Tag: Andrej Karpathy

  • What Is Vibe Coding? How AI-Powered Development Lets Anyone Build Apps Without Writing Code

    What Is Vibe Coding? How AI-Powered Development Lets Anyone Build Apps Without Writing Code

    Building an app once required years of learning programming languages, debugging errors, and understanding complex logic. That barrier is fading fast. Vibe coding has changed how people approach software development, letting anyone describe an idea in plain English and watch an AI model turn it into working code.

    What Vibe Coding Actually Means

    Vibe coding is a method of building software where a person describes their goal in natural language, and an AI model writes, runs, and fixes the code based on that description. Instead of learning syntax or writing every function manually, a user simply says something like “build me a to-do list app with dark mode,” and the AI generates a working version.

    The term was coined by AI researcher Andrej Karpathy in February 2025, when he described giving in to “the vibes” and letting AI handle the coding almost entirely. Collins Dictionary later named vibe coding its Word of the Year for 2025, officially defining it as the practice of using natural language prompts to make an AI program produce the desired code.

    How the Process Actually Works

    Vibe coding generally follows a simple loop: describe what you want to build, the AI interprets the request and writes the underlying code, you review the output and ask for adjustments, and the AI refines it further based on that feedback.

    Tools such as Lovable, Replit, Bolt, GitHub Copilot, and Cursor have built entire platforms around this workflow, handling tasks like database setup, real-time updates, and error debugging automatically. Instead of writing a complex database query, a user might simply describe the dashboard they want, and the AI generates the required logic behind it.

    Why It Matters for Non-Coders & Businesses

    Vibe coding lowers the barrier to entry for building software in a way few previous tools have managed. People with no programming background can now launch functional apps, websites, and prototypes without hiring a developer or spending months learning to code. Businesses benefit too, since ideas can move from concept to a working demo in hours rather than weeks. According to reports, by mid-2026 roughly 25 percent of Y Combinator startups were building products using mostly AI-generated code, showing how quickly this shift has moved from a novelty into standard practice for early-stage teams.

    The Limits & Risks Worth Knowing

    Vibe coding is not without its downsides. Since users often skip reading the generated code line by line, mistakes and security gaps can slip through unnoticed. Early adopters have reported issues ranging from broken features to AI-generated content that misrepresented real data, such as fake product reviews appearing on a demo app.

    Developers who understand the underlying logic can catch these problems faster, which is why many experts recommend treating AI output as a strong first draft rather than a finished product, especially for anything handling sensitive user data.

    The Road Ahead for App Development

    Vibe coding is reshaping who gets to build software, shifting the developer’s role from writing every line of code to guiding an AI assistant toward the right outcome. As AI models keep improving at understanding context and intent, the applications people can create this way will only grow more capable.

    It is not about replacing skilled developers entirely, but about giving far more people the tools to turn their ideas into something real, faster than ever before.

    Frequently Asked Questions

    1. What is vibe coding? Vibe coding is an approach where a person describes an idea in plain language and lets an AI assistant translate it into working software, rather than writing the code line by line.
    2. Does vibe coding replace developers? No, it does not replace skilled developers entirely. It gives more people the tools to turn their ideas into working applications, and experienced developers still play a key role in guiding, refining and reviewing the output.
    3. Can someone with no coding experience use vibe coding to build an app? Yes, vibe coding makes app building accessible to people who never learned to program, since the process relies on describing intent rather than writing syntax.
    4. Is AI-generated code from vibe coding safe to use directly? Experts recommend treating AI output as a strong first draft rather than a finished product, especially for anything handling sensitive user data, so a review step remains important.
    5. How will vibe coding evolve as AI models improve? As AI models get better at understanding context and intent, the applications people can build through vibe coding are expected to grow more capable and complex over time.
  • Karpathy’s Lord of the Rings AI Test Highlights LLM Gaps in Visual Reasoning and Planning

    Karpathy’s Lord of the Rings AI Test Highlights LLM Gaps in Visual Reasoning and Planning

    Andrej Karpathy, a prominent artificial intelligence researcher, has demonstrated a paradigm shift in how large language models (LLMs) are evaluated—moving from abstract coding benchmarks to complex, real-world software engineering tasks. In a recent experiment, Karpathy challenged Anthropic’s Claude Opus 5 to transform the opening paragraph of J.R.R. Tolkien’s The Lord of the Rings into a fully interactive 3D website, using the Three.js JavaScript library.

    Instead of requesting small code snippets or static images, Karpathy provided the model with a token limit of one million (approximately $10 in cost) and instructed it to generate a complete, browser-based 3D scene. This test goes far beyond conventional evaluation methods, which typically focus on isolated tasks like code generation or text completion. Here, the model was required to think, write thousands of lines of code, manage resources, and produce an interactive product—all in one go.

    From Benchmarks to Real-World Engineering

    Karpathy’s approach reflects an industry-wide trend: as AI models grow more capable, engineers want to assess their performance on end-to-end workflows rather than narrow tests. The experiment underscores that the frontier of AI evaluation is shifting toward holistic, production-like challenges that reveal both strengths and persistent weaknesses.

    Visual Reasoning and Scene Composition Still a Challenge

    While Claude Opus 5 succeeded in generating an interactive 3D scene, the experiment also exposed critical shortcomings. According to Karpathy, visual rendering and scene composition remain difficult for even the most advanced AI systems. Additionally, evaluating the quality of such outputs is problematic: while code can be automatically tested, visual data often requires human judgment to assess accurately.

    Raising the Bar for AI Evaluation

    Karpathy’s experiment illustrates that expectations for LLMs are expanding beyond answering questions or writing code. Developers now want AI to build complete applications, reason across multiple steps, and generate interactive digital experiences with minimal human intervention. As AI capabilities continue to improve, real-world engineering challenges like this one will play a bigger role in measuring model performance, highlighting not only what these systems can accomplish but also the technical gaps that remain before they can consistently deliver production-ready software.

  • Andrej Karpathy: Speak Naturally, Not Perfectly — Why Casual Voice Prompts Outperform Polished AI Instructions

    Andrej Karpathy: Speak Naturally, Not Perfectly — Why Casual Voice Prompts Outperform Polished AI Instructions

    Artificial intelligence researcher Andrej Karpathy, the mind behind the term “vibe coding,” is challenging the common obsession with crafting the perfect AI prompt. Instead, he advocates for a more natural approach: simply say what you want, even if your thoughts are messy or incomplete.

    In a post on X on July 21, 2026, Karpathy explained his preference for using voice mode when interacting with large language models (LLMs). Rather than typing carefully refined instructions, he speaks freely, allowing the AI to organize his scattered ideas into a coherent response. He often tells the chatbot upfront that his thoughts may be messy or contain mistakes, so the model focuses on intent rather than literal interpretation.

    “One pattern I find useful for working with LLMs is a nice long ramble session. Sometimes the LLM needs more bits to understand what you’re trying to achieve, but you’re too lazy to type them. In these cases I like to lean back, switch to voice and just ramble.”

    Karpathy argues that many users provide too little context. He believes longer, more natural instructions give the AI a better chance to grasp the real objective than short, polished prompts. His comments come amid growing investment by AI companies in voice-driven assistants designed to facilitate more fluid conversations.

    Karpathy coined the term “vibe coding” in early 2025, referring to a coding method where developers use natural language prompts to generate code via AI instead of writing every line manually. His latest message reinforces that philosophy: forget about perfect prompts, focus on clear communication.