Will Vibe Coding End Engineering as We Know It? I Don’t Think So

I’ve had some unusual conversations with prospective clients over the past few months. They’ve all culminated more or less like this:

“Since you’re now using AI tools to help code, this project could be done for half the price, right?”

If you were thinking of hitting your development agency with that line, let me pre-emptively advise you not to.

It’s 100% true that AI coding has created a perceived devaluation in coding. However, this is mostly among people who don’t code. At Appstem, instead of losing business to supposed vibe-coders, we’ve actually seen an uptick in requests from enterprise clients. Most of them are to help them build out internal AI capabilities.

So why does there seem to be such a disconnect between what the public thinks vibe coding is doing to the development industry and the reality? I want to talk about this a bit more in-depth and tell you what I’m seeing as the future of development as the owner of a small, bootstrapped, enterprise-grade development agency.

Briefly, What is Vibe Coding?

Vibe coding, if you’re unfamiliar, is the act of coding in languages or tools with which one lacks experience using a language Learning Model like ChaGPT or Cursor.

The vibe coder simply writes out how they want the app to function, and the LLM spits out the requisite code in the language that was requested.

The popularization of the trend is largely attributed to one cofounder of OpenAI, Andrej Karpathy, a machine-learning expert and former head of AI at Tesla. He's frequently taken to LinkedIn and X.com to express his thoughts on vibe coding as the future of engineering.

However, Andrej isn't the first or only vibe code evangelist. If you were lurking around on X (formerly Twitter) in late 2024, you’ll remember that famous online entrepreneur Pieter Levels also pushed vibe coding by “vibe coding” a video game that eventually earned him thousands of dollars every month in advertising revenue.

There’s now a sea of evangelists throwing up hastily built apps and bragging about their six figure MRRs gained. This has in turn pushed every aspiring developer in America to clock in after hours at their day jobs to build their own hastily-coded app to varying degrees of success and failure.

It's been a net good for society in terms of collective growth. Thousands of interesting product ideas are actually getting coded instead of dying in the cradle. However, it has caused some trends that should make any enterprise skeptical.

Why Top Vibe Coders Are Misleading You

I want to draw your eyes to the one thing that connects all the original and most successful proponents of the vibe code movement: They are all already experienced developers.

Why can a Pieter Levels or an Andrej Karpathy make well-functioning applications in coding languages they’ve never practiced using AI? It's because they’ve already put the time into understanding the fundamentals of coding.

They know how a functioning application is supposed to work, which security features to include, and how to architect it so that other developers can easily build on top of it.

While I don’t want to discount what can be accomplished with the help of these tools, I’m still strongly convinced the next generation SaaS tools will need more than a non-technical project manager with a ChatGPT subscription.

Already, we’re starting to see the headlines trickle in prophesying the risks of vibe coding complete commercial applications.

The most famous of these in the past month was the recent hacking of the social dating app, Tea, in which thousands of user identities and their government IDs were compromised. Many claim the app was vulnerable due to a vibe-coded user database. In two seconds of Googling, I also found another case where hackers accessed sensitive corporate data via exposed API endpoints in another vibe-coded app.

But I didn’t write this to sound like an old fogey who is against new technology. Let me explain what I like about vibe coding.

Here’s Where I See Vibe Coding Being Successful

Vibe code is amazing for making proofs of concept or demos for investors. It’s a great way to show viability on the fly with less effort. You can see that by how many vibe-coded apps are now going public. It’s definitely MORE than sufficient to get an investor on board.

Founder of OSMOS, Eli Cohen, put this nicely on LinkedIn:

“Think of it like those invention prototypes from the 80s. Crazy contraptions that proved a concept but would never ship to customers.”

Here’s where I think vibe coding will never work:

I think these are the three largest concerns for vibe-coded apps:

1. Security — Our biggest clients at Appstem are in healthcare, fintech, and security. These organizations cannot risk any form of security breach. Furthermore, many of these organizations cannot legally let consumer-grade LLMs access their internal data. This further complicates building any enterprise grade vibe-coded solution.
2. Maintainability — You can vibe code an entire app, but you will eventually need a developer to pop the hood and make sure everything runs correctly. If a new dev is not familiar with your code structure, it’s going to take them much longer to parse through and edit it. In many cases, they may have to rewrite the whole thing.
3. Scalability – If you’re building on a bad code base, you’re never going to have a good product. Once you’ve vibe-coded an MVP, you need to get a real developer structuring the code ASAP so that it can be built more easily.

I also generally think vibe coding is not a good fit for large legacy code bases. LLMs need to train on a certain volume of data before they can actually program well. There likely isn’t enough training data in existence for some of these code bases for AI to ever get the gist.

Beware The Vibe Coding Trap For New Engineers

The last thing I want to leave with is a warning to new developers: Learn to code with LLMs, but make sure you can code without them too.

As part of Appstem’s software engineer vetting process, we make all applicants take part in a coding challenge. Recently, we’ve seen a large uptick in AI-generated responses and we can spot these quite easily. Obviously, AI-generated responses are an immediate disqualification for working at Appstem. I have reason to believe most development-heavy startups are the same.

Don’t believe the hype: the age of the skilled full-stack developer is not ending. Quite the contrary, it’s just beginning.

AI will never replace a fully trained engineer, but it’s making it harder for new developers to get junior roles. There’s much less excuse to turn in a resume with zero completed projects to your name now that you can build out a wireframe with AI.

Want to Build Out Your Enterprise’s AI Capabilities?

If you run an enterprise and are looking to invest in AI, we’d love to give you a free consultation.

At Appstem, we work with enterprise clients like KFC, Tesla, Enterprise, Clorox, Jiffy Lube , and more – but you don’t have to be a multinational to work with us.

We’ve helped clients like you build and maintain useful software tools like kiosks, mobile apps, webpages, and more (we’re always experimenting with new tech).

Our goal is to help you access all the services of an internal development team without needing to hire and maintain one of your own.

We don’t want to be your agency. We’d like to be an extension of your team.

Do we sound like a fit? Fill out our form here and we’ll be in touch shortly.