Ask HN: Is vibe coding viable for real work or operating products/services?
I keep seeing praise for vibe coding everywhere, so I’ve tried it myself a few times. While it works fine for simple scripts or mockups, it seems to hit clear limitations when complexity increases or when you need something more substantial. Am I missing something here? I’d love to hear your experiences. For context, my attempts have been successful with straightforward automation tasks and quick prototypes, but when I’ve tried to build more complex systems or anything requiring robust error handling, maintainability, or integration with existing codebases, the approach falls short. The generated code often lacks the architectural considerations and edge case handling that production systems require. Has anyone successfully used this approach for serious development work, or is it primarily a tool for rapid prototyping and learning? What’s been your experience with scaling AI-assisted coding beyond simple tasks?
There is a reason it is called vibe coding and not vibe architecting or vibe systems design. I have had better results when I come up with the overall architecture first. Then I'll vibe code to stub out classes and methods, and then vibe code the tests, then vibe code to fill in those stubs, making sure the tests pass. You have to handhold the AI like you would an intern or junior dev.
Trying to vibe code a gcc plugin. That A"I" is worse than the most simple tutorial and I end up reading the gcc docs and headers myself. Ah well, at least I got some training now.