Vibe Coding Has a Budget Problem

I thought the challenge with vibe coding, and the Vibing Risk project, would be getting AI to write good code.

It wasn’t. The challenge was paying for it.

I burned through my budget in a single day.

The Dream

I had a chance to play with Devin AI, which gave me my first real experience with full vibe coding. You provide it a ticket, it spins up its own environment, writes the code, and creates a PR. You review, give feedback, and just like that, the ticket is done.

It felt like hiring another developer.

I didn’t realize I was also signing up for the cost of one.

I figured the advertised $20/month wasn’t going to be the real number. But I didn’t realize how far off I was.

I started the project from scratch with Devin writing everything. I was making incredible progress for my first few hours. I was right in the middle of having Devin complete the map feature when I got a notification: I had run out of ACUs.

I was in the middle of a ticket, and it felt like Devin had gone on strike.

No more work until you buy more ACUs.

What is an ACU anyways?

Was the dream already over?

Could I really afford to continue the project if I was running out of credits this quickly?

The Pivot

This was too good to stop.

I was making real progress and could see the future sitting right in front of me. But could I afford it? There had to be a way to make this sustainable.

I had already been using IntelliJ’s AI tools to help me write code. Maybe I could switch to that instead. It wasn’t the pure vibing experience that Devin offered, but maybe I could still finish the project without writing the code myself.

IntelliJ offered three options: Junie, Claude, and Codex.

This seemed promising. I already had an IntelliJ subscription, so maybe this would solve the problem. The dream was still alive.

So I started over from scratch.

STRIKE!

At almost the exact same point in development, IntelliJ refused to write more code until I bought additional credits.

Okay. This was starting to look bad.

Then I remembered seeing an ad for free Codex.

One more try.

I signed up for the cheapest plan and started again.

The Codex interface was honestly a disappointment, but luckily IntelliJ let me use my Codex subscription through their interface.

At the end of my first full day with Codex, actually my third first day, I had finally made it through an entire day without getting cut off.

No strike. No demand for more credits.

I kept hearing that Devin and Claude were better at writing code. Maybe they were.

But Codex was the first one that let me believe I could actually finish the project.

Too Good to be True - As Expected

I had been using Codex for a few days. We had almost gotten to a playable game, and I still hadn’t written a single line of code.

It felt like this project was showing me a completely new way of building software.

STRIKE!

Yep, there it was again. But this time was different.

Codex wasn’t asking for more money. It was telling me that I was pushing it too hard and that it wasn’t going to do any more work until 6pm.

This wasn’t a full strike. It was more like:

You’re not paying me enough, so I’m taking a break.

Honestly, I could live with that, so I took a break too.

Over the next few days, I saw this happen a couple more times. The basic plan wasn’t going to cut it anymore.

So I upgraded to the Pro plan: $100/month.

Which, honestly, was much closer to what I had expected to pay from the beginning.

Vibe Coding on a Budget

The dream is still alive.

For now, I’ve managed to get past the budget problem, but I don’t think the problem is actually solved.

The pricing I’m paying today probably isn’t sustainable long term. Honestly, I’m getting far more value out of Codex than what I’m paying for.

Maybe this is just the phase where AI companies are pricing aggressively to capture the market. Maybe not.

Either way, it’s hard to deny where programming is heading.

Over the last few weeks, I’ve had more fun building software than I’ve had in years.

Not because the AI is perfect, far from it, but because tools like Codex let me spend more time thinking about design, systems, and ideas instead of fighting through every line of implementation.

That’s the part that feels different.

Not that AI can code, but that it changes what I spend my energy on as a developer.

If you want to follow along with the chaos, experiments, and progress of Vibing Risk, you can check out the project on GitHub.