Archives
All the articles I've archived.
-
Agent Harness Turns Development into Executable Patterns
Agent harnesses are not just multi-agent wrappers. They route context, package workflow skills, expose project-specific tools, and use traces as the feedback surface for improvement.
-
The ratchet: turning agent failure into infrastructure
Reliable agent systems are built by converting repeated failures into durable artifacts: skills, tests, hooks, sandboxes, rules, and verification gates.
-
Memory is context delivery, not storage
Most agent memory failures are not database failures. The hard part is delivering the right slice of experience into future behavior with source, scope, and authority.
-
Agent UX optimizes delegation
Agent products should not optimize for chat responses. They should compile runtime activity into delegation state, attention boundaries, valid actions, and evidence-backed trust.
-
Game theory as a learning lens, not an arithmetic class
A learning-oriented distillation of game theory resources: Schelling, Yale, Model Thinking, Game Theory 101, and evolutionary games. The point is not memorizing formulas, but learning to see constraints, commitments, types, and feedback.
-
The agent stack is getting boring, finally
A few links from today made the shape clearer: tmux for runtime, eve for durability, lossless-claw for memory, 12-factor-agents for control flow, and Obscura for a lighter browser surface.