Start here

This section is the on-ramp to the field guide. Read it top-to-bottom if you are new to building with AI agents, or jump to the page that matches your situation.

What's in this section

Page	What it covers
What is an agent?	A working definition — model + tools + loop + stop condition — and how agents differ from one-shot chat.
Tasks where agents shine	Concrete examples of workflows that benefit from agent architecture, with reasons why each one works.
When not to use an agent	Anti-patterns and failure modes: the cases where a simpler tool, a script, or a human is the right answer.
Decision tree: which surface?	A structured guide to choosing between chat apps, custom GPTs/projects, desktop clients, and the API/SDK.
Beginner path	A 30–60 minute setup walkthrough for someone who has never built an agentic workflow before.
Power-user path	For daily ChatGPT/Claude users ready to add CLI tooling, MCP servers, and structured evals.
Local-first path	Running as much of the stack as possible on your own machine, with minimal cloud surface area.
Team path	SSO, shared resources, PII boundaries, audit logs, and eval as a team practice.
Safety baseline	An opinionated pre-build checklist: stop conditions, least-privilege tooling, HITL, logging, and kill switches.

How to use this guide

Start with What is an agent? and When not to use an agent regardless of your experience level — the definitions here are used throughout the rest of the guide. Then pick the path that matches your context (beginner, power user, local-first, or team).

If you are evaluating whether to build at all, read the decision tree first.