AI Is a Lever, Not Magic: Why Your Domain Expertise Is the Real Fulcrum

My friend burned half a year of tokens and still shipped nothing

A friend texted me a while back, complaining he'd blown another three grand on tokens that month. I asked him: so where's the thing?

He hemmed and hawed, said it's still in progress. Except that progress had already dragged on for the better part of six months.

What he was doing has a name now: vibe coding. Plainly, it means letting AI write the code end to end while you just describe what you want and never touch how it's built. Andrej Karpathy put the term out on X on February 2nd; Collins later picked it as their 2025 word of the year, with search interest jumping around sixty-seven hundred percent, and Merriam-Webster added it that March. My friend bought the story — figured if he could describe a product, he could ship one.

And then? Six months in, not even a working demo. The money was real. The product wasn't.

Why his code fell apart the moment pieces met

I peeked at his project once. Pull a single module out and the AI wrote it decently — features checked. Shove them together and the seams split: endpoints didn't line up, data didn't flow, the page threw errors on the first click.

The truth is he never learned how software actually gets built. When something broke he had no idea where to even start looking.

One time I asked him: who is this thing for, and what problem does it solve? He stared at the screen and said nothing. Without that judgment — knowing who it's for — AI just spins in place and hands back confident nonsense. He'd never really decided who the thing was for.

A former colleague: same tools, two finished pieces a day

Different story. A woman I used to manage ran a mom-and-baby account. Same AI tools. She put out two solid posts a day.

She didn't dump the work on the model. She'd internalized eight years in that space and knew what mothers actually lie awake worrying about. AI drafted; she rewrote the headlines and tuned the tone, because she knew which headlines would clear the platform's bar.

Two a day, both publishable. The difference was the judgment already in her head — AI just finished the labor.

The guy who worked in a pet store shipped a paid scheduling system in a week

Another person: he'd worked in a pet store. With AI he cobbled together a scheduling system in a week, and people actually paid for it.

He knew the pain of staffing — who covers weekends, how to shuffle when someone calls in. Three shops ran on his system; one paid him five hundred.

He never wrote a line of code, but he knew staffing cold. The system that earned money was AI pasting over the technical gap for him.

Stories over. What's sitting in your own head worth amplifying?

AI, to me, is one thing: a lever. The fulcrum isn't in the tool. It's in the years of domain sense you've stacked up.

To find your own fulcrum, three questions:

1. Which field do I judge better than most people around me?
2. What mistakes have I made that others haven't, and what did they teach?
3. What need do I see that others walk straight past?

Don't tell yourself your experience is too small to count. I've watched people build real businesses on custom bedtime stories for kids, or localizing indie games into obscure languages. Niche sounds lonely. The loop still closes.

Found the fulcrum? Next step is the tool. A lever with no fulcrum under it just lies there.

However long the lever, your accumulation is the fulcrum

So don't rush to buy the priciest plan first. Write down your answers to those three questions. See what's actually in your head. Once there's something there, let AI extend your lever.

Writing this, I remembered the noodles at lunch were way too salty. Drank half a kettle.

I copied those three questions onto a sticky note by the monitor. The friend who burned the tokens just sent another voice message. I haven't opened it.

The friend, colleague, and pet-store worker here are anonymized stand-ins for people I know; names and details are altered, not real identities. One reminder: never paste client contracts, user data, or internal company material into an AI.

Xiao Zhi. I've long watched how AI tools change the way ordinary people work. On this blog I write about turning technology into real skill.