Technology

Vibe Coding: Just Hit Accept and Ship It

Pete Casson

Pete Casson

· 6 min read
Vibe Coding: Just Hit Accept and Ship It

Trends and fads in coding are nothing new, there's always a new language, framework or tool being hailed as the next great thing. This time it's Vibe Coding, it's not a language, tool or framework but a way of writing software using AI and allowing it to do it all for you.

Andrej Karpathy, co-founder of OpenAI and former Director of AI at Tesla, tweeted in Feb saying that he is embracing a new way of coding that he calls, Vibe Coding. But is this the next great thing in programming or is it just another fad?

What is Vibe Coding?

Andrej describes Vibe Coding as;

Where you fully give in to the vibes, embrace exponentials, and forget that the code even exists.

To put it simply, you let AI write software for you and just hit 'accept' without reviewing any of it. We are in the age of writing 'no code' apps in a few hours, people who have never coded before are able to create a usable app without having to learn to code. YouTube is littered with videos on how people have created a £10k MRR business with 'no code'. I have no doubt there are a small number of apps that have been created using AI and are generating a livable income, but most are not designed to be used by anyone else other than the person who 'wrote' it.

Putting Vibe Coding to the test

I recently put Vibe Coding to the test. I needed a temporary place holder landing page. We had a design ready to go and it needed building, so I asked Cursor to do it. I referenced the Figma file and told it to use NextJS, Tailwind and typescript. Cursor took me through running the relevant commands to create the project, install the dependencies and create the files.

After a couple of minutes I had the foundations of an app that I could fire up and view. It wasn't far off the initial design, I carried on prompting for tweaks to the design to get it to match the Figma file. After a few more minutes I had a working landing page that was ready to go. From start to live on our platform took 36 minutes.

Not bad, but also this was a simple one page site with a predefined design. For this type of project, Vibe Coding is great. This landing page will be removed in the near future, so spending time crafting a beatifully written site isn't necessary here.

Beware the gotchas

Tools such as Cursor and GitHub Co-pilot have made it easier than ever to write code, you just give it an instruction and it'll generate code for you, sometimes it'll write the whole thing. It's genuinely amazing that we are at that stage. This is the back bone of Vibe coding, you just instruct the AI to build an app and ship it. That's great, but there's a few gotcha's to be aware of.

Be specific

AI code generators are very good at taking a specific instruction and creating code to solve a query. For example, "Create a view that shows all of my todo's in a list and allows me to edit, create, delete, mark as complete and upload attachments to a todo item". Tools such as Cursor will create you an app that does all of that. You'll need to keep prompting it to connect the app to a database, create login pages, settings pages, etc. But if you persist you'll get a working Todo app.

While AI excels at generating code for well defined tasks like ToDo apps, complex or novel problems require detailed, step by step instructions to achieve desired results.

Debugging

Debugging is pretty poor in AI tools. It really struggles to understand the bug and find a solution to the problem, particularly if the problem is interlaced over a number of files. There are AIs that will look at your entire code base and try to piece it together but we are still pretty far away from AI being able to fix bugs all by itself.

Deployment

One thing that I have noticed a lot with YouTube videos, you know the ones that say they've earnt £10k+ in the last month, none of them talk about deployment. Writing code is only one part of the problem, making it available to the general public is a completely different ball game. Where do you host your app? Is it web based? Is it a mobile app and needs to be available on the App Store? What about a database? What type of database do I need? Security? and the list continues.

AI will give you a steer on some options and may even write a deployment script for you but you have to know what you are doing. It's very easy to deploy an app and for it to be incredibly insecure and open to abuse.

Scaling

AI generated apps are great for a small group of people, they suck at scale. They're not optimised for scale, tend to be resources intensive and hard to extend. The code becomes a Jenga tower where you hope you don't pull the wrong brick out otherwise it'll come crumbling down. Allowing AI to generate all of the code will give you huge technical debt, and one day it'll become due at a high cost.

Security

This one is probably the most concerning for me. I've seen a lot of AI generated code that does the job asked but completely ignores security. SQL queries are left wide open, no authentication checks are added, inputs aren't validated and the list goes on. This will become a larger problewm over time and will only be fixed by human engineers.

Should I be Vibe Coding?

No, you shouldn't be just prompting and hitting accept and ignoring what the code does. Tools such as Cursor and CoPilot should be used as an assistant, making you a faster coder, removing lower level tasks and enabling you to focus on the bigger picture. As a developer you should absolutely be using AI coding tools, they aren't there to replace you, they're there to help you.

The role of a developer is changing, we're no longer just spitting out code, to be fair if that's all you're doing now then you need to up your game. The role of the developer is changing from coding to product, you need to understand why you are creating X feature or Y app. What's the user trying to achieve? How do I best facilitate that? There's only one way you're going to find that out.

You need to speak to your users!

Yep, we need to get out of our seat and speak to the people that are going to use our software. I get it, this will make some people uncomfortable, and it's not easy to speak to customers. You open yourself to criticism and negative feedback, but if you reach out to the right people you'll get fair, useful and actionable feedback. Buddy up with someone, or watch interviews that other people have done, there's so many sources of information just waiting for you.

The way I see it you have two routes as a developer now;

  1. Become a Product Engineer
  2. Become an Architect

Both aren't going to be replaced with AI any time soon. Someone has to design the layout of the app, manage the services, scalability, up time, maintenance & security that the entire platform sits on.

Equally we need people to create amazing apps for everyone. For me as a CTO and developer, this really excites me, AI is allowing me to focus on creating software that really solves a problem and is an amazing experience for people. I can get features and changes out faster than ever, we can test and get results back even quicker. For a startup this is gold dust, but it is for enterprises as well.

Where can I use Vibe Coding?

If you want to go all in on Vibe Coding then there are a couple of areas where it'll be useful. Prototypes and landing pages. Prototypes are throw away products that allow you to quickly learn and move on, you don't want to be losing time building the prototype. This is where Vibe Coding can be used to get a quick prototype live and get feedback. If the prototype becomes a full product then spend the time coding it correctly, using AI as an aid.

Same with landing pages, you can get AI to generate a bunch of pages quickly, just don't get it to write the copy please 🙏. Use your customer language and A/B test it.

Vibe Coding is the extreme way to use AI for coding and, as always, there's good things to take from it but in reality it's not useful for everything. AI is here to stay and it's developing at an incredible rate, far faster than anyone can really comprehend. If you don't embrace it you will be left behind.

Pete Casson

About Pete Casson

Pete is Chief Technology Officer and co-founder of Collctiv. Prior to this, he was a CTO at Twinkl and Software Engineer at McLaren Formula 1 Team.

Sign up to the HLH newsletter

Get the latest articles straight to your inbox

Copyright © 2025 Pete Casson. All rights reserved.