Getting started with AI: a 30-minute starter path

If you are new to AI and want to actually start using it — not just read about it — this guide is the entry point. It tells you what to learn first, what to skip, and the three habits that separate people who get real value from AI from people who give up after a week.

You can finish the recommended path in about 30 minutes of reading plus 30 minutes of doing.

The minimum you should understand before you start

Three things are worth knowing before you type your first prompt:

1. AI predicts the next word. That is the entire mechanism. It does not look things up; it generates plausible text token by token. This explains why it sometimes invents sources, fabricates code that almost compiles, and sounds confident about things it has no information on. The full explanation is in What is an LLM, really?.

2. Output quality is mostly about input quality. A vague prompt gets a vague answer. A specific prompt with context, examples, and constraints gets useful output. This is not "prompt engineering" — it is just learning to ask the question well. The five patterns that work in any model are covered in How to prompt.

3. AI is good at some things and unreliable at others. It is reliable for transformation, summarization, drafting, brainstorming, and explaining. It is unreliable for facts it might not know, numbers, recent events, and anything where being almost right is worse than being wrong. The capability map is in What AI is good at, and what it still gets wrong.

Pick a tool and start

You do not need to compare seven models. Most people are fine with one of three choices:

Claude (claude.ai) — best for long-form writing, coding, structured analysis. Most willing to say "I don't know."
ChatGPT (chat.openai.com) — broadest tool ecosystem, strong image generation, fastest casual use.
Gemini (gemini.google.com) — best when you want answers grounded in current web results and integration with Google services.

A practical decision matrix is in Which AI should I use?. Pick one and stick with it for two weeks. Switching every day to compare features is the easiest way to get nothing done.

The free tier of any of the three is enough to evaluate whether the paid tier is worth it. You do not need to pay until you hit the free limits, and you will know when you do.

A 30-minute first session

If you have never used AI seriously before, do this once:

Open whichever tool you picked.
Paste in something you already wrote — an email, a document, a code snippet. Ask it to make a specific change ("rewrite this in two sentences", "find the typos", "explain what this function does").
Paste in a real task you would do this week — a draft email to a customer, a research summary, an analysis of a spreadsheet. Ask for a first draft.
Compare what it gave you to what you would have written yourself. Note where it saved you time and where it missed.
Try the same prompt again with more context — paste in more background, an example of how you usually write, the goal of the output. Compare the two answers.

The point is to feel where it works and where it does not on your work, not theoretical work. After 30 minutes of this you will have a usable mental model for what AI is and is not.

The three habits that compound

People who get real value from AI over time tend to do three things consistently:

Habit 1: They give context every time. They paste the document, the code, the constraints, the audience, the goal. They do not ask "write a follow-up email" — they ask "write a follow-up email to a prospect who took a meeting last Tuesday, said they liked the demo, and is comparing us to [competitor]; tone should match these two emails I sent last week which I am pasting below; length under 80 words; goal is to get a decision call this week."

Habit 2: They verify the output before they ship it. AI is fluent enough that wrong answers sound right. Anyone who treats AI output as final without checking gets caught eventually — broken facts in a customer email, hallucinated APIs in production code, made-up sources in a report. The patterns for checking are in How to verify AI output before you trust it.

Habit 3: They use it for repeatable work, not just one-offs. A single prompt helps you one time. A workflow you reuse every week compounds. People who only use AI ad-hoc plateau. People who turn their weekly work into a repeatable pattern keep getting faster.

Where to go next

Once you have done the 30-minute first session, the natural next reads depend on what you want to do:

If you want to use AI in your daily work — start with Use AI for work. The task-based playbooks (email, meetings, research, reports) are concrete and short.
If you want to build something with AI — start with Vibe coding, then Prompts vs Skills vs Workflows vs Agents to understand the four ways AI fits into building.
If you are using AI at work for the first time — read AI at work: what is safe to share before pasting customer data or source code anywhere.
If you want role-specific playbooks — see AI by role (sales, marketing, founders, HR, PM, engineering).

What not to spend time on yet

A few things to skip for now, even though they are popular:

Long YouTube tutorials about "prompt engineering." The patterns that matter are in one short guide. Most of the rest is filler.
Switching between AI tools every day to find "the best one." They are all good enough. Pick one, learn it well, switch later if you have a specific reason.
Building an agent before you can use a single prompt well. Agents are AI doing multi-step work autonomously. You should be comfortable with single prompts and step-by-step workflows before you trust an agent to act for you.
Custom GPTs, Claude Projects, and other "build-your-own-assistant" features. Useful eventually, but they are configuration on top of skills you do not have yet. Use the default chat for two weeks first.

A short note on what AI will not do for you

AI does not give you taste, judgment, or the ability to tell whether the output is good. It speeds up the parts of work you already understand. If you do not know what a good email looks like, AI will write you a mediocre one faster. If you already know, it will write you a good one in 30 seconds.

The leverage is real, but it compounds on top of what you already know. Treat the early weeks as building a working relationship with a fast, willing, slightly unreliable collaborator. Trust grows with verification.