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 August 31, 2026
AI control risk moves from research to legislation
An essay published by OpenAI's chief scientist flagged degrading chain-of-thought monitoring and called for all labs to slow scaling. Separately, legislation to ban superintelligent AI and impose 20-year prison terms for violations was introduced.
- Agents add unresolved dispute resolution layers outside existing legal frameworks.
- Partisan lines are hardening around AI infrastructure and data centers.
- AI behavioral convergence in large model deployments is spreading and poorly understood.
OpenAI scientist warns AI labs to slow down or lose control · @sksq96 AI's near-term trajectory poses risks to human control, essay argues · @merettm Ex-Kamala Harris TikTok now an anti-AI, anti-datacenter hub · @WillManidis Meta faces pressure to address AI flock behavior it calls safe · @CoryOnBrand Bill would ban superintelligent AI and jail violators for 20 years · @AndrewCurran_ AI agents create new layers of dispute resolution complexity · @_nityas
Agent workflows proliferate but measured gains disappoint
Practitioners are sharing reproducible agent workflows including self-grading loops, YouTube transcript learning, and first-principles code audits. Sober reflection over a quarter of heavy use suggests total productivity gains rarely exceed 50%, with LLM failures often eroding the surplus.
- Agentic codebase cleanup shed 375,000 lines in roughly 15 hours.
- Using coding agents well is a teachable discipline with named, transferable skills.
- LLMs enable faster starts but create new time sinks through failures and corrections.
AI self-grading pattern helps agents identify and fix their own gaps · @austin_hurwitz AI agent calls stores by phone to check product availability · @chrismaconi AI agents can learn skills by mining YouTube transcripts · @leonabboud First-principles AI prompt finds unnecessary code in any codebase · @georgepickett AI Engineering Skills Map for coding agents released · @AndrewYNg Hermes Agent repo shed 375,000 lines via 15-hour agentic cleanup · @Teknium Real-world LLM productivity gains may be well under 50% · @andrewho03 Using coding agents well is a teachable discipline with named skills · @AndrewYNg Agentic cleanup shed 375k lines and made the repo legible for all · @Teknium
Venture capital polarizing as LP skepticism reaches blind pools
Capital is concentrating in a handful of hypergrowth outliers, leaving traditional Series A rounds unfunded and pushing LPs to question blind pool economics. The seed-to-Series-A conversion rate has roughly halved since 2015, and SPV fees face long-overdue scrutiny.
- Fewer than 50 VC-backed companies have gone public annually since 2022.
- Young GPs under 25 are deploying small AngelList funds into personal networks.
- Founder satire implies perceived fraud carries little reputational downside in venture.
LPs push GPs to justify blind pool funds over secondaries · @MeghanKReynolds Under-25 GPs raising small AngelList funds, many are nepo babies · @pavelprata SPVs face long-overdue reckoning over opaque, high fees · @tbpn Hype-driven VC rounds echo Bird's 2018 cautionary tale · @NYCounihan YC S26 startups show steep fundraising premium over non-YC peers · @NWischoff YC startups still raise at a meaningful premium over comparable peers · @NWischoff Funding concentrates in hypergrowth startups, squeezing the rest · @FrancisPSantora IPO market stalls as unicorn backlog tops 950 companies · @pavelprata Odds of a seed startup reaching Series A have halved since 2015 · @whoisnnamdi Founder says fraud is the dominant strategy in venture · @Jeffreyw5000 Raising almost $18M to build a smart toilet took 178 rejections. · @ScottHickle
AI market makers counter bundlers on tokenized-stock chains
Robinhood Chain's tokenized stock pairs are enabling AI-run market-making vaults that absorb launch supply before bundlers can corner the float. In early tests, agents captured over 70% of supply and sustained orderly markets against aggressive early demand.
- Memecoins paired against tokenized stocks create strong weekend LP opportunities.
- AI vaults replace thin bundler-cornered floats with agent-managed inventory at launch.
New platform lets users launch coins against any stock or token · @econoar AI market-making vaults snipe bundlers on Robinhood chain · @0xSammy LP tokenized stocks vs. stablecoins to capture memecoin volume · @0xKarim AI market makers can hold launch inventory to counter bundlers · @0xSammy Tokenized stock and memecoin pairs offer strong weekend LP returns · @0xKarim Mosh DeFi protocol pairs every token with an AI market maker · @justinbebis