is not the same as free · '@perrymetzger'
Crypto infrastructure moves toward institutional on-chain finance
DeFi vaults are being reframed by institutional analysts as the on-chain equivalent of managed funds, positioning them as future core infrastructure for tokenized real-world assets rather than crypto-native experiments.
- Paradigm's October conference signals continued institutional convergence around stablecoins and AI.
- Operational efficiency gains from vaults introduce new risk considerations at scale.
- Regulatory and structural legitimacy, not just technical readiness, will gate adoption.