“Vibe Coding” ain’t it

At MoonPay, we like to keep our finger on the pulse of the latest software trends, and this of course includes developments in AI. Some of the questions we get asked include:

• Is software development now easy?
• Will developers be replaced by non-coders?
• Will AI make coding obsolete?

No, no, and no. The hype will fade, and the real work of weaving these tools into everyday workflows will take center stage.

In this article, I explore why software engineering isn’t going away any time soon.

Delusions of grandeur

The project manager who creates their first app with ease may think they’ve just unlocked the tools to create whatever their heart desires.

But as they’ll quickly find out, they’re wrong. Just as the proud dad who finishes his daughter’s tree house doesn’t believe he can build a skyscraper, the budding software engineer who can vibe code a Flappy Bird clone in 15 minutes shouldn’t believe they can develop sophisticated software. There are levels to this game.

Ask yourself: Where’s the breakout company whose software has been built, deployed and maintained exclusively with AI, and scaled to a degree that anyone cares about? I haven’t seen it, and I’d be willing to bet you haven’t either.

AI can accelerate the process, but it doesn't replace the craft.

Close enough to miss the mark

For the right task, AI coding can work like magic. It can add buttons, write tests, create database tables—all with requests in plain English.

To the uninitiated, this seems like having an obedient developer at one’s beck and call, but we developers can see it for what it is: these models have been trained indiscriminately on every right and wrong answer, every spaghetti code example from the original fountain of coding knowledge.

The truth is, AI can’t do everything well. For tasks such as novel coding problems, it is especially ill-suited. For resolving edge-case bugs in real-time, it misses subtle nuances. For designing scalable system architecture from scratch, it lacks foresight.

AI is magic where it works well, but it stumbles where the path isn’t already paved.

“Good enough” is never great

Part of the fun of these large language models is that they are in some sense randomized. You can prove this easily by repeatedly giving the AI the same prompt and watch it produce a different output each time. That’s partly why it can give you an endless stream of novel answers to your deep and meaningful questions about life and the universe. This randomness is a feature, not a bug.

But in finance, especially crypto, “good enough” is never really good enough, and it’s certainly never great. The reality of AI software development at the enterprise level is that you’re still writing the code, you’re just using the far less precise language of AI prompts to do it. It’s like trying to type with your feet—with enough time you’ll get better at it, but it’s still not optimal.

You can’t rely on AI for exceptional results.

To the moon

As the rest of the world prepares for a future without software engineers, we here at MoonPay have a different perspective. We’re constantly looking to expand the boundaries of what’s thought possible in crypto and elsewhere. We've already hosted our first AI-themed hackathon, and we're encouraging our teams to test ideas before sharing their results more widely.

Ultimately we want to use AI not to solve our engineering problems, but to aid us in building impactful, cutting-edge products.

If you’re excited by a future where AI amplifies the developer experience, MoonPay is the place for you. Join us as we push boundaries together–we’re hiring!