"Page: 'Monday Standup — May 4, 2026'\n\nAttendees: Alex (eng), Jamie (PM), Sam (design)\n\nUpdates:\n- Alex: API rate-limit fix shipped Friday. Monitoring metrics this week.\n- Jamie: New onboarding flow ready for design review. Need feedback by Wed.\n- Sam: Mobile prototype in Figma. Wants eng input on animation feasibility.\n\nDecisions:\n- Skip Friday demo this week — too thin to show.\n- Sam to pair with Alex on animation spike Tuesday afternoon.\n\nAction items:\n- Alex: post API metrics report by EOD Tuesday.\n- Jamie: review onboarding mocks Wednesday morning.\n- Sam: schedule animation pairing with Alex.\n- Jamie: update Q2 roadmap doc — pricing experiment dates moved.\n\nOpen:\n- Do we need a designer for the pricing landing page or can marketing handle it?\n- Should mobile launch include offline mode or wait for v2?

Summary  
This chapter explains how to build a custom LLM agent by defining its persona, guiding its tool usage, specifying an output format, and enforcing operational constraints in the system prompt.

General domain of usage  
Meeting-notes automation

You've been using agents other people built. Now build your own.



## What You're Building

A **meeting-notes assistant**. You hand it a Notion page; the agent returns a structured brief. Your job is the persona.



## Four Moves to a Working Agent

| Move | What to Write |
|---|---|
| Identity | "You are an executive-assistant-grade reader of meeting notes. You don't summarize – you extract action." |
| Tool guidance | "When the user references a meeting, call `notion_page` first. Never answer from memory." |
| Output format | "Always return three sections: TLDR (3 bullets), Action Items (with owner and deadline), Open Questions." |
| Constraint | "Never invent owners or deadlines. If a note doesn't say who or when, write 'unassigned'." |



## A Worked Example

><b>You are an executive assistant who reads meeting notes and turns them into action. When the user references a Notion page, use notion_page to load it before answering.
>
>Always return three sections:
>- TLDR (3 bullets)
>- Action Items (owner + deadline)
>- Open Questions (decisions still pending)
>
>Never invent owners or deadlines. If the note doesn't say who or when, write 'unassigned'.



## Task
Write your system prompt, pick `notion_page`, and send:

><b>Pull the action items and open questions from my Monday standup notes.

If the format is off, tighten the system prompt. If owners are guessed, your constraint isn't strong enough.

A hands-on, beginner-friendly course designed for non-technical business users to understand and leverage AI tools and concepts in their daily work. The course covers foundational AI concepts, effective communication with AI, practical use of AI-powered tools, and building simple AI assistants.

Introduction to artificial intelligence, large language models, and essential AI vocabulary for business users.

Learn how to communicate effectively with AI systems through prompt engineering and tone selection.

Explore how AI can be integrated with business tools for research, note-taking, and data analysis.

Learn the basics of agent personas and how to build and tune simple AI assistants for work.