Has vibe coding actually gotten easier?

A few months ago, I was catching up with a PM friend in San Francisco and telling him about all the AI coding tools I'd been experimenting with.

His response surprised me:

"I'm just going to wait until it's all solved."

His thinking was that AI is improving so quickly that eventually, he'll be able to type an idea and have a fully functioning app appear.

Meanwhile, I've spent the last two years jumping between tools, experimenting with coding agents, and trying to keep up with every new development.

And it got me thinking: Has vibe coding actually gotten easier?

In some ways, absolutely. Getting an app to 70% has never been faster. But getting that last 30% right? That's a different story.

What started as prompting an LLM for code has evolved into managing agents, context, loops, workflows, and increasingly complex development environments.

Earlier this year, I tried the opusfived.dev  coding challenge that simply asked me to change an "Add to Cart" button to blue.

Thirty minutes of prompting Claude later, I'd somehow ended up with two buttons, gradients, and a lengthy discussion about design tokens. 😂

It made me wonder whether we're confusing more powerful tools with more accessible tools.

As a non-engineer building my own product, I think there are three things that would make AI coding genuinely accessible:

1. Speed to launch: Can I get something into production quickly?

2. Control: Can I actually build what I envisioned, down to the details?

3. Iteration: Can I keep improving it without breaking everything else?

We've made extraordinary progress on the first. I'm less convinced about the other two.

I've been thinking a lot about this while building Ableten, particularly around the idea of malleable software — products where users can directly manipulate and customize what AI generates, rather than relying on an endless cycle of prompting.

Curious what other makers think:

Has vibe coding actually gotten easier for you over the past year? Or have we just traded writing code for managing increasingly complicated AI systems?

And do you think the future is better AI agents, or better ways for humans to control what those agents create?