How agents work
The basics¶
An agent is an AI that can do things — read files, write drafts, run steps — not just chat. This section explains how they work, in plain words, so you can supervise one with confidence.
The six pieces of every agent¶
- Instructions — the job and the rules you give it.
- Context — what it can currently see (your message, files, notes).
- Tools — what it's allowed to use (files, web, apps).
- Actions — what it actually does.
- Checks — how you (or it) verify the work.
- Stop points — where it pauses and asks you.
Every problem you'll ever have with an agent traces back to one of those six. That makes fixing things much less mysterious.
Read these in order¶
- The agent loop — how an agent thinks, acts, and checks, and where it goes wrong. Read →
- Context & memory — what the AI can see right now, what it forgets, and where real memory lives. Read →
- A vault for your agent — a simple folder of notes that becomes your AI's long-term memory. Read →
- Subagents — when one AI hands work to helper AIs, and when that's worth it. Read →
- Tools & connections — how AI plugs into your apps, and what to allow. Read →
- Keeping it safe — the short list of habits that prevent almost every disaster. Read →
If this helped
If this section made agents make sense, a small tip on Ko-fi keeps every guide here free. Tip on Ko-fi