A continuously evolving collection of notes on artificial intelligence, software systems, machine reasoning, agents and computation. This site is less a blog than a public working memory — an attempt to turn reading, experiments and engineering experience into durable knowledge.
ORDER: LAST UPDATED ↓
Prompt engineering optimizes a message. Context engineering designs the information environment around a model. As agents gain memory, tools and persistent state, that distinction becomes architectural rather than semantic.
Retrieval is a runtime policy, not a storage primitive.
Episodic, semantic and working memory in practical systems.
Latency, specialization, routing and bounded tasks.
State transitions, retries, validation and observability.
Why reasoning budgets will become configurable infrastructure.
284 DOCUMENTS / 39 TOPICS
Every note has links, provenance, revision history and a place in the larger graph. Ideas are expected to change. Old conclusions remain visible, but newer evidence can supersede them.
The goal is not to collect information. The goal is to reduce the cost of thinking about the same problem twice.