Using AI is just asking questions. You're already good at that.

If you’ve ever asked a coworker for a favor, you have the only skill this takes. There are five sentences you already say without thinking. Every one of them works on an AI, exactly as spoken.

And if you haven’t actually opened one of these AI tools yet, you’re not behind. You’re normal. Being excellent at your job has never required this before now, and plenty of people who are very good at theirs are starting from zero with AI. Quietly, mostly.

The examples below are deliberately tiny and obviously fake. That’s the point: small enough to try, close enough to real work that it transfers.

LEVEL 1

One thing, one ask

The five sentences

LEVEL 2

Two sources, one question

Where the money is

LEVEL 3

The same job, every week

Asks become recipes

TRY IT RIGHT NOW · 30 SECONDS

Never opened one? You just need somewhere to type. Your company may already have an AI solution — if you use Microsoft at work, look for Copilot; if you use Google, look for Gemini. It might be sitting in a toolbar you already have open. Otherwise open ChatGPT, Claude, Copilot, or Gemini in a browser tab; the free version is plenty for this.

Wondering if you’re even allowed to use these at work? Fair question, and one for IT before your first real task. This one’s invented. No files, no data, no send button.

Paste this in:

“Write a short, friendly work email for me: I can’t make Tuesday, I’d like to propose Thursday at 2pm instead, and I’m sorry for the shuffle.”

Read what comes back. Edit one sentence if you want. That feeling of “oh, that’s all it is”? That’s the whole page.

It won’t be quite right the first time. That’s fine — revisions are free.

Asking a coworker for a third rewrite costs you something socially. Asking AI costs nothing; it never sighs. “Shorter.” “Friendlier.” “Less corporate.” “Try again as a table.” Say it the way you’d say it to a person. This is the part nobody tells beginners, and it’s the whole difference between one disappointing answer and the one you actually wanted.

LEVEL 1

One thing, one ask

You’re ready right now. Each of these takes one file or one thought, and one sentence.

DRAFT

“Can you write it up?”

You know what needs saying — you just don’t want to spend twenty minutes making it sound right. Hand over the bullet points; get back a draft you edit and send.

Doing this with real work? Swap the real names and numbers for fake ones first — it works exactly the same — then check your company’s AI policy on which tools are approved.

You give it three bullet points

can’t make Tuesday
propose Thursday 2pm instead
sorry for the hassle

At work this becomes:The delicate vendor pushbackThe policy reminder nobody wants to writeMeeting recap to the team

SUMMARIZE

“What does this actually say?”

Contracts, policies, that 14-page PDF from corporate. Somebody has to read it. It no longer has to be you — ask for the plain-language version, then spot-check the parts that matter.

Doing this with real work? Swap the real names and numbers for fake ones first — it works exactly the same — then check your company’s AI policy on which tools are approved.

You paste in the dense thing

“Perishable items introduced into the communal refrigeration unit shall bear affixed identification of ownership. Items lacking said identification, or exceeding a residency of five (5) business days, shall be subject to disposal during the weekly sanitation cycle occurring each Friday at 16:00.”

At work this becomes:The 40-page vendor contractA year of exit-interview notesThe all-hands recording

EXTRACT

“Can you pull out the details?”

The information exists — it’s just buried in a rambling email, a scanned form, or someone’s meeting notes. Ask for it as a tidy list and it comes out structured.

Doing this with real work? Swap the real names and numbers for fake ones first — it works exactly the same — then check your company’s AI policy on which tools are approved.

You give it the messy note

“hey! mtg w/ Sam moved to Tues 3pm, re: Q2 budget. bring the vendor numbers & coffee ☕”

At work this becomes:Handwritten timecards → hoursInvoices → line itemsInterview notes → scorecard

COMPARE

“Why don’t these two match?”

Every office has two lists that should agree and don’t — and checking them by hand is the job everyone avoids. This is where AI finds actual money. Hand over both lists; get back exactly what doesn’t line up.

Doing this with real work? Swap the real names and numbers for fake ones first — it works exactly the same — then check your company’s AI policy on which tools are approved.

You give it both lists

ORDERED

Apples ×10

Granola ×6

Pretzels ×12

Seltzer ×24

DELIVERED

Apples ×10

Granola ×6

Pretzels ×8

Seltzer ×24

At work this becomes:Timecards vs. agency invoicesPurchases vs. vendor price listsAssets vs. the roster

RETRIEVE

“Can you find out and get back to me?”

Prices, options, what competitors charge, how other companies handle a policy. Ask, and it comes back organized instead of as forty open browser tabs.

Doing this with real work? Swap the real names and numbers for fake ones first — it works exactly the same — then check your company’s AI policy on which tools are approved.

You ask the question

“Find me three red staplers under $15.”

At work this becomes:Vendor price sweepPolicy benchmarkingThree quotes, one table

LEVEL 2

Two sources, one question

You’re ready when you’ve tried three of the five sentences on something real and checked the answers yourself. Not yet? Go back to “Can you write it up?” — it takes thirty seconds and doesn’t need a file.

The bigger wins come from pointing AI at two workbooks at once. These are the same moves chained together, and every one hands back something you can verify against the originals.

Doing this with real work? Swap the real names and numbers for fake ones first — it works exactly the same — then check your company’s AI policy on which tools are approved.

Find the exceptions

COMPARE

YOU HAND IT: This month’s timeclock export and the payroll register.
YOU ASK: “Which people don’t match between these two?”
YOU GET BACK: A short exception list with the reason for each: on one file but not the other, hour totals that differ, a name that appears twice.

Spot the pattern

COMPARE + SUMMARIZE

YOU HAND IT: A year of weekly sales by location and the staffing schedule workbook.
YOU ASK: “Do sales dip in the weeks we ran short-staffed?”
YOU GET BACK: The pattern in plain English, with the specific weeks that prove or disprove it. Sometimes the answer is “no pattern,” and that is worth knowing too.

Plan from history

COMPARE + DRAFT

YOU HAND IT: Last year’s order history and the current inventory counts.
YOU ASK: “Build me a reorder plan for next quarter.”
YOU GET BACK: A draft plan: what to order, how much, when, plus the two items trending up that history alone would miss. You adjust it; you know things the spreadsheets don’t.

Explain the variance

COMPARE + SUMMARIZE + DRAFT

YOU HAND IT: The budget workbook and the actuals workbook.
YOU ASK: “Where are we off plan, and is it timing or a real trend?”
YOU GET BACK: A short variance story per department, with one-time surprises separated from things that will keep happening. The version your boss actually reads.

Merge the masters

COMPARE + EXTRACT

YOU HAND IT: The customer list from billing and the customer list from sales.
YOU ASK: “Make one clean list and show me every conflict.”
YOU GET BACK: One merged list, plus a conflicts tab for you to settle: same company under two names, two different addresses, the contact who left in March.

LEVEL 3

The same job, every week

You’re ready when you catch yourself typing the same ask a second time. Not there yet? Pick “Find the exceptions” above and run it once this month. You’ll be back here with a recipe worth saving.

Here’s the level-three secret: a good ask, written down, is a recipe. Save it. Next week, hand over the new files and use the same words. You’ve just written a standard operating procedure for your new assistant — and refining it a little each time is the whole skill.

The month-end packet

THE RECURRING JOB: Every month: the same four spreadsheets need reconciling and summarizing for the close.
YOUR SAVED ASK: “Compare these the usual way: match totals, flag exceptions, one-paragraph summary per file.”
WHAT CHANGES: Month two takes minutes, not the afternoon. The recipe gets sharper every close.

The Monday report

THE RECURRING JOB: Every week: last week’s numbers and notes become the status update your team expects.
YOUR SAVED ASK: “Same format as last week: wins, numbers vs. plan, blockers, next week. Keep it under a page.”
WHAT CHANGES: A consistent report that reads like you, because you edited the recipe until it did.

The action register

THE RECURRING JOB: Every meeting: decisions and to-dos scatter into notes and are never seen again.
YOUR SAVED ASK: “Add this meeting’s commitments to the running list: who, what, when due. Resurface anything overdue.”
WHAT CHANGES: One living list instead of eleven forgotten ones. The follow-up person becomes you, effortlessly.

Want this to run every week without you? That’s the next rung. I built an agent that does exactly this recipe on one command — it reads the week’s calendar and last week’s numbers, writes the briefing, and leaves it sitting in my drafts. Here’s how that went, breakages included.

The one rule that never changes

You check everything. AI drafts, flags, and finds — you decide. Nothing on this page acts without a person reading it first. That’s not a limitation; it’s why you can trust the whole arrangement.

COMING EVENTUALLY

There’s a sixth sentence: “Can you handle this every week?”

That one’s different — it’s where AI stops being a favor and starts being a teammate. I’m building toward it in the open, one project at a time. Meanwhile, at the top of this ladder: I put three frontier AIs through the same exam — three models, one judge.

That’s the whole thing: five sentences, three levels, and the rules of trust. Enough to start tomorrow morning — and the first one takes thirty seconds.

These are personal learning notes shared for education, not professional advice. Every example uses invented data. Your judgment, and your company’s policies, always come first.