Gemini за 233 секунды создал приложение для огорода — и сам исправил баг

A Verge author gave Gemini a detailed prompt and stepped away from the computer for five minutes. When they returned — a garden organization app was running in the preview window, and next to it a bug message was blinking. This is what vibecodeing looks like in 2026 for non-programmers: no code, only words — and a working product as the result.

The garden as a test case

The task was quite concrete: the yard was gradually falling into disrepair, and they wanted a tool for planning plantings, scheduling watering, and tracking plant health. They didn't want to buy a ready-made app — they needed something personal. Learning to code wasn't an option either. The solution came in the form of Gemini and vibecodeing. The author wrote a long, detailed prompt describing everything they wanted to see in the app. They clicked "run" and went off to do other things. In five minutes, there was a finished product on the screen. Though with a problem.

233 seconds to fix a bug

The screen displayed a message: "The channel is irretrievably broken and will be destroyed." It sounded catastrophic. But right below it was a button labeled "Fix bug." The author clicked it. Gemini provided an explanation — something about "locks" and "race conditions." The journalist honestly admits they didn't understand a word. After 233 seconds, the AI reported: the problem was fixed, the app works.

"It was strange: I asked the computer to create an entire application based on one request — and it asked me to click a button to fix a bug.

I clicked. It was exhilarating."

An important detail: this was already the second or third attempt. The previous ones apparently ended less successfully — which is itself an honest picture of vibecodeing. Iterations are inevitable here, but the threshold for overcoming them is fundamentally different. Unlike traditional programming, errors don't require a debugger or hours on Stack Overflow. It's enough to rephrase the request — or click the button that the AI itself offers.

Who is trying vibecodeing now

A year ago, vibecodeing was primarily associated with startup founders — people with technical thinking but no time to write code themselves. Gradually, the audience expanded:

• Managers and analysts — automating Excel routines and internal tools
• Marketers — building simple dashboards and forms without the IT department
• Freelancers and small businesses — creating client tools quickly and cheaply
• Ordinary people — making apps for personal tasks: garden, budget, habit tracker

Gemini, Claude, ChatGPT, and specialized tools like Cursor and Bolt have lowered the technical threshold to almost zero. The main skill now is being able to articulate the task precisely. Errors are inevitable, but the AI often offers to fix them right in the interface.

What's changing in development

The boundary between "user" and "developer" had been getting blurrier, but vibecodeing seems to erase it entirely — at least for a certain class of tasks. Professional development isn't going anywhere: complex systems, security, scaling — all of that still requires expertise. But for personal use, for prototypes, for small internal tools, the barrier has collapsed. The story about the garden app is telling precisely in its ordinariness: not a startup, not a technical user, not a conference demo — just a person whose yard is withering away. They weren't trying to create a commercial product. They were simply solving their problem.

What this means

Vibecodeing has reached everyday life. When an ordinary journalist with no programming skills creates a personal tool with one prompt — this is no longer a trend for tech enthusiasts. It's a signal: AI development has become part of ordinary life.