The AI application skill stack: grounding, agent loops, evals, production
Source
x.com
Author
Andrew Ng
Date
Context
Building an AI application requires a way to decide what to change after observing its mistakes. In his application-engineering article, Andrew Ng describes a repeated loop: inspect outputs and execution traces, choose measurements that reflect the product's needs, then use the results to direct the next change. The tests may use code, model judgments, or human review, and the tests themselves need evaluation. This expands the application-building branch of his skills map, connecting model and workflow choices to evidence about how the resulting system behaves.
Terms in this piece · Glossary
LLM — A large language model — the neural network behind tools like Claude and ChatGPT, trained on huge amounts of text to predict what comes next.
grounding — Tying a model's answers to checkable sources — retrieved documents, live data, tool results — instead of letting it answer from memory alone.
RAG — Retrieval-augmented generation — fetching relevant documents first and pasting them into the model's context so it answers from your data instead of memory.
agentic loop — The cycle an agent runs in: decide, call a tool, read the result, decide again — repeating until the goal is met or a stop condition fires.
