👾 Vibe Coding: How AI is Redefining the Way We Code

Vibe coding is a software development approach that relies heavily on AI to write, refine, and debug code, so developers can focus on the desired outcome (hence the “vibe”) rather than digging into the details of the code. The term “vibe coding” was coined by AI researcher and OpenAI co-founder Andrej Karpathy in early 2025. Karpathy described it on X as follows: “There's a new kind of coding I call "vibe coding", where you fully give in to the vibes, embrace exponentials, and forget that the code even exists… It's not too bad for throwaway weekend projects, but still quite amusing. I'm building a project or webapp, but it's not really coding - I just see stuff, say stuff, run stuff, and copy paste stuff, and it mostly works.”

Vibe coding focuses heavily on iterative refinement. The developer uses plain English to tell the AI what they want the software to do; the AI handles low-level coding based on user prompts. The developer then provides feedback, and the AI adjusts code accordingly. 

A prime example of vibe coding at work comes from Rene Torcios, the “hackathon king of SF”. Torcios has attended over 200 competitions where he’s competed against Stanford students and professional software engineers and won cash, software credits, and clout, all without knowing how to write a single line of code. Instead, he has been relying on ChatGPT and Replit, a vibe coding platform, to write code for him since 2023. “Anyone can build anything they want,” he said. Turcios has recently started building his own AI agents startup as a solo founder, having AI do all the coding. 

Even Google CEO Sundar Pichai says that he is vibe coding a webpage for fun. "I've just been messing around - either with Cursor or I vibe coded with Replit - trying to build a custom webpage with all the sources of information I wanted in one place," Pichai said. Coding had "come a long way" from its early days. 

Vibe coding has a lot of obvious benefits. It saves time for developers, especially for simple projects; it helps developers create personalized tools without having to build the code from the ground up; it makes software development accessible to people without coding experience; all while allowing developers to focus on the overall vision and functionality of software rather than getting bogged down in code details. (Not to mention that it can work for beta products or testing).

If we were to create a poll, the three most popular vibe coding tools today would probably be Cursor, Windsurf, and Lovabale. While they are all great tools, they differ in terms of technicality and use cases. 

Cursor is an AI-powered vibe coding tool integrated into VS code, an open-source and highly popular code editor developed by Microsoft. It also integrates deeply with Github copilot. Cursor is ideal for experienced developers or existing VS Code users who want full control over their code but also added “superpowers” that give them smart code suggestions. 

Windsurf, recently acquired by OpenAI, is an AI-first development environment with a heavy focus on enabling collaborative development. Windsurf is best for dev teams that want to generate entire applications using AI in a short period of time, making it especially ideal for startup teams. 

Loveable is simpler technically and built for solo developers and freelancers. It is a personal dev agent that allows developers to automate simple, repetitive tasks like setting up the project and fixing bugs. While it is less commonly used for entire app generations, Loveable is great at providing task-oriented guidance during project build. 

Cursor offers a “tab-tab-tab” flow, providing multi-line code suggestions. It provides smart debugging and refactoring features, including inline bug-fixer and smart rewrites. However, Cursor does not provide preconfigured templates and is focused on editing existing code instead. 

Windsurf’s Supercomplete is a smart, multi-line autocomplete that also lives in the tab key and predicts the insertions, deletions, and edits the developer will make next based on their history and clipboard. It also has strong debugging capabilities, where its agent detects bugs and runs fixes across multiple files. Windsurf also provides full-stack templates for frontend and backend applications. 

Loveable provides task-oriented suggestions instead of writing the entire code structure. It has limited debugging features and lacks deep code refactoring capabilities. However, it provides many ready to use templates for apps, landing pages, MVPs, and more. 

All three platforms provide built-in agents to help with planning, debugging, and background tasks. In particular, Windsurf’s Cascade is the strongest. It is a full agent that watches everything you do - file edits, terminal commands, clipboard, chat history - and then plans multi-step “Flows” to get the task done. Loveable’s agent is conversational, workflow native, and more task-oriented than file-centric. Cursor has inline agents that are laser-focused on code completion and background agents that solve coding tasks without user supervision. Cursor even launched a Slack integration that allows users to assign tasks to background agents in Slack. 

The rise of vibe coding is rewriting the startup playbook. A notable case was Base44, a 6-month old, bootstrapped vibe coding startup sold for $80M to Wix in June 2025. Base44 hit its first 10,000 users 3 weeks after its launch and grew to 250,000 users, generating $189K in profit in May 2025. It’s been a “crazy f***ing journey so far,” Base44’s founder Maor Shlomo posted on LinkedIn when announcing the news of the acquisition. With vibe coding, many founders no longer need to recruit a team of software developers. Instead, they rely on AI to write most of the applications, making it possible to move at venture-scale speed without venture-scale headcount. Large engineering teams are replaced by single developers who could act as full-stack engineer, UX designer, and data analyst at the same time. Base44 is a glimpse into the future where solo startups are not only possible but highly profitable. 

However, vibe coding can also bring in bad consequences if used incorrectly. One prime example is the creation of “technical debt”: developers may become overly reliant on AI to write the code and end up not even understanding the code that they produced, potentially making the code difficult to maintain and expand in the long run. In addition, if AI-generated code for user-facing applications is not properly reviewed, it may lead to security flaws. Microsoft Azure CTO Mark Russinovich cautioned that vibe coding tools often break down when handling complex software projects, especially ones that span multiple files and have different parts of code that rely on each other. “These things are right now, still beyond the capabilities of our AI systems… You’re going to see progress made. They’re going to get better. But I think that there’s an upper limit with the way that autoregressive transformers work that we just won’t get past.”

While vibe coding works like a charm for building prototypes and fun side projects, when it comes to full stack development, the technology doesn’t seem to be quite there yet. Writing code that runs isn’t the same as building software that’s durable, expandable, and intentional. Anyone could prompt an LLM to write an app, but making sure that it does not break under pressure and can be expanded in the future requires thoughtful design and testing that the AI is not equipped with yet. The prioritization of speed over structure in vibe coding leads to things breaking as a result. 

How should we use vibe coding to maximize its potential? Tammuz Dubnov, founder of AutonomyAI, points to a new direction: vibe engineering. While it still uses the power of AI, vibe engineering contains AI within structure. Instead of prompting the AI to write an entire application, the developer asks it to fix and improve specific lines of code, giving AI context and accountability instead of simply freelancing. The developer isn’t writing every line; neither is the AI. Instead, the developer designs a system that writes code and tests it, with AI embedded in specific workflows. This is supported by Russinovich, who said that the future lies in AI-assisted coding, where AI writes some code but developers maintain control over architecture and complex decision-making. 

Although vibe coding is still in its early days, it has shown incredible potential in redefining software development, and it’s just the beginning. As AI becomes more deeply integrated into our daily workflows, vibe coding hints at a broader trend: we’re moving from telling machines how to do things to simply telling them what we want. It’s not just about writing code faster; it’s about changing who gets to build, and how. Whether you’re a seasoned dev or someone with zero tech background, the future of software and more might just be built on vibes. 

Written by Eva Jia .