Why choosing a model per workload belongs in configuration, why the routing table has to be benchmarked through the full harness rather than guessed, and why it needs to be rebuilt on a cadence rather than trusted forever.
A comparison between infrastructure testing and agent evaluation, focused on hidden state, behavioral contracts, and feedback loops.
How a personal LLM wiki can separate raw evidence, curated memory, context packs, privacy boundaries, and evaluation loops.
A systems-through-line from Chef, infrastructure automation, and TDD in operations to context engineering for agentic AI.
Why reusable prompts and context packs deserve versioning, review, privacy boundaries, failure modes, and evaluation loops.