Artem Grigorenko, a team leader with expertise in software development and testing, as well as the author of a blog about QA and Management, shared his experience in vibe-coding a solution for automating Random Coffee — organizing paired meetings for members of his community.
Artem Grigorenko, a team leader with expertise in software development and testing, as well as the author of a blog about QA and Management, shared his experience in vibe-coding a solution for automating Random Coffee — organizing paired meetings for members of his community.
«Vibe coding is fun, but it hurts. I started this week in the style of 'let’s just vibe code, what could go wrong?' And, as always, everything didn’t go as I expected», — Grigorenko said on his LinkedIn.
He said that his community had suggested automating Random Coffee. This is the name given to a casual, friendly, and informal coffee meeting between two or more team members. The concept was originally introduced as an attempt to break down barriers between teams and departments, encouraging employees to engage in conversation.
«Before that, I was adding participants in ChatGPT and it itself created the appropriate pairs, etc. I wrote a script that was supposed to solve this whole thing, but the community said that „it’s not hipster-like, and we need to add UI“. Well, I think, okay. Cursor, your time has come. Siv generated the requirements and prepared the story in the chat and threw the prompt into Cursor. And then the magic began,» said Artem Grigorenko.
He noted that he used Next.js on the frontend and FastAPI for the backend.
«I decided to make Cursor in Python, because the script for it was already ready.»
«I decided to trust AI and just approve everything it tells me to do. By the way, I didn’t approve everything. And yes, it works, but then all sorts of jokes and gimmicks start,» the IT guy admitted.
He says that creating an MVP product with vibe coding is easy, but support is hell. He adds that «AI coding is great when you want to quickly put together something that works,» but the problems start when you need to make changes.
«AI can generate an entire project in an hour, but each change is hours of debugging. And if you are a Jun, you simply don’t know where to look. And that’s why the gap between Jun and AI seniors doesn’t disappear. It’s growing!», says Artem Grigorenko.
He explained that a high-level specialist reads the code, sees the architecture, and «debugs point by point,» while a junior can get lost «in prompts, configs, incomprehensible variables, and doubts.» This makes the debugging process very long.
«Cursor broke part of the working logic. And it still didn’t save the history of changes. And it doesn’t have undo (or I didn’t find it). And it doesn’t remember what happened,» the IT guy listed his problems with vibe coding.
Aitivets advised making commits and keeping a history of changes to avoid an «AI-driven local apocalypse.»
Recall that Andrey Karpaty, a former OpenAI researcher and now the director of artificial intelligence and computer vision for autopilot at Tesla, in February described a new programming technique called «vibe coding,» which involves using AI chatbots. It is currently gaining popularity among programmers.
80% of IT professionals who read dev.ua admit that they use artificial intelligence to write code. So-called vibecoders simplify and speed up their work.
Previously, we conducted a small survey and found out in which cases this method is appropriate and useful, whether IT professionals are satisfied with the quality of the code written by artificial intelligence, and how to correctly formulate a prompt for AI so that the result satisfies the requester as much as possible.
