Ask first.
The one habit that quietly makes you better at AI than almost everyone. No coding required.
Here is something I see constantly.
Smart, capable people, good at their actual jobs, open an AI tool, stare at the empty box, type one careful sentence, read something generic, and quietly decide AI is overhyped.
The tool was never the problem. The empty box was.
Most people open Claude, type a question, read the answer, and close the tab. That is about a tenth of what it can actually do.
I have spent 20 years in tech, and right now I lead a few of the teams building AI models at Google. I am in this every single day. And the people who get the most out of it almost never know more than you. They just do one thing differently.
This guide is that one thing, plus the six places it pays off. By the end you will be able to talk to your computer, turn a messy folder into a finished spreadsheet, let Claude read your email and calendar before it answers, teach it to work your way, and build a working web page without writing a line of code.
Two small asks before you start:
Save this and block 15 minutes this week to actually try it.
Send it to anyone who still says they are “behind” on AI.
Before anything: set up Claude (skip if you already use it).
Go to claude.ai, or download the desktop app at claude.ai/download.
In the box where you type, bottom-right corner, pick the model. Free runs on Sonnet. The smarter paid one is Opus, on the $20/month plan.
Pick Opus, turn on Extended thinking, and set it to high.
That is your setup done. The thinking setting matters more than people expect. On a hard task, it is the difference between a fast guess and a careful answer.
I - The habit: make Claude interview you.
Whenever you ask Claude to do something, add one line at the end.
“Ask me questions first.”
That is the whole trick. Instead of you guessing the perfect prompt, Claude turns around and interviews you. Three to five questions. You answer in a sentence each, or click the options it gives you. Then it builds the thing around your answers instead of a generic guess.
Most people do the opposite. When the first answer is weak, they write a longer, more detailed prompt, then an even longer one. They try to predict every instruction up front. It rarely works, because you cannot know what the model is missing until it tells you.
Try it on something you actually need. Paste this:
Help me write a one-page proposal for a new client. Ask me questions first.
It comes back asking about the client, the budget, the deadline, what you are pitching. A minute later you have a proposal built on your real situation.
Why this matters: a blank prompt forces the model to assume. Its assumptions and yours rarely match, so you get something close-but-wrong and spend twenty minutes fixing it. When it asks first, that gap closes before it writes a word.
Everything below is the same line, pointed at something bigger.
II - Stop typing. Talk.
You speak about four times faster than you type. And most of us think out loud better than we write.
The real reason this matters is not speed. When you type, you stay terse. You write “draft a client update” and stop. When you talk, you ramble in the good way: you mention the client is nervous, the project slipped a week, you want to sound calm but honest. All of that context lands in the request without any effort.
You do not need an extra tool. Claude has a microphone button built into the same box you type in.
Click the mic.
Say what you want out loud.
Add “ask me questions first.”
Now you are having a real back-and-forth at the speed you think, not the speed you type. The friction of typing is the reason most people keep their requests short and vague. Talking removes it. And more context is the whole game.
III - Real files, not just answers.
This is where most people stall. The chat answers you on screen, and then you are stuck copying it somewhere useful.
Open the desktop app and click the Cowork tab, at the top next to Chat.
A normal chat talks back to you. Cowork makes the actual file and saves it on your computer.
The mistake people make here is treating Cowork like a smarter chat. It is not. It works inside a folder on your computer, and the folder is the point. Whatever you put in there, it can read. Whatever it makes, it saves back.
The setup:
On your desktop, right-click and make a new folder. Name it anything.
In Cowork, click Work in a project and pick that folder.
Drop in whatever you have. A messy export, last month’s numbers, a half-finished doc.
Then use the habit again:
Build me a spreadsheet to track my monthly expenses. Ask me what you need first.
It asks how you sort your costs, what you want totaled, how you want it laid out. A couple of minutes later there is a clean spreadsheet sitting in your folder. Every cell editable.
The same trick works in reverse. Drop a messy three-page document in the folder and ask it to turn it into a clean one-page summary. You get a real file back, not a wall of text to copy out by hand.
IV - Give it your world.
This is the one that makes Claude feel like it knows you.
So far you have fed it everything by hand. Connectors let it reach into the apps you already use. In the app: Customize, then Connectors. Link your Gmail, your calendar, and your meeting notes, whether that is a tool like Granola or the notes Gemini saves to your Drive after a Google Meet call.
One honest note, because most people skip it: connect the ones you will actually use, and know you are letting Claude read them. For personal and small-business work, that trade is fine. Just make it on purpose.
Now you can ask things you could never ask a blank chat:
Go through my client meetings from this week and draft the follow-up emails. Ask me questions first.
It reads the transcripts, checks your calendar, asks a couple of things to get the tone right, and hands you drafts you just review and send. Want options? Ask for three versions and pick one.
Think about what that replaces. The thirty minutes after every call where you try to remember what you promised and write it up. Now the notes, the calendar, and the first draft are in one place, and the draft is already written.
V - Teach it your standards.
Connectors give Claude your context. This next one gives it your standards.
A skill is a saved playbook. You teach Claude how you like something done once, and it does it that way every time after.
Think of a new hire who learns your procedure and then never forgets it.
In Customize, then Skills, there is one built in called skill-creator. Type slash skill-creator and describe what you want it to learn:
Build my spreadsheets the way I like them: dark header row, subtotal lines, a summary tab. Ask me questions first.
It walks you through the details and saves the whole thing under a short name, say slash expenses. After that, you type that name at the start of any request and it runs your entire playbook.
Spreadsheets are just the easy example. The real payoff is anything you do the same way every week: the way you write a client update, the format of your weekly report, the tone of your cold emails. Build the skill once, and you stop re-explaining yourself.
So you explain yourself well one time. After that, it is one word.
VI - Build the thing.
You do not know how to code. Fine. Neither do most of the people building useful things with this right now.
In the app there is a Code tab. Turn on the setting that lets it work without stopping to ask permission at every step. Then give it the same kind of request as everything else:
Build me a page where people can book a 30-minute call with me. Ask me questions first.
It asks whether you want it dark or light, whether it should connect to your calendar or just collect emails, and what fields go on the form. You click through. It builds the form, the buttons, the layout.
A few minutes later it hands you a working page. Want it on your phone? Ask for that too, in plain English.
A booking page is the demo, not the ceiling. The same approach builds a simple internal tool: a form that sorts incoming requests, a small dashboard for your numbers, a one-page calculator for your pricing. The kind of thing you would normally pay someone to build, or just never get around to.
Your first 15 minutes.
Reading this does nothing. Doing it does. So block 15 minutes this week and run this:
Minutes 0 to 3: open Claude, pick Opus, turn on thinking.
Minutes 3 to 8: take one real task you have this week. Type it or say it, and end with “ask me questions first.” Answer the questions. Watch what comes back.
Minutes 8 to 12: download the desktop app, open Cowork, point it at a new folder, and ask it to build one real file.
Minutes 12 to 15: open Connectors and link one app you actually use. Just one.
You do not need all six on day one. You need the first one today.
The honest limit.
I am not going to oversell this.
It will not make you a software engineer. It will not run a system for a million users or carry a complex product a real engineering team should own.
What it covers is most of what you hit in a normal week: the reports, the spreadsheets, the small internal tools, the first version of an idea you want to test. Start there. The harder stuff has a ceiling, and you are nowhere near it.
I do not pick sides between AI models. I am around all of them for work, and I am sharing what actually changes how I get things done.
Here is the part that outlasts any one app.
The buttons I mention to you are Claude’s. The habit underneath is not. Asking the AI to question you first, handing it your context, teaching it your standards, that works just as well in Gemini or ChatGPT. The menus differ. The habit does not.
So you did not learn one tool. You learned how to work with any of them.
Most people who feel behind never actually tried. You read this far, which means you are already past them.
See you next week,
Guney