News
AI's Takeover of Software Development Gets a Name: 'Vibe Coding'
As far as we can tell, the first mention of "vibe coding" in Microsoft documentation appeared in a post from the Visual Studio Code dev team less than a month ago, apparently the company's first nod to the new term for letting AI coding assistants like GitHub Copilot do almost all of that pesky typing stuff that used to be the way software applications were developed.
"Happy vibe coding! Isidor and the VS Code team" is how Isidor Nikolic signed off on a Feb. 24 VS Code dev team post introducing a preview of GitHub Copilot agent mode in the open-source-backed code editor, possibly the most popular coding tool in the world.
The internet is now awash with "what is vibe coding?" posts, describing a practice that sees developers just accepting Copilot suggestions (see "VS Code Copilot Gets Closer to Tab, Tab, Tab Coding").
The term "vibe coding," much pithier than "tab, tab, tab coding," also popped up in a Visual Studio Magazine article last week about AI prompting techniques (see "Visual Studio Devs Share Copilot AI Prompts to Improve Code").
One of the techniques shared in response to a social media post asking for prompting tips was:
CRUD operations. I ask Copilot to enhance it. The AI expands my brief outline into several pages. I review, adjust, and follow up if necessary, unless I'm vibe coding. Then I prompt Copilot, "Make it so."
The way things are going, "vibe coding" is just a precursor to the time when "developers" can just verbally describe what they want in a new application and let Copilot or some other AI take over and build the thing. Such functionality is advancing at a rapid pace, as we recently detailed how easy it was for Copilot to create a bare-bones website just by examining, with "vision" capabilities, a poorly drawn mockup image (see "Hands On: New VS Code Insiders Build Creates Web Page from Image in Seconds").
Similar functionality was even demonstrated last October with Copilot Workspace (see "Copilot Workspace Does Web App in Minutes, No Coding Required").
With "vibe coding" now a mainstream term, its official Wikipedia entry explains it was first introduced last month by computer scientist Andrej Karpathy, a co-founder of OpenAI and former AI leader at Tesla. The term quickly caught fire.
"The concept refers to a coding approach that relies on LLMs, allowing programmers to generate working code by providing natural language descriptions rather than manually writing it," Wikipedia says. "Karpathy described his approach as conversational, using voice commands while AI generates the actual code. 'It's not really coding - I just see things, say things, run things, and copy-paste things, and it mostly works.' Karpathy acknowledged that vibe coding has limitations, noting that AI tools are not always able to fix bugs, requiring him to experiment with changes until the problems are resolved."
Indeed, there's already a backlash to vibe coding.
A
Reddit post on vibe coding just published yesterday says:
Has this become the only topic here? And the fact that some people are so triggered by it is really laughable. If someone spends money, they can do whatever they want with it. And I don't mean to say that I'm a vibe coder; I'm just concerned that my entire Reddit consists of vibe code topics. It's like you have nothing better to do in your life than think about how people use an AI program. You can't make that up, you just have to see it for yourself. So, as I said, it's laughable.
[Click on image for larger view.] 'The real pros know: Use AI as a tool. Not as a replacement for your brain.' (source: Kush Creates).
And a Hacker News post from two days ago is titled "Why Vibe Coding Makes Me Want to Throw Up." It references a Kush Creates post published on March 17.
"So Karpathy dropped this absolute gem about 'Vibe Coding' and I nearly spit my third coffee of the morning all over my keyboard," the post says. "Let me get this straight - we're now PROUD of not understanding our own code? Cool cool cool. Totally sustainable approach to Software Engineering."
So it appears vibe coding has rapidly progressed through research firm Gartner's "hype cycle" all the way to the famed "trough of disillusionment."
One reader replied this way to the Hacker News post: "I certainly don't hate vibe coding as much as this person, but having tried it a few times, the thing that really worries me is 'the slippery slope' he describes at the end and that's true with AI assistance in general. I think learning new programming languages probably expands your mind in ways that is somewhat immeasurable, and using AI to work with unknown languages takes away from that, and in general, I wonder what happens if you go from troubleshooting problems like 5 times a week, to once every 3 weeks because most problems fix themselves. And I do wonder if that will have a broader impact on the quality of your thinking as an engineer."
That discussion is happening all through the internet and among software developers everywhere.
In the meantime, "vibe coding" is certainly a better moniker than our "tab, tab, tab coding."
Stay tuned to see where vibe coding takes us. Here's what ChatGPT thinks will happen with vibe coding:
- From Writing Code to Directing AI
- The role of developers will shift from manual coding to guiding AI with natural language instructions.
- AI coding assistants will become more context-aware, understanding a project's structure and goals.
- Explosion of Indie Developers and Solo Startups
- Small teams and solo developers will achieve what once required entire engineering teams.
- Indie software projects will thrive, similar to how WordPress democratized website building.
- AI-Centric Programming Paradigms
- Traditional syntax-heavy programming will give way to higher-level conceptual programming.
- New programming languages or frameworks might emerge, optimized for AI-driven development.
- Challenges in Code Maintainability and Debugging
- AI-generated code can be difficult to debug and understand.
- Expect a rise in AI-driven code explainers to help developers optimize and debug vibe-coded projects.
- Security and Ethical Concerns
- Vibe coding could introduce security vulnerabilities if developers don't fully understand the AI-generated code.
- Automated security auditing tools will become crucial to ensure compliance and security.
- AI-Powered No-Code/Low-Code Convergence
- Vibe coding will merge with no-code/low-code platforms, making software development more accessible.
- This will blur the lines between traditional developers and power users.
- Resistance from Traditional Developers
- Some developers will embrace vibe coding, while others will resist it due to concerns about code quality and maintainability.
- Over time, AI-driven development will likely coexist with traditional software engineering.
About the Author
David Ramel is an editor and writer at Converge 360.