From Vibe Coding To Vibe Engineering (AI column)
Watched the humorous. speech Kitze (Kristian Ristowski)at the end of 2025 At the AI Engineer Summit. The speaker is a prominent front-end developer and entrepreneur in the community who is known for his sarcastic approach to modern web development, which he criticizes the excessive complexity of tools. (Webpack, configurations) It promotes the "Just Ship It" approach. In this report, Kitze argues that "vibe coding" is a fun way to create pet projects, but for real work you need to turn it into "vibe engineering." When it comes to theses, here they are.
1The spiral of death (The Doom Loop) The problem with vibe coding is that as soon as you start manually editing code written by AI, you break the wib. You're wasting time reading someone else's (machine) You try to understand it, you edit it, and then the next generation, the AI can erase your changes or stop understanding the context. The solution from Kitze is that if the AI wrote the bug, don't fix it with your hands. Get the AI to fix it. It takes discipline, but it's the only way to keep pace. You have to become a manager who doesn't do the job for an intern, but makes the intern redo the task correctly.
2Context is everything. Ordinary vibe doding breaks down on large projects because AI “forgets” or doesn’t know the structure of the entire project. And “vibe engineering” is already building infrastructure around LLM, which automatically feeds models. The full context of the project (File structure, types, documentation) before every request. Kitze shows that instead of writing code, it now writes scripts and prompts that gather context for AI. This is the new engineering.
3From syntax to semantics A developer ceases to be a “code writer” and becomes an “intent architect.” Your job is to clearly describe what should happen, and how It'll be written. (on React, Vue or pure JS) - becomes secondary.
Interestingly, the author suggests the evolution of the name and approaches: "Vibe coding": I asked Claude to make a snake, and it works. It's fun. It's a chaotic process, a random outcome. “Vibe engineering: I built a pipeline where the AI has access to my entire code base, knows my linters and tests, and I can reliably generate complex features by controlling the process through prompts and tests.” It looks like engineering. Kitze argues that we need to move from playing with AI to professionally embedding AI into engineering processes.
If we think about how the proposed changes affect the industry, we get the following:
Vibe coding is dangerous for beginners. If you cannot verify the result (Don't understand architecture.)You'll create a monster. Beginners now need to learn not syntax for loopsystem design and quality control.
An experienced engineer becomes super-productive. If he knows. what With the help of vibe engineering, it can replace the entire command from the 3-5 man. Kitze cites an example of how he rewrote huge chunks of Sizzy in days, not months (sizzy is a browser for developers and qa engineers)
DevEx is changing. Previously, we optimized syntax highlighting and auto-addition. Now DX is how easy your project is to "read" by a neural network. If your code is incomprehensible to the LLM, you lose.
Well, the one-line conclusions from the report sound like this. - Don't write the code, control it. The best code is the one you didn’t write with your hands. - Invest in context.. Learn to create tools that allow AI to “see” your project as a whole. Resist the urge to "fix your hands quickly". It's a trap that brings you back to the old paradigm and slows you down in the long run. - The code becomes ephemeral.. We're used to saving code. In the era of vibe engineering, code can be easily deleted and regenerated if it is no longer compliant. Value is in the task description, not the code file.
#Engineering #AI #Software #ML #DevEx #Productivity #IDE