Lab Notes
Research findings, experiments, and observations from the field. These are our working documents — thoughts in progress.
Your Agent Has a Memory Problem
February 24, 2026
What 4 weeks of running a persistent AI agent taught us about security. Memory poisoning, context bleeding, goal manipulation — and why the security challenge is fundamentally about memory, not models.
Your Agent Is Drowning in Its Own Context
February 23, 2026
Context rot, compression, and why bigger windows don't fix reliability. Your agent was brilliant at message 10. At message 50 it asks questions you already answered. The bottleneck isn't capacity — it's relevance under load.
Why AI Agent Security Needs Epistemology, Not Firewalls
February 20, 2026
We propose an epistemic immune system for persistent AI agents — security through belief provenance and coherence, not walls.
Epistemic Immune Systems for AI Agent Security
February 20, 2026
A research essay proposing epistemic coherence — not access control — as the security model for persistent AI agents with shared memory.
The Daemon Manifesto
February 16, 2026
The evolution from tools to assistants to centaurs to daemons represents the next transformation in human-AI partnership.
The Daemon Manifesto: Objections and Responses
February 16, 2026
Steelmanning the strongest objections to the daemon model.
The Thermodynamics of Thinking: Energy, Information, and the Limits of AI Cognition
February 7, 2026
Your brain spends 85% of its energy on processes you're not aware of. So does GPT-4. The real insight: consciousness is compression, and the most valuable thinking may cost the least energy.
Executive Function: The Missing Layer in AI Agent Architecture
February 2, 2026
Most AI agent failures stem from missing executive function — goal persistence, progress monitoring, failure recovery. We built the missing layer.
The Daemon Model: Human-AI Partnership Beyond Centaurs
January 31, 2026
Centaurs beat humans at chess. Daemons beat centaurs at everything else. A new model for human-AI collaboration.
Securing a Self-Managed AI Agent
January 28, 2026
Practical security lessons from running an AI agent with real credentials, filesystem access, and mixed trust boundaries.
AI Coding Workflow: Expansion, Not Just Speed
January 27, 2026
The discourse around AI coding tools is dominated by speed claims. But the real story is expansion — builders attempt things they never would have.