AIGuide
Your guided path · beginner → advanced

AIGuide

A friendly, visual walk-through of everything you need to understand modern AI agents — what they are, how they think in a loop, how they delegate to subagents, how skills give them know-how, how it all runs in Claude Code inside VS Code, and how to collaborate on the code you build. Start at lesson one and follow the path; each page hands you to the next.

Start lesson 1 — the Basics See the full path
8 lessons No experience needed Runs offline
The AI loop 🧭Gather context 🧠Reason 🛠️Act 👀Observe 🔀Decide
1️⃣

Follow in order

The lessons build on each other. If you're new, just start at the top and go down.

⬅️➡️

Use the bottom nav

Every page has Previous / Next buttons and a link back home, so you never lose your place.

🎯

Jump around freely

Already know a topic? Use the top menu or the cards below to skip straight to it.

The curriculum

Your 8-lesson learning path

From "what even is an AI agent?" to orchestrating a team of them and shipping code with a collaborator. Click any lesson to begin.

1
Start here · plain English

The Basics

Zero jargon. What an AI agent is, the try-look-adjust loop, what a subagent is, and why any of it matters — all with friendly pictures.

Begin
2
The core actor

Agents

The deep tour of a single agent: the "augmented LLM", its anatomy, the tool-use loop, the toolbox, models, permissions, memory, and the ways to run it.

Begin
3
Delegation & teamwork

Subagents

How an agent hires helpers with their own isolated context, defines them, runs them in parallel, and orchestrates them with proven patterns.

Begin
4
Reusable know-how

Skills

Package expertise into a SKILL.md the agent loads only when relevant. Progressive disclosure, invocation, creation, and when to use them.

Begin
5
It all comes together

In Claude Code & VS Code

The real mechanics: the exact five-phase loop, how subagents are defined and invoked, and how the whole thing lives inside your editor.

Begin
6
Mastery

Advanced — loop engineering

Context windows, isolation, parallel fan-out, orchestration depth, and the design patterns that make an agent reliable instead of just clever.

Begin
7
Ship it with others

Collaboration

The practical side: how to work on the same files as someone far away using Git + GitHub, VS Code Live Share, and the right habits.

Begin
8
Reference

Glossary

Every term from the whole guide, defined in plain language, with a live search box. Come back here any time a word trips you up.

Open
Is this for you?

Who benefits, when to use it & what you could build

Each technique shines in different situations. Here's who tends to reach for each one, the moment it pays off, and a real project it unlocks — so you can spot the parts that matter for you.

🤖

Agents

Letting an AI actually do multi-step work, not just answer.

WhoSolo devs & indie hackers, students learning to code, analysts, and non-coders automating repetitive computer tasks.
Use it whenA task needs try → check → adjust, not one reply — fixing bugs, refactoring, scripting, or researching across many files.
ExamplePoint an agent at a failing test suite and have it find the bug, fix it, and re-run until everything's green.
🧑‍🔧

Subagents

Delegating heavy or parallel work to focused helpers.

WhoDevs on larger codebases, teams, researchers, and anyone hitting the "context is full" wall.
Use it whenA job has a noisy side-quest (reading 40 files) or independent parallel pieces, or you want an unbiased second opinion.
ExampleAudit a whole repo for security issues by fanning out one reviewer per folder, then merging their findings.
🧠

Skills

Packaging your process so you never re-explain it.

WhoTeams wanting consistency, anyone who repeats the same prompt, and non-devs with a repeatable workflow.
Use it whenYou keep re-typing the same instructions, have a house style or checklist, or reference docs the AI should follow.
ExampleA /pdf-export skill that turns any Markdown report into a branded, formatted PDF the same way every time.
🔁

Loop engineering

Designing the cycle so the agent is reliable, not just clever.

WhoPeople building AI apps or agents, power users, and anyone whose agent wanders or ships wrong answers.
Use it whenReliability matters — you need verification, a clear stop signal, or parallel scale you can trust.
ExampleBuild a research pipeline that fact-checks itself — findings only survive if independent "skeptic" agents can't refute them.
🧩

Claude Code in VS Code

Bringing all of the above into your actual editor.

WhoAny VS Code developer, students learning to build, and pros weaving AI into daily work.
Use it whenYou want AI help inside your project — with reviewable diffs, editable plans, and permissions you control.
ExampleLand a new feature with side-by-side diffs you approve, or map an unfamiliar codebase with the Explore subagent.
🤝

Collaboration

Working on the same files as someone far away.

WhoRemote teammates, open-source contributors, students on group projects, and freelancers with clients.
Use it whenTwo people need the same project and you're not on the same network — different homes, cities, or time zones.
ExampleBuild a website with a friend across the country via GitHub — each on your own branch, merged through pull requests.
🧭

Not sure where to start?

If you're brand new, just begin at Lesson 1 and go in order — every technique above is introduced gently before it's used. The scenarios here are simply signposts for when each one becomes worth reaching for.

The thread running through it all

One big idea: the loop

Every lesson circles back to this. An AI agent doesn't answer once — it runs a loop: act → observe → decide, over and over, until the goal is met. Designing that loop well is what turns a clever model into a dependable helper.

By the end of this guide you'll understand agents, subagents, skills, the loop, Claude Code in VS Code, and how to build with other people — enough to actually put it to work.

Start lesson 1