Notes from the system.
Long-form essays and architecture breakdowns. Written in a documentation style — short on rhetoric, long on diagrams, named patterns and explicit tradeoffs.
State Engineering: Designing Agent State Systems That Don't Pollute
State engineering designs the typed schemas, reducers, scoping rules, and immutable ledgers that allow multi-agent systems to coordinate without state pollution or context bloat.
Agent Observability: Designing Systems You Can Debug, Trust, and Improve
Agent observability creates an evidence trail for what an agent saw, decided, changed and proved—so production behavior can be diagnosed, governed and improved.
Graph Engineering: Designing Agent Systems That Coordinate
Graph engineering makes the handoffs, state, authority and verification rules between agent loops explicit—so multi-agent work can coordinate without drifting.
Loop Engineering: Designing Agent Loops That Converge
Loop engineering designs the triggers, handoffs, evidence, state, budgets and stop rules that make repeated agent work safe and verifiable.
Harness Engineering: The Missing Reliability Layer in Agentic AI Systems
Harness Engineering introduces a runtime discipline for building reliable, production-grade agentic AI systems through deterministic execution boundaries.
The Shape of Agentic Systems
Most agent failures are not model failures — they are shape failures. A taxonomy of the loops, hierarchies and graphs that actually work in production.
Memory Is the System
Agent capability is mostly a function of memory design. A breakdown of working, episodic, semantic and procedural memory in practice.
RAG Is a Retrieval Problem, Not a Generation Problem
Why most RAG systems plateau, and the retrieval-side investments that actually move the quality curve.
Evals as Product Engineering
Treating evaluation as a first-class system — datasets, judges, traces, regressions and the discipline of continuous quality.
Subscribe via RSS for new essays.
More in the queue —