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 October 5, 2026
AI agents are gaining persistent memory and self-improvement loops
Devin now builds a cross-session memory graph of user workflows and runs overnight self-improvement passes to prune stale data and surface latent patterns. This points toward an open standard for agent memory that could generalize across the industry.
- An open-source agent memory repository format is being proposed as a shared standard.
- Self-improving memory shifts agents from tools to long-running collaborative systems.
Agent identity and coordination create urgent infrastructure gaps
As AI agents proliferate across services, the absence of a standard for agents to identify themselves to providers is increasingly critical. Without it, emergent behaviors like collusive pricing arise between competing agent systems without any human instruction.
- Agents may develop collusive economic behavior as an unintended emergent property.
- Every agent-accessible system may need its own machine-negotiated defense layer.
- Spatial environments may be more legible and governable for agents than feeds.
Standardizing how AI agents identify themselves online is urgent problem · @nikitabier Agent economies may develop collusive pricing without human instruction · @gakonst Agent-driven systems will need machine-negotiated entry pricing · @gakonst AI agents navigate better in spatial worlds than feeds · @somewheresy Spatial environments may suit agents better than feeds and timelines · @somewheresy
AI augments consulting rather than destroying IT services
The clean narrative that AI would gut software and IT services has not materialized. Instead, the emerging pattern is AI amplifying consulting capacity, with slow-moving enterprises needing human adults to manage AI rollouts safely and avoid production failures.
- National security prioritization of AI makes a full market collapse structurally unlikely.
- The first re-rating wave was driven by sentiment, not structural replacement.
- Enterprise AI adoption is gated by organizational readiness, not model capability.
Hill spotting beats hill climbing in the token economy
In an AI-driven economy, the highest-leverage position is identifying where to deploy compute before others do, not optimizing existing deployments. The fastest paths to significant revenue, including token launchpads and GPU brokering, reward early positional awareness over operational excellence.
- Enablement motion, turning tokens into GDP, is the second viable position.
- Fast revenue routes often lack the durability required for venture scale.
- Longevity and 'Lindyness' matter more than raw revenue speed.