You're using AI wrong. You're still the one working.
If you're still typing prompts and doing the work, you're using a 2026 tool with a 2023 mental model. Here's what the real shift looks like.
Here's the thing nobody tells you about AI.
If you're still typing prompts, reading answers, and doing the work — you've missed it. You're using a 2026 tool with a 2023 mental model.
The shift isn't that AI helps you work faster. The shift is that AI does the work, and you don't.
You direct. You decide. You check. The doing happens without you.
Most people reading this have never actually seen what that looks like. ChatGPT doesn't show you. Your colleagues don't show you. LinkedIn shows you screenshots that look like magic but don't translate into anything you could do on Monday.
So here are three real examples. Different shapes of work. Different tools. Same underlying move: set it up once, it runs without you, you do something else.
Example 1: Finding new customers (Manus)
What I used to do
Block half a day. Open LinkedIn Sales Navigator. Type filters. Scroll. Copy names into a spreadsheet. Open each company's website. Find the right person. Try to find their email. Guess what they care about right now. Write the outreach. Send. Repeat 30 times. Hate myself by the end. Skip it next month.
What I set up
A Manus run that finds Swiss KMUs in specific industries, identifies the L&D or HR decision-maker, finds their verified email, and pulls three signals about what's currently keeping them busy — recent press, hiring activity, content they've posted.
I wrote the brief once. Took an afternoon to get it sharp.
What happens now
I press start. I close the laptop. Two hours later, I have a list of 40 prospects with names, verified emails, current context, and a draft opening line per person. I review it over coffee, kill the obvious mismatches, send the good ones.
What I do instead
I think about who I actually want to work with. I write the angle. I make the calls. The grunt work that used to eat half my day is gone — not faster, gone.
Example 2: The weekly content loop (Claude)
What I used to do
Sit down on a Sunday with a vague feeling I should "post something this week." Stare at LinkedIn. Write something polished and lifeless. Hate it. Don't post. Or post and get six likes from people who like everything.
What I set up
A Claude Project that knows my voice, my frameworks, my client examples, my no-go phrases. Connected to my calendar and my email so it sees what's actually been happening that week. I dump voice notes into it whenever something in my work makes me think.
What happens now
Friday morning, I open the project. It already knows what I worked on, who I talked to, what I noticed. I tell it the half-thought I had on Tuesday on the train. Forty seconds later it gives me three angles for a post, in my voice, with the specific example I mentioned. I pick one, sharpen the first line, post it.
Total time: under ten minutes.
What I do instead
I think. I notice things during the week and let them register, knowing I won't lose them. The week becomes the source material. The writing isn't the work anymore.
Example 3: A task you'd normally just do yourself (Cowork)
This one is different. Manus does research. Claude does thinking. Cowork does tasks — the kind of thing you'd open a browser and do yourself.
What I used to do
Get an email asking me to send my CV, the company info, three references, and a one-pager about the program I run, formatted to match their template. Forty-five minutes minimum. Half of it spent finding files I knew existed somewhere.
What I set up
I told Cowork once where my files live, what my standard cover note looks like, and how I want references formatted.
What happens now
I forward the email to Cowork with one sentence: "Reply to this with the standard package, adapted to their template." It opens the email, finds the files, formats the response, drafts the reply, and waits for me to send.
I check it. I send. Done in ninety seconds.
What I do instead
I'm not at the computer. I'm in a call, or on the way to one. The task ran while I was elsewhere.
The repetition is the whole point
Notice the pattern in all three: the setup is the work. After that, it runs.
This is the part most people miss when they think about AI. They imagine the dramatic one-shot — the perfect email written in three seconds. That's not where the value is.
The value is that the email gets written next week too. And the week after. And the month after.
Doing something once with AI saves you twenty minutes. Setting something up so AI does it every week saves you twenty minutes a week, forever, while you're somewhere else.
The first one feels impressive. The second one changes your life.
What to actually do this week
Don't go build ten agents. Pick one task. The kind that:
- You do regularly (weekly is ideal) - You don't enjoy - Has a clear input and a clear output - You've avoided automating because it felt fiddly
Spend an afternoon setting it up. Not perfectly — just well enough that it produces something you can review and ship.
Then run it next week. And the week after. And the week after.
That's the whole game.
The real shift
The mistake isn't using AI badly. The mistake is using it at all in the way you currently do.
If you're still the one doing the work, you're not using AI. You're using autocomplete with a personality.
The actual shift is uncomfortable, because it means letting go. Trusting an agent to do something you used to control. Reviewing instead of producing. Directing instead of typing.
That's the discomfort that separates people who get leverage from people who keep nodding at LinkedIn posts about leverage.
If AI isn't doing it for you, it's probably not AI.
You're still doing it.