I'm not a developer. I run marketing for a living.
But like everyone else this year, I opened Lovable one night, described an app, and watched it appear on my screen in four minutes. Working. Clickable. Mine.
So I built another. And another. Over a few months I built twelve — a pricing calculator, an internal dashboard, a little tool for a client, a landing page, a thing that scratched a personal itch I'd had for years.
All twelve are dead. And here's the part that took me embarrassingly long to notice: they all died in exactly the same place.
The graveyard
Let me walk you through it, because the pattern is the whole point.
App #1 — a pricing calculator (Lovable). Beautiful. Worked perfectly in the preview. I hit "Publish" and got a something.lovable.app URL. I needed it on my own domain, in front of a client, looking like it belonged to me. That last 5% was a wall. The app never left the preview.
App #3 — an internal dashboard (Bolt). Ran great until it needed an actual backend and environment variables. I spent a Saturday in config files I didn't understand, copying error messages into three different chatbots. Sunday night I closed the tab.
App #5 — a Claude artifact. Genuinely the cleanest thing I'd ever made. It lived inside a chat window. I could not send anyone a link that survived past the conversation. It was real and unreachable at the same time.
App #8 — a landing page (v0). It handed me a Next.js repo. I'm a marketer. You might as well have handed me a jet engine and wished me luck.
I'll stop there. You get it. Twelve different ideas, twelve different tools, one identical cause of death.
The thing nobody tells you
Every AI builder sells you the same magic: describe it, and it exists.
And that part is true now. Building is solved. For people like me — people who can't write the code but know exactly what they want — the hard part of software just evaporated.
Except "it exists" and "it's live" are two completely different things, and the entire industry quietly lets you confuse them.
"It exists" means it works on your screen.
"It's live" means a stranger can open a URL, on your domain, and it still works tomorrow.
The gap between those two sentences is where all twelve of my apps died. It's deployment, hosting, domains, environment variables, the stuff that has a hundred years of accumulated developer assumptions baked into it. AI got me from zero to a working app. Then it dropped me at the foot of a wall it never mentioned existed, and the wall was built for someone with a CS degree.
That's the cruel joke of vibe coding right now. The tools democratized the creative half and left the boring, load-bearing half exactly as gatekept as it always was.
What changed
I'd love to tell you I learned DevOps. I didn't. I have a job.
What actually happened is I got annoyed enough to go looking for the missing step — the thing that takes "works on my screen" and turns it into "here's a link" without asking me to understand a single config file. I ended up close enough to the people solving it that I became, more or less, user zero of the fix.
I'm not going to pitch you here. dev.to can smell a pitch from three scrolls away, and honestly the product matters less than the realization, which is this:
If you're building with AI and your apps keep dying, you are probably not bad at building. You're hitting the one step the demos skip. Naming that wall out loud changed how I work more than any new model did. I stopped treating "it runs" as the finish line. The finish line is a URL someone else can open.
Eleven of my twelve apps are still dead. I'm fine with it — they were practice. The twelfth one is live, on my domain, and someone I've never met used it last week.
That's the only version of "done" that counts.
If you've vibe-coded something that's stuck in a preview right now — what tool was it, and where exactly did it stall? I'm collecting these. The death points are weirdly consistent and I don't think it's a coincidence.
Top comments (1)
You didn't fail. The tools failed to close the loop. The fix is choosing tools that treat deployment as step 1, not step 99, and refusing to accept a preview URL as "done."
Your 13th app should take 4 minutes to build and 10 minutes to ship. That's the standard now.