Start Here With AI

Welcome. If you've been hearing about AI for two years and still haven't really used it, you're in exactly the right place. Nothing here assumes you know what a model is, and nothing here asks you to install anything.

Here's the honest version of what AI is today: a very well-read assistant that will help you with almost any kind of thinking work, and will occasionally be confidently wrong. That combination sounds scary until you realize it describes a new hire. You wouldn't hand a new hire your biggest decision on day one. You'd give them something you understand well, look at what they did, and go from there. That's the whole method, and it's the reason most beginners get stuck — they start with the flashiest thing instead of the checkable thing.

What AI is actually good at

Think of it in two buckets. It's genuinely strong at preparing: drafting, summarizing, organizing, comparing, rewriting, explaining something in plainer words, turning messy notes into something readable. It's shaky at anything that depends on facts it can't see — your numbers, your commitments, your client's actual situation.

So the right way to think about it isn't "AI does my work." It's "AI does the assembly, I do the deciding." Keeping those two jobs separate gives you a smaller job to check — not a reason to stop checking.

Open one assistant. Just one.

You need a free account and a browser. ChatGPT at chatgpt.com is the most common starting point. Google Gemini is handy if you already live in Gmail and Docs, and Claude is pleasant for writing. Pick one and stay there for a week — comparing tools is a great way to feel busy and learn nothing.

Then talk to it the way you'd talk to a person. You don't need magic words. "Here are my notes from a meeting. Turn them into a short update for my team, and don't add anything I didn't say" works better than most of the clever prompts floating around online.

The one question that saves beginners

Before you choose what to try, ask yourself something that sounds too simple to matter: which result would I be able to judge?

That's the whole trick. The tool sets what's possible. You set what's safe. If you know the work well enough to notice when something's off, you can use AI on it today. If you don't, a smooth, confident, wrong answer will sail right past you — and it will sound great while it does.

Here's a test you can run in ten seconds. Before the AI shows you anything, name two or three things a good result absolutely must contain, and one or two mistakes you'd catch instantly. If you can do that, go ahead. If you can't, pick something you know better first. That's not a setback; it's the difference between "this sounds good" and "I know why this is good."

What a first task looks like

Boring is perfect. A weekly update. A summary of notes you wrote yourself. A first draft of an email you send all the time. A checklist you rebuild every month.

And what to leave alone for now: anything legal, medical, or tax-related, anything with confidential client information, and anything where a wrong answer costs you money or trust. Not because AI is dangerous, but because you can't check it yet.

Watch it happen: one real example

A project lead I work with had five scraps of notes: the checkout copy was approved Wednesday, the analytics work is blocked waiting on data-team access, the Friday launch hasn't been reconfirmed, Priya is sending a revised QA checklist Monday, and one mobile test failed for reasons nobody has figured out yet.

Before opening AI, she wrote down her bar: every owner and date has to come from the notes, uncertain things stay uncertain, and risks don't get promoted into decisions. Then she asked for a draft update using only those notes, with instructions not to invent owners, dates, causes, or approvals.

The draft came back organized and mostly right. It also contained two sentences she never wrote: "The launch remains on schedule for Friday," and "The data team will complete access by Friday."

Those two lines are the most useful thing on this page. They aren't bizarre hallucinations — they're plausible. They sound exactly like something a teammate would say, which is precisely why nobody double-checks them. The dangerous sentence is almost never the strange one. It's the reasonable one that happens to be untrue.

She cut both lines, kept the rest, and added a rule to her prompt: don't assume a blocked item resolves by the deadline unless the notes say so. Ten minutes of assembly work gone, and not one fact handed over. That's the whole win, and it's much less glamorous than the demos.

Count the whole loop, not the fast part

This is where beginners fool themselves. The draft takes ten seconds, so it feels like a win. But if you spend twenty minutes cleaning up the source, checking every line, fixing errors, and pasting the result somewhere else, you may have saved nothing at all.

AI often moves work rather than removing it. So count the full loop: getting your material ready, writing the request, reading the result, fixing it, and putting it where the work actually lives. If checking takes longer than doing it yourself, you haven't saved time yet — give it a smaller piece, hand over better notes, or try a different task. A faster wrong answer is just rework that arrived early, which is a very efficient way to stay busy.

The part nobody tells beginners

Something quietly useful happens after a few weeks of this: your corrections start describing your own process. Keep fixing the same tone? You found a rule worth writing into the request. Keep pasting in the same missing background? That background belongs in the request every time. Keep stepping in at the same moment? You found judgment that should stay human, permanently.

Your edit history turns out to be documentation you didn't know you were writing. And the better AI gets at producing things, the more valuable your judgment becomes — because someone still has to know what matters, notice the exception, and own the outcome. That's not a consolation prize. That's the job getting more interesting.

It works outside of work, too. Planning a family trip, AI can sort flights, hotels, and activities into a draft itinerary while you keep the parts that actually matter: the budget, the fixed days, and the fact that nobody in your house is doing a 6 a.m. flight. Same shape, smaller stakes.

Your first ten minutes

Don't try to become an AI person today. Do this instead: pick one task you already understand. On paper, write down where the real information lives, three things a good result must include, two mistakes you'd spot, and which decisions stay with you. Then hand AI one small, clearly defined piece of it and keep your source visible while you read what comes back. Check the claims, not the vibe.

If it holds up, run the same handoff again tomorrow. If it holds up twice, save the prompt somewhere you'll find it — that saved prompt is the real win, not the one good chat you'll never repeat. And if reviewing cost more than doing it, you still learned exactly where your line sits today.

Before you ask AI to do more, make sure this one task works reliably.

The free AI Starter Action Guide is help with exactly that: the same steps on paper, with room to write your source and your must-haves, the request to copy, a seven-day plan, and the five things to leave alone for now. Print it, use it on one real task, and you'll be further along than most people who've been reading about AI for a year.

Open ChatGPT in a new tab

Help me find one useful thing to try with AI today.
Ask me a few short questions, one at a time, to find work I already know well enough to check myself. I'd like to know:

what kind of work I do most weeks
which parts of it are repetitive
where the real information lives (my notes, a document, real numbers)
what would tell me a result is wrong

When you have enough, give me three things:

One task to start with, and why it fits.
A request I can copy and paste, written so you only use the material I give you and never invent names, dates, numbers, or approvals.
A short list of what to check in the result before I use it.

Keep it plain. No jargon, no generic tips.