Vibe coding at the edge of superintelligence

Here’s a thought experiment we like to try on people. Imagine you could describe any piece of software to a machine and get it back instantly, and correct. No loop, no bugs, no “it worked a minute ago”. What would you build?

Almost everyone goes quiet. Then, after a pause, they name something surprisingly small: a better way to plan the family holiday, an app for their running club, a tool that does the boring half of their job.

That pause is the whole future of vibe coding in miniature. When building stops being hard, deciding becomes the work.

The gap is closing from one side

Every new generation of models asks less of the human in the loop. Not long ago, AI could reliably write a function and you stitched the pieces together. Then it could build a screen. Now agents work through whole codebases for long stretches, plan their own steps, run the tests and fix what they broke.

Draw that line forward and you don’t need to believe in any particular date for superintelligence to see where it points. The mechanical part of software, turning a clear intention into working code, is heading towards free.

What doesn’t get automated (as quickly)

Some things get harder to automate the closer you look, because they aren’t really about code at all:

• Knowing what’s worth building. Ideas are infinite. Problems people will pay to have solved are not.
• Knowing who it’s for. Software is a guess about a person. Somebody has to understand the person.
• Taste. The difference between an app people tolerate and one they love is a thousand small calls about what to leave out.
• Saying no. The most valuable feature is often the one you didn’t add.
• Being accountable. When it goes wrong, someone has to own it. Machines can’t sign a contract or apologise to a customer and mean it.

When anyone can build anything, the scarce thing is knowing what’s worth building.

The brief becomes the source code

The best vibe coders already write like good product managers: clear, specific, honest about what’s out of scope. That’s not a coincidence. As the models improve, the description is increasingly the thing you’re really making, and the code is just what the description compiles into.

This is good news for anyone who can think clearly and explain themselves well, whether or not they ever learned to program. Writing a great brief is a learnable skill, and every project you ship is practice.

Trust becomes the bottleneck

Here’s a problem that gets bigger, not smaller, as AI gets better. If machines can write software faster than any person can read it, how does anyone know it does what it claims, and nothing else?

Part of the answer is technical: better automated testing, AI checking AI, and formal proofs for the things that really matter. Part of it is very old-fashioned: track record. Who built this? What else have they built? Did those things hold up?

That’s a big part of why Vibers exists. A public, permanent record of real things people have shipped, how they built them, and what the community thought, is the kind of trust that’s hard to fake and gets more valuable as everything else gets easier to generate.

What about the jobs?

Nobody honestly knows, and you should be suspicious of anyone who sounds certain. But a few things seem likely.

There will be fewer jobs that consist only of turning a spec into code. There will be many more people building software who never would have before: the café owner, the teacher, the nurse with an idea for a better rota. And the world’s appetite for software is nowhere near met. Every small business, club and niche community has problems no software company will ever bother to solve.

There’s a precedent. Spreadsheets wiped out a great deal of manual bookkeeping, and the people who learned to use them well became more valuable, not less. The work moved up a level. That’s the most likely shape of this, too, just faster.

About the .si

Yes, .si is Slovenia’s domain, and we have nothing against Slovenia (the lakes look lovely). But lately people have started reading .si another way: super intelligence.

Vibers is a bet on that reading. As the machines get closer to it, the interesting part of building moves to the humans doing the steering, and those people deserve a home. Allegedly.

What to do about it now

• Ship things. A record of good judgement builds up over time in a way that no single clever prompt ever will.
• Practise writing briefs. Treat every prompt as a small exercise in saying exactly what you mean.
• Learn to check, not just to build. Skim the changes. Test your own app as if you were trying to break it. Ask “how do we know this works?”
• Stay curious rather than anxious. The people who do best in every big shift are the ones who started playing with it early.

If you’ve never tried any of this, what is vibe coding is the place to start. If you have, the challenges is a good excuse to stretch.