Vibe Coding Is Fun Until Production Shows Up

I get why vibe coding took off.

Karpathy gave it a name, the internet grabbed it, and now half the timeline acts like software is just vibes, prompt history, and a deploy button.

Want to see your calendar in a new way? Great. Want a tool your family can use to manage school pickup and drop-off? Also great. Want a tablet app that controls the lights in your house like you are living in a slightly underfunded sci-fi movie? Incredible.

Small apps can be world changing for the people using them. That part is real.

But we need to stop pretending that "I got an LLM to make something passable" is the same thing as learning how software works.

The Good Part Everyone Is Right About

For simple personal tools, AI is honestly great.

• CRUD screens
• quick forms
• dashboard scaffolding
• glue code
• "please make this annoying manual process go away"

If you want to spin up a weird little app for your own life, I am fully in favor of that future.

The problem is when someone builds one shiny weekend demo and immediately starts posting like software engineering is dead and the only thing standing between them and a startup empire was not having enough prompt credits.

Useful Is Not The Same As Durable

This is the part the hype keeps skipping.

A tool that works for you, your spouse, and maybe your one organized friend is not the same thing as a product that works for 5,000 people who all forgot their passwords on the same Monday morning.

It is very easy right now to generate software that looks fine from ten feet away. Kind of like a lot of AI content, honestly. Smooth UI. Nice buttons. A few charts. Maybe even a Stripe checkout duct-taped on with pure confidence.

Then you pop the hood and it is a haunted house.

• no real auth model
• no permissions story
• no migration plan
• no rate limiting
• no logs worth reading
• no backups
• no clue what happens when multiple users touch the same data
• absolutely no plan for "what if people actually use this"

That is not "the future of engineering." That is a demo with flattering lighting.

The Part That Should Worry People

What bothers me is not that people use AI to code. I use AI tools all the time.

What bothers me is watching newer engineers start to treat code like a slot machine.

Pull lever. Get component.

Pull lever. Get API route.

Pull lever. Get deployment config.

Never read it. Never question it. Never learn why it broke.

That is where the brainrot kicks in.

If you never learn how state works, how auth fails, why queries slow down, how systems get abused, or what makes software secure, then you are not leveling up. You are just becoming dependent on autocomplete with better branding.

AI should multiply skill. It should not replace having any.

Vibe Coding Has a Ceiling

This is not gatekeeping. It is just scope.

You can absolutely vibe code your way into a personal app that genuinely improves your life. You probably should not vibe code your way into healthcare, fintech, or enterprise software and then act surprised when the bill comes due in security holes, scaling issues, and late-night incident calls.

The hard part of software was never just typing syntax.

It is making decisions that still hold up later:

• data models
• access control
• observability
• failure handling
• disaster recovery
• maintenance
• cost
• handoff to the next person who has to touch this thing

None of that gets solved just because the generated UI looked pretty decent on the first try.

Companies Keep Hitting The Same AI Wall

This is also why so many companies that want to "go AI" keep face-planting into the same wall.

They are not blocked by the fact that AI cannot generate things. It can. The blocker is that nobody slowed down long enough to answer the adult questions:

• what problem are we actually solving
• who is this for
• what data does it need
• what happens when it is wrong
• how do we measure whether it is helping
• where does a human stay in the loop
• should this even exist in the first place

That is the wall.

Just because we now have the ability to bring things to life faster does not mean we should build every possible AI feature that wanders into a strategy deck.

Half the time "we need AI" is really "we need better workflows, cleaner data, and someone willing to say no to a bad idea before it becomes a roadmap item."

And that is really the next conversation: once building gets cheaper, product judgment gets more expensive. Why that makes product management and vision more important than ever is its own topic, and honestly, probably the more important one. See When Building Gets Cheap, Product Matters More.

The Most XKCD Outcome Possible

Also, if we are being honest, a lot of vibe coding has strong old-school XKCD automation energy.

"I am $1,000 in tokens and 8 weeks into building my own app so I can avoid a $4.99 a month iPhone subscription."

That is not disruption. That is a side quest with a revenge budget.

And I say that with love, because I am exactly the type of person who would absolutely do this, justify it as "saving money," and then accidentally build half a backend for no reason.

To be clear, there is nothing wrong with that kind of curiosity. Building weird little apps, taking chances, and growing your skills is valuable.

But let us calm down a bit. You are learning. You are experimenting. You are not one prompt away from kicking off the SaaS apocalypse and taking down Facebook with your new app.

My Actual Take

Use AI to build the little calendar tool. Use it to make the school pickup app. Use it to build the light-control dashboard. Use it to prototype faster and remove friction.

Just do not confuse "I made it exist" with "I made it good."

And definitely do not confuse "the AI gave me code" with "I now understand software engineering."

Those are not the same milestone.

The best developers I know are not valuable because they type fast. They are valuable because they know the difference between a fun shortcut and a future incident report.

Closing

Vibe coding is real. It is useful. It is not fake.

But a passable app is not the same thing as a durable system, and AI slop does not magically become architecture just because it compiled on the first try.

Build the app. Ship the toy. Learn from it.

Just do not let the demo fool you into thinking the hard parts disappeared. And no, your favorite X influencer didn't vibe code a working quantitive Polymarket bot in a weekend making thousands of dollars a day in profit.

Written from home, while trying very hard not to spend six weeks rebuilding a perfectly good five-dollar app.