‘Vibe coding’ trend grows

A year ago, “I built this in a weekend” usually meant a rough demo and a lot of missing pieces. In 2026, it often means someone described an app in plain English, let an artificial intelligence model write most of the code, and shipped a clickable prototype before Monday. (arxiv.org) (cursor.com) The phrase people use for that is “vibe coding,” and Andrej Karpathy pinned it down in a February 2, 2025 post describing a workflow where you “forget that the code even exists” and keep steering with prompts instead of typing every function by hand. The term spread because it named a habit that coding tools had already made possible. (arxiv.org) (aiwiki.ai) The basic trick is simple: the prompt becomes the blueprint. Instead of writing a login page, a database schema, and an error handler line by line, the builder writes instructions like “make a meal-planning app with sign-in, saved recipes, and a weekly calendar,” then keeps revising the result in conversation. (developers.openai.com 1) (developers.openai.com 2) That only works because the tools changed. Anthropic said in mid-2025 that Claude could build, host, and share interactive apps directly inside Claude, and later said millions of users had already created more than half a billion artifacts ranging from games to productivity tools. (anthropic.com 1) (anthropic.com 2) Cursor pushed the same shift from the code editor side. Its own documentation now describes the product as an artificial intelligence editor and coding agent that can understand a codebase, plan features, fix bugs, and review changes, which is a very different job from old autocomplete tools that only finished the next line. (cursor.com 1) (cursor.com 2) That is why the prompt itself is starting to matter like a work sample. OpenAI’s prompt engineering guides treat the instruction as the thing that determines output quality, and builders are increasingly saving those instruction sets the way designers save mockups or writers save drafts. (developers.openai.com) (help.openai.com) Recruiters are noticing because a prompt-built app can show the whole flow of a product in one link. A candidate can demo onboarding, search, payments, and a dashboard in a portfolio piece even if the underlying code would not survive a production code review. (recruit2.com) (questera.ai) That has made “good enough to click through” a much bigger category than it was in 2024. Anthropic’s artifacts pitch is explicitly about skipping deployment hassle, and that convenience is perfect for internal tools, hackathon demos, client mockups, and job portfolios where speed matters more than handcrafted architecture. (anthropic.com) (anthropic.com) The backlash is growing at the same speed. Karpathy’s original description included accepting changes without reading every difference, and that shortcut is exactly what scares experienced engineers who have to maintain security rules, test coverage, and code that still works six months later. (arxiv.org) (anthropic.com) The split is becoming clearer in public data. The Next Web reported this week that artificial intelligence coding tools helped drive an 84 percent year-over-year jump in new Apple App Store submissions in the first quarter of 2026, while Apple also started pulling apps that broke platform rules, which is what happens when making software gets easier faster than reviewing it. (thenextweb.com) (gizmodo.com) So the new skill is not just “can you code.” It is whether you can describe a product clearly enough that a model can build version one, then spot which parts are demo glue and which parts need a real engineer before anyone trusts the app with money, health data, or payroll. (developers.openai.com) (anthropic.com)