I have four production apps running solo. Real users, real money moving, real 2am pages. Solo devs shipping full production apps with AI is real — that half of the hype checks out.

The other half of the hype needs a reality check.

What “production” means when I say itLink to heading

• Where Am I Serving? — real-time mission-call reveal game for LDS families. Paid events, live-sync across every device in the room, real event nights riding on it working.
• Ward Page — multi-tenant SaaS for LDS wards. Paid plan, free trial, active tenants.
• Lehi 31st Ward — my ward’s actual community site. Two years live, quietly the most code I have running anywhere.
• mx-workflow — Claude Code plugin published to the marketplace.

None are demos. All shipped from one person.

What AI actually made possibleLink to heading

The velocity claim is real. I ship features in an afternoon that would’ve taken me weeks alone. E2E test suites I’d never have written by hand — Claude writes them and I review them. Database migrations I’d have dreaded — Claude drafts the diff, I read it, we ship. Refactors that used to be six-month backburner items, done and merged before dinner.

This part is not marketing. The code half of the work compressed hard.

What AI didn’t take off my plateLink to heading

Then the reality check.

Still all me
Customer support when someone’s paid event breaks on the day of
Marketing — no agent has grown my user base
Pricing and packaging — what to charge, what to bundle, what to cut
Product taste — does this feel right for Latter-day Saint wards is a call I make
Incident response — 2am pages don’t route to Claude
Realizing I’m solving the wrong problem — the model helps me code, not think

AI makes you faster at the code half of the work. It does nothing for the parts that aren’t code.

The math that surprised meLink to heading

Being a solo with AI is more work than being a solo without AI. Not less.

The reason: you ship more, so you have more surface to maintain. Four production apps means four incident channels, four sets of customer questions, four deploy pipelines that could break, four backups I have to trust are running, four sets of dependencies quietly going out of date. Each one costs a fraction of what it would’ve cost to build five years ago. Each one costs the same as ever to run.

I traded “one app I couldn’t finish” for “four apps I can barely keep up with.” Different problem. Not obviously better.

What actually works for me right nowLink to heading

If you want real production apps solo with AI, the shape I keep coming back to:

• Pick a stack you can defend at 2am. SvelteKit + Supabase + Stripe on Cloudflare Pages for most of mine. Boring. Reliable. Every piece has a runbook.
• Build the review layer before you scale. Wrote about this a while back — the AI generates, you verify, or you’re storing up debt for future you.
• Say no to more apps until the current ones run without you touching them daily. This is the one I keep failing.

The honest ceilingLink to heading

Fully-fledged apps at real scale, shipped by one person? Yes. Multiple of them at once without burning out? That’s the ceiling nobody’s talking about honestly.

The gap between “possible to build” and “possible to run” has never been wider. AI closed the build gap. The run gap is still yours.

That’s the reality check I wish someone had written for me a year ago.