Vibe Coding
How AI is changing how and who builds software...forever
What if we never needed to write a line of code in the future?
Are entry-level software engineers at risk?
Has the ‘Coding’ boom ended?
These are all questions I try to answer today, with a sprinkle of personal experience and examples from my personal and professional life with coding.
Let’s dive in!
WTF is ‘Vibe Coding’
Let’s start with the basics. What the hell is Vibe Coding?
Andrej Karpathy (founding team OpenAI, Tech Director Tesla and more) recently coined the term that describes how AI-powered apps are enabling coding without programming/technical skills. Forgetting for a second the very Californian name (though I admit it’s catchy) it marks a clear distinction between AI-coding assistants (think of Github Co-Pilot, and any IDE where engineers write code and get AI suggestions on code) and natively built apps that center around a chat interface.
So, anyone, without coding experience, can pretty much get into the game just by chatting with the app. Here’s an example of the starting screen by Lovable:
This is a true paradigm shift, well because not only it centers around using english as the coding language, but also provides an interface which lets you see in real-time changes you’re making to what you’re building.
Here’s an example from a very simple app that I asked it to build for ‘Layoff'-Lifeline’, a small set of resources and interactive tools I had been thinking to build for a while, but never had the time (or capability) to do so. This weekend, in just half-an hour, I got a functioning prototype going on, with a financial calculator, job search assistant that actually gets your CV and provides job opportunities and more.
It wasn’t all perfect, and I haven’t built a rocket ship - but the joy it brought me to see my idea come to life so quickly and effortlessly, definitely excited me.
The idea of ‘vibe’ is that given this experience, it becomes both easy for non-technical people to build their own projects, but also for slightly more technical folks to do changes quickly (i.e. ‘turn all of the buttons to green, make the icons larger and so on). It’s instant gratification for building. Better than social media, if you ask me.
So who are the main players in this up and coming space? Here’s a short list:
Cursor (cursor.so)
Key Features: AI code generation, built on VS Code, user control, AI buttons for common tasks.
Total Funding: $173.25 million.
Stats: Over 360,000 paying developers and 40,000 paying customers. Used by engineers at OpenAI, Midjourney, Perplexity, and Shopify. Achieved $100 million ARR in 12 months .
Replit (replit.com)
Key Features: Ghostwriter AI for code generation, browser-based IDE, cloud-based environment, user-friendly interface, extensive language support, real-time collaboration, built-in hosting.
Total Funding: $272 million.
Stats: Over 30 million software creators . Early focus on young developers and educational institutions .
Lovable (lovable.dev)
Key Features: AI-powered full-stack app builder, natural language interface, rapid prototyping, integrates with GitHub and Supabase.
Total Funding: $27.7 million.
Stats: Over 140,000 registered users . Achieved $4 million ARR in four weeks and $17 million ARR in three months . Processes over 1,000 product builds daily .
Bolt.new (bolt.new)
Key Features: AI-powered development with natural language commands, browser-based development environment, integrates with existing tools, rapid development.
Total Funding: $105.5 million.
Stats: Over 2 million user registrations, achieved $20 million ARR in two months . Claims to reduce development costs by 99% .
Rosebud AI (rosebud.ai)
Key Features: Creates AI-powered characters and games, offers game development tools, simplifies game development with features like Sprite Sheet Generator and AI NPCs.
Total Funding: $8.8 million.
Quick note about Cursor: it’s been the fastest growing app to reach 100M$, in just 2months.
It's Not All Sunshine and Roses
Now, before we all get carried away with the hype and start planning our early retirement from coding (just kidding!), it's important to remember that vibecoding, for all its magic, isn't without its limitations. Even the most advanced AI tools still have some kinks to work out.
AI Can Be a Bit of a Diva: Like a temperamental artist, AI sometimes throws a tantrum and produces unexpected results. You might describe what you want perfectly, but the AI might misinterpret your instructions or generate code that doesn't quite hit the mark. This means you still need a basic understanding of what's going on under the hood to spot errors and guide the AI back on track.
Complexity is Still a Challenge: While AI excels at automating simple tasks and generating basic code structures, it can struggle with more complex projects. Think intricate integrations, custom features, or anything that requires a deep understanding of software architecture. For those heavy-duty projects, you'll still need the expertise of a seasoned developer.
The "Black Box" Problem: One of the challenges with AI is its lack of transparency. Sometimes it can feel like a "black box" where you input a prompt and get an output, but you don't really understand how the AI arrived at that solution. This can make it difficult to debug or modify the code, especially if you're not familiar with the underlying logic.
The Creativity Conundrum: While AI can generate code, it can't (yet) replicate true human creativity and ingenuity. It can follow instructions and learn from existing data, but it can't come up with truly original ideas or solve problems in novel ways. So, if you're looking to build something truly groundbreaking and innovative, you'll still need that human spark.
The best software will likely be built through a collaboration between humans and AI, where humans provide the creativity, critical thinking, and problem-solving skills, while AI handles the grunt work and automates repetitive tasks. It's like a dynamic duo, each playing to their strengths.
How is this poised to change the Coding Job Market?
Fred recently posted a chart on Software Development Jobs in the US. We're seeing a 35% drop in software developer job listings compared to just five years ago. That's not just a little dip; it's a dive. And it's not just some US thing; it's happening globally, though Australia's oddly bucking the trend. Now, you might be thinking, "Okay, so what? Markets fluctuate." But this feels different. It's not just about the usual economic cycles.
Think about it: interest rates are up, VC funding is down, and suddenly, everyone's talking about how AI is going to make us all super-productive. Companies are probably thinking, "Why hire a bunch of new engineers when we can just get our existing team to do more with AI tools?" It's that whole "wait and see" approach, but it's having real-world consequences. And let's be real, after the hiring frenzy of 2021-2022, a lot of companies probably over-hired anyway. They're probably still working through that backlog - and adjusting for macro-economic context (looking pretty grim at the moment, yes).
We're also seeing these smaller teams, like at Linear and Bluesky, proving that you don't need a massive engineering department to build awesome stuff. So, it's a perfect storm: AI making coding easier, companies being cautious, and the realization that maybe, just maybe, smaller, more efficient teams are the way to go.
If 'Vibe Coding' takes off, maybe we'll see a surge in non-developers building software. And who's going to maintain and improve all that AI-generated code? Maybe that's where the next wave of developer jobs will come from. Or maybe, just maybe, we're looking at a fundamental shift in how software is created and who creates it.
Software for One?
One of the most exciting aspects of vibecoding is the potential to create "software for one" – highly personalized apps tailored to individual needs and desires. Think of it as bespoke software, handcrafted just for you.
This concept, highlighted by Kevin Roose in his New York Times article, opens up a whole new world of possibilities. Imagine:
A fitness app that adapts to your unique body and goals. It analyzes your activity levels, sleep patterns, and dietary habits to create a truly personalized workout and nutrition plan.
A news aggregator that filters out the noise and delivers only the stories that are relevant to you. It learns your interests and preferences over time, becoming your own personal news curator.
A productivity tool that helps you manage your time and tasks in a way that works best for you. It understands your work style and helps you optimize your workflow.
These are just a few examples of how "software for one" could revolutionize the way we interact with technology. It's about moving away from one-size-fits-all solutions and embracing hyper-personalization.
This shift towards "software for one" has the potential to democratize software development like never before. It empowers individuals to become creators, not just consumers.
Imagine a world where anyone, regardless of their technical skills, can bring their software ideas to life. A world where a teacher can create a customized learning app for their students, a musician can build a unique sound editing tool, or a small business owner can develop a bespoke inventory management system.
This is the promise of vibecoding – to unlock the creative potential within each of us and enable us to build the software that we want and need, without limitations. It's about putting the power of creation back into the hands of the people.
It's a bit of a wild ride, and honestly, I'm just as curious as you are to see where it goes.
Ciao,
Matteo
Excellent informative well written article! Your insights are food for thought and the potential and possibilities are endless. Appreciation for your work and insights