16
Oct
Frontier models: pick for the job, not the score
14 min read
Field notes from Amrita Thavrani
AI, without the hand-waving.
For product and program managers who keep asking "what was that again?"
Start here
01
Understand the term
Plain-English definitions, when the meeting turns into alphabet soup.
Open the glossary → 02Find the broken layer
Prompt, context, loop, or harness? Find the failure before you rewrite the prompt.
Read Four Layers → 03Run the AI program
Make evaluation a working practice, with ownership beyond launch.
Read LLM Evaluation →AI, in plain English
Context window
The token budget for the input and output a model can handle in one request.
RAG
Retrieving information and adding it to the model's input to help generate an answer.
Agent
A system in which a model chooses actions and tools to work toward a goal.
Recommendations
Other newsletters worth reading.