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 June 1, 2026
Agentic execution scales faster than verification or human adoption
AI output has expanded dramatically while enterprise adoption and verification tooling lag significantly behind. The asymmetry between cheap agentic execution and costly outcome validation is becoming the defining friction of the current deployment moment.
- Large org deployment requires navigating seven layers of internal process
- Execution cost falls fast while claim verification cost barely moves
- Recursive self-improvement tooling may signal actual exponential takeoff has begun
AI output is exploding while human adoption stays flat · @jenzhuscott Even working AI faces seven gates of operational hell in big orgs · Brandon Carl AI execution is getting cheap faster than verification can keep up · cgi.md Recursive self-improvement tools mark early exponential liftoff · Elad Gil The last six months may be among the most important in history · Elad Gil
One operator with parallel agents achieves team-scale output
A single person orchestrating 20 to 30 parallel agents can now match the throughput of engineering teams, investment analysts, and content operations. The emerging unit is one orchestrator setting objectives and reallocating compute, not a traditional team structure.
- One engineer shipped 40 pull requests a day using parallel agents
- In YC's spring batch, 60% of one-liners mention AI or agents
- The org chart flattens as compute replaces headcount scaling
Solo engineer ships 40 PRs a day running 20-30 parallel agents · Peter Yang (featuring Kun Cheng) A solo investor replaces a whole team with AI agent workflows · @zeroxkyle Naval's rule for the AI era is waste tokens to save human time · Naval Founders will run teams of agents like portfolio managers setting OKRs · deana Three-agent content pipeline cuts post production from 30 hours to 4 · Payton Kaleiwahea (summarizing Alex Lieberman) YC Spring '26 is the agents batch, 60% pitch AI in their one-liners · Ollie Forsyth Distill frontier models into local agents for a private personal assistant · Tomasz Tunguz
Open models match closed ones, pricing gap stays wide
The capability gap between open-weight and closed models has closed faster than expected while pricing has barely moved, creating immediate arbitrage for builders who route across providers. Application vendors that stay provider-neutral and charge on outcomes rather than tokens are finding structural cost advantages.
- Lindy's switch to DeepSeek cut costs and improved performance simultaneously
- Model routing becomes a primary lever for cost and risk management
- Charging on outcomes rather than inference tokens realigns incentive structures
Open and closed model capability converged, but pricing didn't · Chamath Palihapitiya AI apps beat the labs by staying Switzerland and routing across every model · Brad Menezes Lindy moves all traffic off Anthropic to DeepSeek v4, cheaper and better · Flo Crivello Raw intelligence will get cheap, but outcomes don't compress · Sierra AI conventional wisdom keeps reversing, from wrappers to desktop GUIs · Greg Isenberg OpenAI and Anthropic will move up the stack and absorb Cursor and Lovable · Jason Calacanis
Venture incentives misalign as fees rise and returns fall
Top VC funds now extract fees at many times their historical rates while returns have weakened relative to public markets. Early-stage culture has shifted toward performative input metrics, rewarding token burn and launch visibility over genuine product building.
- The venture power law at fund level is largely self-inflicted
- Benchmark's new growth fund signals a structural shift in firm strategy
- AI-native company speed is outpacing traditional slow-moving venture processes
Benchmark raises $2B and launches its first growth fund · Kate Clark AI investing hits cost walls as power-law belief breeds concentration · Brandon Carl Traditional VC's slow process clashes with AI-native company speed · Dave Fontenot Silicon Valley traded weirdos and tinkerers for input-maxxing strivers · Andrew Ziperski The venture power law is self-inflicted at the fund level · Dan Gray (Odin) Early-stage venture now rewards performative input-maximizing over building · Dave Fontenot Striver founders input-max performative metrics over real product fit · Andrew Ziperski Top VC firms now extract 10x more fees as their returns weaken · Dan Gray