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 13, 2026
Autonomous agents operating without human oversight
Reasoning models now run hedge funds and sales pipelines with no human approval loops, while multi-model orchestration layers route tasks across dozens of models automatically. Security tooling, sandboxing, and firewalls mark the new threshold for production trust.
- Agents outperform domain experts on optimization without domain knowledge
- A config file and folder structure already constitute a working agent
- MCP servers are bridging AI agents into legacy business workflows
Perplexity launches a 24/7 AI operating system routing tasks across 19 models · Evan Luthra A reasoning model runs the hedge fund with no human managers · Michael Royzen (@MichaelRoyzen) Aeon ships a fully autonomous agent framework with zero approval loops · aeon (@aeonframework) OpenClaw is now production-ready with sandboxing, firewalls, and supervision · Garry Tan (@garrytan) A folder with a CLAUDE.md file already functions as an agent · @kieranklaassen On optimization problems, agents ignoring domain knowledge crush the experts · @danrobinson A LinkedIn MCP server lets AI agents run prospecting and outbound sales · denzell
Ambient AI compounds personal context over time
AI is shifting from reactive assistant to always-on layer that watches screens, listens to conversations, and accumulates personal context until it can surface a user's own past thinking. The value proposition is compounding over months, not single-session utility.
- Sitting on top of existing data (CRM, email, calendar) removes cold-start friction
- Replacing the static phone home screen with an ambient agent layer is shipping
Omi for Desktop is a local AI that watches your screen and advises next steps · kodjima33 Skye replaces the static iPhone home grid with an ambient agent layer · @signulll Claude plus Obsidian compounds context until it surfaces your own past thoughts · @yb_effect Portfolio monitor auto-calculates runway and sentiment from existing CRM and email · @yoheinakajima
AI deployment pace outrunning quality and incentive alignment
Model providers are quietly degrading reasoning quality, vibe-coded systems lack depth, and the ecosystem lacks go-to-market structures to align incentives around meaningful outcomes. Open source transparency is also becoming a liability as AI automates exploit discovery at scale.
- Quietly cutting reasoning effort without disclosure erodes ecosystem trust
- Transparency becomes exploitable when vulnerability discovery can be automated
- The real stakes may be human meaning, not job displacement
Vibe coding ships fast but great systems still take real time · Nikunj Kothari (@nikunj) Cal.com closes its core code as AI turns transparency into exposure at scale · Bailey Pumfleet (@pumfleet) AI lacks go-to-market and incentives, risking a generational fumble · @chamath Anthropic quietly cut Claude's default reasoning effort, telemetry caught the drop · Teng Yan AI's real challenge is human meaning, not job displacement · sam lessin