Effective harnesses for long-running agents
Practical patterns for preserving context, making incremental progress, and leaving work in a state another agent can pick up.
Reference shelf
Library
There are tons of great resources out there about AI Engineering, LLMs and AI/ML in general, these are the ones that I found most useful and refer back to frequently.
AI Engineering
Harness design for long-running application development
A follow-on look at decomposition, separate evaluators, and testable contracts for longer-running coding work.
Agents
Chip Huyen's engineering-oriented treatment of tools, planning, failure modes, and evaluation for agentic systems.
Agentic Engineering Patterns
Simon Willison's evolving field guide to working with coding agents, including Git, subagents, testing, and manual QA.
LLM Internals
Deep Dive into LLMs like ChatGPT
Andrej Karpathy's broad, code-aware walkthrough of how language models are built, trained, and turned into assistants.
Transformers, the tech behind LLMs
The cleanest visual overview of the next-token machine, embeddings, model weights, and the transformer block.
Attention in transformers, step by step
The visual explanation I would point to first when query, key, value, and multi-head attention still feel abstract.
Transformers from Scratch
Brandon Rohrer's patient, matrix-first construction of a transformer from one-hot encodings through attention and tokenization.
LLM Visualization
Brendan Bycroft's explorable 3-D view of a working GPT, especially useful for connecting tensor shapes to the data path.
Transformer Explainer
A live GPT-2 in the browser with attention, probabilities, temperature, top-k, and top-p exposed for inspection.
TensorFlow Playground
A simple interactive way to watch a small neural network learn and see what layers, features, and learning rates change.