What is Vibe coding?

The new era of programming according to Andrej Karpathy

Photo by Florian Olivo on Unsplash

In this era of ChatGPT, human habits are changing at a rapid pace. During this revolutionary time, Andrej Karpathy (ex-OpenAI co-founder) coined a new term “Vibe Coding”

Subscribe to datasciencepocket on Gumroad

What is Vibe coding ?

Vibe coding is like letting an AI buddy handle the nitty-gritty of coding while you chill and give it high-level vibes about what you want. It’s coding on autopilot, where you trust the AI to do most of the heavy lifting.

According to the message by Andrej Karpathy, “vibe coding” is a coding approach where you:

1. Fully Embrace the AI’s Capabilities: You trust and rely heavily on AI systems, such as large language models (LLMs), to handle coding tasks.
2. Interact with AI through Natural Language: Instead of writing code manually, you communicate with the AI using natural language, asking it to perform tasks or make changes.
3. Minimize Direct Coding: You barely touch the keyboard because you’re using tools like SuperWhisper to dictate commands to the AI.
4. Accept Changes Without Reviewing: You accept all changes made by the AI without reading the differences (“Accept All”), trusting the AI’s suggestions.
5. Resolve Errors by Copy-Pasting: When encountering error messages, you copy and paste them into the AI without comments, often fixing the issue.
6. Work Around Bugs: If the AI can’t fix a bug, you work around it or ask for random changes until the problem is resolved.
7. Build Projects with Minimal Coding: You’re building projects or web apps, but it feels less like traditional coding and more like seeing, saying, running, and copying/pasting stuff, which mostly works.
8. Forget About the Code: You embrace the process so much that you almost forget the code exists, focusing more on the outcome and the interaction with the AI.

If you notice, you don't need to know programming. Just a few clicks here and there with some natural language commands

This approach is described as being suitable for quick, experimental projects or “throwaway” weekend projects, where the goal is more about exploration and amusement rather than creating robust, production-ready code. It’s a testament to the growing capabilities of AI in software development and how it can change the way developers work.

How will Vibe coding change the future?

In many ways …

• Democratizing Coding: More people can jump into making cool stuff without needing to be a code wizard. It’s like everyone gets a magic wand to build apps and websites.

I’ve seen noobs creating intermediate apps

• Boosting Productivity: Developers can chill more and let AI handle the heavy lifting. It’s like having a super-helper that gets stuff done while you sip your coffee.

I’m coding things at lighting speed

• Innovating Faster: With AI doing the boring bits, devs can focus on the fun, creative parts. It’s like having more time to brainstorm and come up with killer ideas.

Executing 10 ideas simultaneously is possible now

• Smoother Teamwork: Teams can vibe better together, with AI bridging gaps and making sure everyone’s on the same page. It’s like having a DJ that keeps the party going without any awkward pauses.
• Changing Education: Learning to code might shift to learning how to work with AI, which is like learning to speak a new language with machines.

Tech is due a massive shift

• Low-Code Love: It’s all about making coding easier, and vibe coding fits right into the low-code movement. It’s like DIY but for digital stuff.

You don't need to know programming at all

But the grass is always greener on the other side

Problems with Vibe coding

• Quality Control: AI might not always hit the mark, leading to bugs or wonky code that needs extra effort.
• Humans may lose programming abilities: Relying too much on AI might make devs a bit lazy or out of touch with the nitty-gritty of coding.
• Readability: AI-generated code can be a hot mess, making it hard for other devs to understand or work with later on.
• Security Risks: AI might not catch all the security flaws, leaving apps and websites open to sneaky hacks.
• Learning Curve: There’s still a need to learn how to effectively communicate with AI, which can be its kind of hustle.
• Job Impact: Some devs might worry about AI taking over their gigs, even if it’s just for certain tasks.
• Ethical Concerns: There’s a debate about who’s responsible when things go south with AI-generated code.
• Scalability: AI might struggle with complex projects that need a more human touch and strategic planning.
• Cost: Using advanced AI tools might come with a hefty price tag, not everyone can afford to play. You would require a really good LLM to try this way of working.

What’s my take on vibe coding?

• If you’re a programmer, it makes you a 10x programmer for sure
• If a non-tech guy, you now have something to build baseline apps quickly.
• Can you rely solely on vibe coding? Baseline problems: Yes, Intermediate and domain-specific problems: No
• Would coding be a thing of the past: Maybe yes!

If you wish to try your hands on Vibe coding, I would strongly suggest you opt for the best coding LLM (say Claude 3.5 Sonnet or DeepSeek) alongside some good AI IDE (like cursor or cline) to make it a smoother ride else this can be a big headache.

Concluding,

Vibe coding is a fresh, laid-back approach to programming where AI takes the wheel, handling most of the coding while you focus on the bigger picture. It’s great for quick projects and can supercharge your productivity, but it also comes with its own set of challenges like quality control and security. In the end, vibe coding isn’t about replacing traditional coding but enhancing it, making it a powerful tool for both seasoned devs and coding newbies.