It’s February 2nd, 2025. I’m sitting in busy lunch spot in Los Angeles, sipping coffee, when my phone flashes with a Twitter notifications. I almost ignore it—another day, another viral tech thread—but something catches my eye.
Andrej Karpathy, the former Tesla AI director and OpenAI co-founder, has just posted: “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.”
I pause, read it again. He continues: “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.”
Like naming a constellation that’s been in the sky all along, Karpathy’s term immediately crystallized what thousands of developers had already been experiencing: a fundamental shift in how software gets built. What began with simple code completions had evolved into a radically different approach where AI handles the technical implementation while humans focus on creative direction.
Three months later, what Karpathy named has morphed into nothing short of a revolution. Job postings for “Vibe Coders” are appearing on LinkedIn, Y Combinator’s latest batch is packed with startups building entirely AI-generated products, and an entire ecosystem of tools has emerged to support this paradigm that’s been evolving since the first coding LLMs appeared.
While the term “vibe coding” is new, the practice has been evolving organically since OpenAI released Codex in 2021. Early adopters were already letting AI generate substantial portions of their codebase, but the process still required significant human oversight and technical knowledge.
Each successive generation of models brought capabilities closer to what Karpathy described. GitHub Copilot began suggesting longer snippets. Claude started understanding multi-file projects. GPT-4 could reason through complex architectural decisions. The incremental changes were adding up.
“We’ve been doing something like vibe coding for nearly two years,” explains Mohammed Patel, CTO at Dataflare, “but we didn’t have language to describe it. We’d just say we were ‘heavily leveraging AI’ or ‘pair programming with GPT.’ Karpathy’s term captures not just the technology but the feeling—this sense that you’re vibing with the AI rather than writing code yourself.”
The history matters because what we’re experiencing isn’t a sudden emergence but the culmination of a trend that’s been hiding in plain sight. Developers have been gradually relinquishing manual coding tasks to AI assistants, initially with simple autocompletions but eventually with entire functions and architectures.
Now, with the most advanced models, the balance has fully tipped toward human-as-director and AI-as-implementer—a shift significant enough to deserve its own terminology.
The ecosystem that’s developed around this evolving practice has become remarkably sophisticated. Based on my conversations with dozens of developers and extensive testing, here are the standout platforms driving the vibe coding revolution as of May 2025:
The undisputed flagship of vibe coding, Cursor has become the standard bearer for AI-assisted development. Built on VS Code architecture, its “Composer” feature acts as an AI agent that can generate, edit, and debug code based on natural language prompts.
What makes it exceptional is its integration with Claude 3.7 Sonnet, widely regarded as the most accurate model for coding tasks. The combination of a familiar interface with cutting-edge AI capabilities has made it the tool of choice for both experienced developers and newcomers.
“I’m spending 80% less time actually typing code,” says Elena Varga, CTO at Quickshift, a Y Combinator-backed startup. “Instead, I’m thinking more deeply about system architecture and user experience—the stuff that actually matters.”
While Cursor dominates the desktop experience, Replit has claimed the cloud development space. Its browser-based platform eliminates setup entirely, making it particularly popular for rapid prototyping and collaborative projects.
Replit Agent functions as an AI software engineer, capable of not just generating code but also configuring databases, automating deployments, and fixing bugs autonomously. Most impressively, Replit supports over 50 programming languages and offers mobile app access—perfect for coding on the go.
Windsurf has emerged as the power user’s choice, particularly for managing complex repositories and large-scale projects. Its “Cascade” feature enables developers to orchestrate sweeping changes across massive codebases with simple natural language instructions.
The platform excels at maintaining context across large projects, making it ideal for enterprise applications where understanding the relationships between components is critical.
Anthropic’s entry into vibe coding takes a unique approach, operating as a command-line tool rather than integrating with a specific IDE. Launched in February 2025, Claude Code is designed to understand your entire codebase and help you code faster through natural language commands.
What sets it apart is its deep integration with terminal workflows and direct access to the Claude 3.7 Sonnet model, known for exceptional reasoning abilities. It excels at executing routine tasks, explaining complex code, and managing git workflows. While still in beta as a research preview, Claude Code has already demonstrated impressive capabilities, particularly for developers comfortable with terminal-based workflows.
One of the newest entrants, Augment Code has made waves with its ability to understand and manipulate massive codebases with remarkable precision. What distinguishes this platform is its context engine, which allows it to navigate complex dependencies and maintain awareness of the entire project structure.
Recent benchmarking showed Augment Code outperforming other tools by up to 70% on certain tasks, particularly in handling larger repositories with sophisticated architectures. Its ability to automate updates across entire projects makes it especially valuable for enterprise settings with established codebases.
The most beginner-friendly option in the ecosystem, Lovable embodies pure vibe coding through its innovative two-pane interface: AI chat on the left, live preview on the right. Users describe what they want, and the interface materializes in real-time.
“It’s the first true no-code platform that actually works,” explains Tariq Johnson, a graphic designer who built his portfolio site without writing a single line of code. “Previous no-code tools still required understanding their specific logic. With Lovable, I just described what I wanted, and it appeared.”
Specializing in frontend development, v0 has gained traction for its ability to transform design concepts into pixel-perfect implementations. While less versatile than general-purpose tools, its specialized focus makes it unbeatable for UI/UX tasks.
Many developers use v0 in conjunction with other tools, creating a workflow where v0 handles the interface while platforms like Cursor or Replit manage backend logic.
In a signal of how rapidly this trend is professionalizing, CO/AI, a Venice Beach startup, posted what appears to be the first “Vibe Coder” job listing on LinkedIn in April 2025. The post drew 300+ applicants inside three days and spun off sibling roles such as “Vibe QA Analyst” focused on “testing the vibe as well as the code.”
The listing emphasized translating conceptual ideas into shipped products using AI tools, with the provocative tagline: “Keyboard optional.”
This evolution reflects an astonishing productivity claim echoing through Silicon Valley: Fortune reports that some CEOs believe vibe coding “lets 10 engineers do the work of 100.”
Perhaps most telling is data from Y Combinator, the influential startup accelerator. In March 2025, YC reported that 25% of startups in its Winter 2025 batch had codebases that were 95% AI-generated, reflecting a dramatic shift toward AI-assisted development.
Many founders are embracing this approach not just for efficiency but as a competitive advantage, allowing small teams to ship features in days rather than months.
Kevin Roose, a New York Times journalist with minimal coding experience, experimented with vibe coding to create several small-scale applications in February 2025. He described these as “software for one,” referring to personalized AI-generated tools designed to address specific individual needs, such as an app named LunchBox Buddy that analyzed his fridge contents to suggest items for a packed lunch.
This democratization effect is perhaps the most profound consequence of vibe coding—turning consumers into creators without requiring years of technical training.
Not everyone is embracing the vibe. Critics point to serious limitations:
Security Vulnerabilities: MIT Technology Review notes that while vibe coding “can help make a vague idea for a website or a game into a reality, it can’t make it reliable or secure.” This was dramatically demonstrated when a vibe-coded SaaS application posted on social media was immediately compromised by security researchers.
Knowledge Gaps: When developers don’t understand the code they’re deploying, diagnosing and fixing subtle bugs becomes nearly impossible.
Maintainability Problems: As AI-generated codebases grow, they often lack the organized architecture that human developers would create, leading to what some critics call “LLM mush.”
Auditability Concerns: For regulated industries like healthcare and finance, the lack of explainability in vibe-coded systems poses significant compliance challenges.
As the movement matures, a playbook for responsible vibe coding is taking shape:
What Karpathy’s tweet did was give a name to a transformation that’s been underway for years. Vibe coding isn’t just a new approach—it’s the recognition that the relationship between humans and code has fundamentally changed.
The tools will continue to evolve, but the core shift is permanent: humans are becoming orchestrators rather than implementers, focusing on what to build rather than how to build it. This redistribution of cognitive labor allows for unprecedented creativity and productivity, though not without risks that demand careful attention.
As I look ahead one thing is clear: we didn’t start vibe coding in February 2025, but naming it has accelerated its adoption and expanded its possibilities. The question now isn’t whether AI will transform software development—it’s how quickly this transformation will spread to other domains.
Anthony Batt is a technology executive and founder in Los Angeles California. His work focuses on the intersection of technology, creativity, and cultural change. X & Linkedin
Join hosts Anthony, Shane, and Francesca for essential insights on AI's impact on jobs, careers, and business. Stay ahead of the curve – listen now!
Join hosts Anthony, Shane, and Francesca for essential insights on AI's impact on jobs, careers, and business. Stay ahead of the curve – listen now!
No hype. No doom. Just actionable resources and strategies to accelerate your success in the age of AI.