AI Engineer Reading List 2026

The transformer paper. Read it once, even if you skim.

GPT-3. Where in-context learning was first observed at scale.

InstructGPT RLHF in production.

An alternative to RLHF that uses written principles.

The original RAG paper.

Why long contexts have a U-shaped attention curve.

Operational, opinionated, free. Best practical RAG learning resource.

The cleanest mental model for agent design we've seen.

The reasoning + acting pattern most agent frameworks use.

Models teaching themselves to call tools.

Practitioner interviews. Always 6 months ahead of mainstream.

Sober, deep posts on Claude internals and applied AI.

The best running commentary on the AI ecosystem in 2026.

Frontier research without too much hype.

The agent worth installing first.

For evaluating and tracing LLM apps.

Open-source eval framework worth learning.