Agree with this series of posts. Long term maintainability is one of the difficulties of vibecoding and friends. For a small one-off project or a prototype, that might be fine. If you are hoping for the project to still be used a year or even ten years later you could have problems.
My take reading through some of these examples is that for non-experts, generative AI vibecoding is to programming what Dreamweaver was to web design back in the day—they give you the short-term code artifact (which is neat, so far as it goes) but they don't create long-term support infrastructure
One of my CS professors helped me consider long term maintenance for all projects. He asked things like:
- What parts of the project are hard to understand or sloppy?
- If you got a bug report in a few months, how long would it take you to find the cause?
2/
- Will it still work if you upgrade hardware, OS, or language?
- Will your docs make sense to anyone else or yourself in a year?
- How hard would it be to add a new feature in a year?
Not everything needs this analysis. When others use your project or security is relevant it is a good idea.
3/
- Do you take untrusted user input? Then you probably want to update packages as vulnerabilities are found
- How will you know about vulnerabilities or upgrades? You might want to use code analysis tools or follow a feed for the package/language/tool
4/
These things are challenging from the start. I feel like rapid code generation and less upfront planning may increase the difficulty. But that is just my vibes.
Anyway, that is a long winded way of hoping people are careful.
Also that I am a perfectionist.
5/
Jun 6, 2025 23:20
