langchain-ai/ambient-agent-101 ? reverse-engineered prompt

Build me a notebook based demo for an ambient email assistant that teaches the basics of LangGraph while actually being useful.

I want it to read an email, decide whether it needs a reply or some other action, draft a response, and when needed ask for human approval before sending anything. It should also remember user preferences over time, so it gets better with feedback. Use mock email and calendar tools for the main version, but make it easy to swap in Gmail later.

Please organize it as a clear learning flow in a notebook, with the supporting code in a Python package, and include a simple evaluation setup so I can test email replies, tool use, and triage decisions. If you need current docs for LangGraph, LangSmith, or Gmail integration, look them up online. Also make sure it runs cleanly with a normal Python setup and includes tests so I can verify everything works.