There is a gap between people who say AI is useful and people who say it is overhated rubbish, and it is almost never about the tool. Both groups are usually paying for the same subscription.
The difference is 8 habits. None of them are technical. None of them need a developer. Most take a fortnight to build and then run in the background forever.
Here they are in the order they compound, because each one makes the next one work better.
1. Ask first
What it is: When you hit something you do not know, your hand goes somewhere by reflex. To Google, to a colleague, or to the pile of things you are avoiding. Change that reflex to point at AI instead.
Why it matters: This is the only skill on the list that costs nothing and takes no learning. It is also the one that decides whether the other 7 ever happen, because a tool you open twice a week never becomes a tool you are good at.
The test is not whether AI gives you a better answer than Google. Sometimes it will not. The test is whether you now have somewhere to put a half-formed question at the exact moment you have it, instead of parking it.
Start with the messy ones. The instructions that make no sense, the photo of a part you cannot name, the email you have rewritten 4 times. Those are the questions a search box has always been bad at.
2. The second look
What it is: Assume the answer is roughly right and specifically wrong. Check anything you are going to act on, spend money on, or send to someone else.
Why it matters: AI is fluent, and fluency reads as confidence whether or not the content is true. A wrong answer arrives in exactly the same tone as a right one, which is a genuinely new problem. You have spent your whole life reading hesitation as a signal, and that signal is gone.
The practical version is narrower than "check everything", which nobody does. Check 3 things: numbers, names and anything with a date on it. Those are where it goes wrong most and where being wrong costs the most.
If it gives you a source, open the source. A link that is not clicked is not a check.
3. Feed the brief
What it is: The quality of what comes back is set almost entirely by how much you put in. Give it the job, the background, the limits, and permission to ask you questions.
Why it matters: This is where most people quietly decide AI is not very good. They type 6 words, get something generic, and conclude the tool is generic. They gave a new starter a one-line instruction and blamed the new starter.
4 parts, in order:
- The job. What you want, said plainly.
- The background. Who you are, what the business does, who it is for.
- The limits. What it must not do. Words you never use, audiences you do not target, the length, the format.
- The question invitation. End with: "Ask me anything you need before you start."
That last line is the cheapest upgrade available. It costs 4 words and it turns a guess into a briefing.
Try this one now. Open your AI and type: "I want to achieve [your goal]. Do not answer yet. What do you need to know from me to give me the best possible answer?"
The list that comes back is the context you have been leaving out of every prompt you have ever written.
4. Buy strategy, not basics
What it is: Use AI to get yourself up to speed on a subject before you talk to the person you are paying for it.
Why it matters: Expert time is the most expensive thing a small business buys, and most of it gets spent explaining foundations. Your accountant charges the same rate for "what is a BAS" as for "should we restructure". You choose which one you buy.
Spend 20 minutes getting AI to explain the landscape, the vocabulary and the obvious questions. Turn up knowing what things are called. The conversation you have then is a different, much more valuable conversation.
This works for accountants, lawyers, insurance brokers, freight, web developers and anyone else whose first meeting normally starts from zero.
5. Train it like a new hire
What it is: Stop expecting it to be right on day one. Give it feedback repeatedly over weeks, the way you would with a person who just joined.
Why it matters: Nobody hires someone on Monday and calls them useless on Tuesday, but that is exactly what most people do with AI. They try it once, get a mediocre result, and never go back. Meanwhile the people getting real value have been correcting theirs for months.
The mechanical version: when it gets something wrong, do not just fix it yourself and move on. Say what was wrong and what right would have looked like. In tools that keep memory, ask it to save the correction so it survives past the conversation.
90 days is the number for a person. It is much faster than that for AI, but it is not one afternoon.
6. Make it mark its own work
What it is: Instead of checking every draft yourself, tell it the standard and make it check its own output before it hands anything over.
Why it matters: This is the step that changes AI from something you supervise into something that produces. Most people sit in a loop of read, correct, re-ask, read again. You can hand that loop to the machine.
You do it by giving it a rubric and a floor:
Write [the thing]. Then grade your own draft out of 100 against these criteria: [your 3 or 4 criteria]. If it scores under 90, rewrite it and grade again. Do not show me anything until it passes. Then show me the final version and the score.
It will genuinely go around 3 or 4 times. You see attempt 4 instead of attempt one.
The criteria are the whole game. "Good" is useless. "Opens with a specific number or a specific moment, no words from our banned list, under 90 words, ends with one clear action" is a rubric it can actually mark against.
7. Write it down once
What it is: Take any job you do more than about 3 times a week, write down every step of how you actually do it, and hand that document to AI.
Why it matters: The thing standing between you and automating a job is almost never the technology. It is that the process only exists in your head, in a slightly different form each time. Writing it down is the work. Once it is written, AI can read it and run against it.
The trick is granularity. "Do the invoicing" is not a process. "Open the job sheet, check the hours against the diary, flag anything over the quoted amount, draft the invoice, send it, add it to the tracker" is a process, and 4 of those 6 steps are automatable.
Pick one job. List every step, including the ones that feel too obvious to write. Then paste the list in and ask which steps can be automated or half-automated, and how.
The steps that feel too obvious to write down are usually the ones carrying all your judgement.
8. Plug it in
What it is: Connect AI to the apps you already use, so it can do things rather than just tell you about them. This is all the word "agent" really means.
Why it matters: Everything above still ends with you doing the doing. You get a great draft, then you copy it, open the other tab, paste it, adjust it, send it. Connecting the tools removes that last stretch, and it is usually the stretch that eats the afternoon.
Once your email, your calendar, your files and your store are connected, the requests change shape. Not "write me a reply to this" but "read this thread and draft the reply". Not "what should I post" but "look at what actually performed last month and tell me".
Start with one. Connect your email, then ask it to summarise the 3 things in your inbox that actually need you today. That single connection is usually the moment it stops being a novelty.
Where to start
Do not try to build all 8. They stack, and the first 2 are free.
This week: skill 1 and skill 2. Change the reflex, check the numbers.
This month: skill 3, and one pass at skill 7 on your single most repeated job.
When those feel automatic: the rest, in order.
The people who get value out of this are not the ones who understand it best. They are the ones who kept going past the first mediocre answer.
Try it on your own last answer
Open the last real thing you asked AI to do, and ask it this:
Look back at the request I gave you. Rewrite my prompt the way I should have written it: with the job, the background, the constraints, and a question back to me. Then tell me what you would have produced differently.
That one exchange teaches skills 2, 3 and 5 at the same time, on work you already care about.