The AI-native dimensions of AI engineering
Company-agnostic, interview-prep-led guides for engineers moving into AI-native roles. Organized by what AI-engineering interviews actually test and the documented reasons strong candidates fail — taught for transfer to the novel problems real interviews throw at you, not memorization.
Guides
- Evaluation & benchmarking — arguably the scarce skill of 2025–26 as model training commoditizes: deciding what “good” means for a real system, measuring it honestly, and knowing how the measurement itself can lie. Complete — 13 chapters.
- LLM application engineering — building reliable systems on top of models: prompting as a discipline, retrieval & RAG, agents and tool use, and the cost/latency trade-offs that decide a design. Complete — 13 chapters.
- Production AI systems — operating AI systems after the demo works: safe deployment, latency and cost at scale, observability, evaluation in production, and the bad day. Complete — 13 chapters.
- Working with AI — engineering with an AI assistant the way interviews now judge it: governing, verifying, and explaining code you didn’t type. The durable skill behind AI-assisted coding rounds — communication and judgment over prompt fluency. Complete — 13 chapters.
Most guides are paired with a hand-rolled companion library you
build chapter-by-chapter (mini_eval for evaluation,
mini_rag for retrieval, mini_agent for agents,
mini_prod for production) —
deliberately minimal and for learning, not production — then bridge
to the production tools (eval-toolkit, RAGAS, LangChain).
Working with AI instead ships workflow artifacts — a preregistration
template, a review rubric — since it teaches a practice, not a system.
Part of the guides.brandon-behring.dev hub.