Articles & Resources
Long-form writing on the engineering side of AI systems — architecture, reliability, and cost — plus references I keep coming back to. For a running log of what is shipping in the field, see Trends.
Articles
Designing AI Systems That Hold Under Load
The hard part of production AI stopped being model quality and became system quality. A working guide to the three constraints that actually bind — memory rather than compute, statistical rather than binary correctness, and the optimization order that pays — with the arithmetic to size each one.
Resources
A short list of references I return to rather than an exhaustive bibliography.
- Scaling Laws Meet Model Architecture: Toward Inference-Efficient LLMs — treats inference cost as a first-class term in the scaling objective rather than an afterthought.
- Sebastian Raschka's LLM research paper lists — the most reliable running survey of what is worth reading each quarter.
- Model Context Protocol specification — the spec itself is short and worth reading before adopting any agent framework that wraps it.
- Google SRE Book & Workbook — predates the current wave and still the clearest treatment of error budgets, load shedding, and graceful degradation.
- World Model for Robot Learning: A Comprehensive Survey — a good entry point to where learned simulators are heading for embodied work.