The 5 foundational AI concepts that matter beyond learning tools

Zachly · x · 2026-11-14

A learning-roadmap post arguing that learning AI is mostly tool-learning, but a handful of foundational concepts matter far more:

Context engineering: context windows, token limits, managing what the LLM sees, compression/retrieval/memory
RAG: embeddings and vector similarity, chunking strategies, semantic search, reranking and grounding
Agent architecture: tool calling, planning and task decomposition, state management, single- vs multi-agent systems
Model fundamentals: tokens/embeddings/attention, transformers and next-token prediction, temperature and sampling, fine-tuning vs prompting vs retrieval
Evaluation and reliability: hallucinations, failure modes, deterministic vs probabilistic systems

More from coding & agent