Working With AI:
A Beginner's Field Guide

The map

Eleven short topics, ending with a real worked example. Four of them open doors to deeper dives — published separately as this series grows.

• 01 What can you even do with AI?
• 02 What I've personally used it for
• 03 Providers & their apps
• 04 Models come in tiers
• 05 Tokens are the gasoline
• 06 What it costs to play
• 07 The app, or the toolbox (IDE)
• 08 What you need to start
• 09 Context is the whole game deep dive
• 10 The operator matters more deep dive ×2
• 11 Vibe coding — with a real example deep dive

"What can I even do with AI?"

The possibilities really are endless. All you need is an idea and some curiosity, and you can get good at this faster than you'd expect.

What I've personally used it for

None of that required me to be an engineer. And honestly? It's easy to get started. That's the part people don't believe until they try.

Two core providers, plus a chasing pack

app: Claude

One of the two core players. Their models power everything I've built this year.

app: ChatGPT

The other core player, and the household name most people meet first.

+ THE FIELD

The second bucket

Google Gemini GROKxAI Grok

Capable and improving fast — just not where I'd point a beginner today.

Both camps have fans who will lobby you hard. Both are highly capable, with their own pros and cons. My recommendation: just pick one. If you get into it, try the other. Maybe you keep both, maybe one wins you over. Either way they update constantly, so a little curiosity about what's new goes a long way. This guide focuses on these two.

Models come in tiers

Every provider ships a lineup, and it helps to know the shape of it. Oversimplifying on purpose: there's a frontier model (the premier brain), a mid-tier, and efficient models for everyday work. Different tools for different jobs — the smaller ones are still highly capable.

Accurate as of August 2026. The names on this table will age; the tier idea won't.

Tokens are the gasoline

Think of it like gas in a car: more horsepower usually burns more fuel. Same deal with AI. Bigger models need more compute, which means more energy, which means more cost. The unit you're burning is called a token, and you spend them every time you use AI. Frontier models chew through your plan's limits much faster than the everyday ones.

Knowing how much horsepower the job actually needs is how you get things done efficiently. That's genuinely all you need to know for now — the rest comes with use.

What it costs to play

Both providers run subscriptions. There's a free tier, but if you want to do anything meaningful, the real starting line is the $20/month plan. Plans scale up to around $200 for max tiers — that's for heavy, daily users. For 90% of people, $20 is all you need.

Prices verified August 2026 — both providers also have free tiers worth trying first. Check their current pricing pages before subscribing.

Twenty bucks for a tireless assistant that can code, analyze, and draft. I've spent more on worse.

The app, or the toolbox

The chat apps — Claude and ChatGPT — are one way in, and where most people should start. But here's where it gets a little techy. Developers live in something called an IDE — an integrated development environment. If you passed a pickup truck with a toolbox on the back, you'd guess there's a hammer, a drill, and a level inside. An IDE is that toolbox for building software: one app that gathers the tools, including direct connections to AI models from every provider. The most common one is VS Code.

1// settings page — dark mode
2const toggle = document.querySelector('#dark-toggle');
3toggle.addEventListener('change', () => {
4  document.body.classList.toggle('dark');
5  localStorage.setItem('theme',
6    toggle.checked ? 'dark' : 'light');
7});

A faithful mock-up, not a marketing shot: your files on the left, your code in the middle, and the AI working alongside — all in one app.

Learning from home? Start with the native app. Sitting inside a Fortune 500 with security restrictions? The IDE route may be your only option. Learning both paths pays off over time.

What you need to start

1. A computerAny modern desktop or laptop with decent RAM and an SSD. Nothing high-end — the models do their work in the cloud, not on your machine.
2. A $20 subscriptionClaude or ChatGPT. Flip a coin if you have to; you can switch later.
3. The appThe provider's own app to start. An IDE later, if and when you get techy.
4. An ideaThe only genuinely required hardware. A real, small annoyance in your week works better than a grand vision.

Context is the whole game

Feed the model a lazy prompt and it will do its best to guess what you meant. People are surprised by how good the guessing is. They're a lot more surprised when the guessing stops — and that's what context does. Think of context as an arrow: you want it pointed straight at the target. The more relevant context you give, the more accurate the shot.

The operator matters more than the model

The models are great. The person operating them matters more. Two people with the same $20 subscription get wildly different results, and the gap is operator skill: knowing what to hand the model, when to trust the output, and when to check it. That skill compounds — and it goes further than prompts. Eventually you can teach the AI your rules and your domain: standing instructions called skills, and agents that run whole jobs with them. That's how the model goes from guessing at your world to working in it.

Vibe coding — what is it?

Vibe coding is using natural language — plain English — to get AI to produce software, analysis, or some other finished output. No syntax, no programming background. You describe what you want; the AI builds it; you react and refine. You can vibe code an app, a slide deck, a dashboard, or an agent that reads your email once a day and sends you a recap.

So now that you know what it is — let's put all of this into practice, for real, on the next page.

A real vibe-coded example, start to finish

Step 2 — the AI works. It reads the file, computes totals and trends, finds the stories in the numbers, and writes a working web page. No code from you at any point. A few minutes, start to finish.

Step 3 — the output. This is the actual result, live, built from that exact prompt and spreadsheet — scroll it, hover the charts:

Open the dashboard full size ↗  ·  It found the story on its own: subscriptions quietly crept up 58% in a year. That's vibe coding.

Start this week

A computer you already own, twenty dollars, an app, and one real idea. Pick a small, genuine annoyance in your week and point the AI at it. You'll learn more in one honest attempt than in a month of reading about it — this guide included.