Writing
Engineering notes.
Open Source
Open source as proof
A threat model without tests is a story. Publish the suite, the limitations, and the parts that are still experimental.
Aug 2026
System Design
System design for probabilistic software
Treat models as untrusted components. Put the invariants in the runtime, the tests, and the protocol — not in the prompt.
Aug 2026
RAG
RAG without faithfulness is search
Retrieval is not reliability. If you cannot score context, citations, and unsupported claims, you have a nicer grep.
Aug 2026
Agentic AI
Agents need a state machine
A demo is a path. A system is a set of invariants that still hold when the path changes.
Aug 2026
AI Security
Allowed is not verified
Tool permission answers whether an agent may attempt an action. It does not prove the approved effect actually happened.
Jul 2026
LLM Evaluation
Evaluation is the product
If you cannot fail a pull request when an agent regresses, you do not have quality. You have anecdotes.
Jul 2026