Don't bother with vibe coding

Vibe coding is everywhere nowadays, ever since Andrej Kaparthy from OpenAI coined the term on X. Silicon Valley is already jumping on the trend and hiring Vibe Coders. As with a lot of things AI, I think it's a trend that's actively harmful. I'd even go as far as saying that you should reject anyone applying for your startup if they claim to have vibe coding experience.

Why? Because it'll cause you to lose a lot of velocity, increase tech debt and burn a lot of your money in the long run. These are all things a startup cannot afford, especially early-stage.

Overwhelmingly, vibe coding is currently embraced by those with little to no experience in professional software development. When I say professional, I'm explicitly not talking about coding. This is easy to learn, and is something that AI is actually somewhat good at (depending on prompting, language and framework). I'm talking about things surrounding the code:

• A good CI/CD setup, mindful of security and automation
• Good documentation
• Knowledge about infrastructure needs
• Unit and integration tests
• Deep understanding of the code

Early-stage, every engineer should have a deep understanding of these four points. Enterprise teams may have the luxury of having separate teams, but early stage startups do not. I've explicitly highlighted the last point, because the entire premise of vibe coding is that you do not need to understand the code at all. This also means that vibe coders won't be able to document their code or set up independent tests, let alone have any knowledge about the surrounding infrastructure needs and CI/CD setup.

As an aside: What I mean by independent tests is that letting the AI set up the unit or integration tests for its own code somewhat defeats the purpose of having tests in the first place.

Vibe coding may give you an initial velocity boost when everything works seamlessly, but when it comes to "boring work" such as documentation and testing, it's impossible to keep up the same kind of velocity if there's no deeper understanding of the code.

That's not to say AI is useless. I use AI daily, whether it's for repetitive work in coding or otherwise running my business. I've also sometimes let AI do quite heavy lifting; for example, the frontend for my latest weekend project, SquadUtils, was done almost entirely by Claude. It's important to note, though, that I was always able to reason about the code and guide Claude in the correct direction if it made mistakes.

And that brings me to the main point: AI can bring a net velocity increase to your team if it's employed by professional software engineers with years of experience building products, precisely because they can tell when and where the AI made a mistake. Put it into the hands of someone that just graduated a coding bootcamp two weeks ago, though, and you'll pay the price down the line once the initial velocity wears of.