Vibe coding is everywhere. Even AI writes its own code for itself
How Claude Code is changing the way you create software
New AI-powered tools make it so writing code starts to feel like dictating ideasrather than tedious typing of syntax. Claude Code, developed by Anthropic, is one of the first solutions that really gives you the feeling that anyone can create a working programusing only natural language. Previous platforms of this type existed, but often required technical knowledge to go beyond simple designs. The key here is that the tool quickly understands problems and can correct them based on a short description.
Claude Code also stands out in that it was designed as a system “helpful, honest and harmless”which translates into more natural interactions. Users emphasize that talking to him resembles a dialogue with a competent colleagueand not with a tool requiring precise commands. At the same time, there is competition on the market – OpenAI, which previously created Codex, has returned to the game, developing its own solutions and making them available in ChatGPT. This led to tensions between the companies, especially when Anthropic restricted OpenAI’s access to its models, citing commercial use rules.
In recent years, Claude’s capabilities have grown rapidly. Increasing the context to 100k tokens allowed him to analyze huge chunks of code and documentation at once, and subsequent updates got the tool going create even complex websites and applicationsincluding 3D designs. As a result, some engineers started talking about “vibe coding”i.e. programming based more on the description of intentions than on technical details. Some even say that most of their code is created today thanks to Claude Codeand the tool itself writes much of its own software.
A specific culture has also developed around the platform – from memes about “claudeholism” to jokes about the tool being “opiate for those who hate tedious coding”. Even little things like funny verbs in the progress bar (e.g. crunching, moseying, zesting), they create the impression that using Claude Code is not only work, but also pleasure. All this makes programming truly accessible for many people for the first time.
However, not everything is rosy
But who spent a lot time for coding with use AI agents (even in the SWE paradigm), for sure Is aware that you can also go for a ride with them. If It’s nice to prototype things with them, they understand the entire life cycle and implementation of software, it they can also be a pain.
On the one hand, they make direct changes to the code and can queue them in a longer sequence (understanding the cause and effect sequence thoroughly) and will repeat them until something works properly. On the other hand, sometimes they come from the wrong side, which means you can spend a lot of unnecessary time debugging the code. It depends on the personwhether he will pick up on the wrong suggestion at all, instead of taking his word for it and then waking up with his hand in the potty that something went wrong. AND probably looking at Microsoft’s recent adventures with Windows (and this company boasts that 30% of its code was created using AI), you can see the negative sides of this phenomenon with your own eyes.
