Categories

Model watch

Notes about capability boundaries, small-model positioning, practical evaluation, and multi-model design choices.

• Small models matter again, and not just because they are cheaper

Applied AI

RAG operations, eval loops, context strategy, costs, review design, and the unglamorous parts that make systems stable.

• Which LLM evaluation metrics actually help product decisions
• Seven operational questions I ask before a RAG system goes live
• Bigger context windows make application design more disciplined, not less
• From demo to delivery, the missing piece is often responsibility, not model quality

Agent workflows

Task decomposition, tool use, state handling, audit trails, and human handoff points that survive beyond framework cycles.

• After the agent framework hype, what capabilities actually remain

Monthly signals

Low-frequency observations about how the field is actually changing once the launch-week excitement settles.

• Early March AI signal: less talk about model size, more talk about workflow closure
• How I keep an AI reading list without being dragged around by every new term
• Why a heavier AI tool stack often calls for a simpler personal workflow