Comment by techblueberry
19 hours ago
When you say with no coding experience, what does that mean to you? I can get near miraculous results from vibe coding, but it often gets stuck in weird “bug loops” where it goes back and forth between broken states, and I have to understand either like bracket formatting, or be able to research library failures and conflicts. I suppose with no coding experience I could maybe muddle through, but itw would require hours of patience and probably learning some coding fundamentals to do something I can identify in sometimes seconds.
Literally could only do basic html css before. Now I can build solid Sveltekit web apps from scratch. It’s still hard and frustrating. I’ve spent all day trying to get OTP auth working it’s very frustrating it misses things. But other times you can one shot whole features. Supabase MCP has been an absolute game changer. Database fluidity in development is really key. Trying to push migrations and generate types manually beforehand add so much time.
> I can get near miraculous results from vibe coding, but it often gets stuck in weird “bug loops” where it goes back and forth between broken states, and I have to understand either like bracket formatting, or be able to research library failures and conflicts.
In my experience this is mainly caused by a lack of investment in tests.
Vibe-coding excels when paired with test-driven development, because TDD approaches serve as validators and problem constrains. Often coding agents get stuck on but loops because they have neither context nor feedback on what represents a broken state. Tests fix both problems, and if you stop to add one then your bug loops quickly vanish.
So vibe coders need to know how to write tests? I doubt that lowers the effective barrier of entry to coding very much.
I assume you can't trust the LLM to write these tests, since you are writing tests so the LLM will stop it's bug loop...
> So vibe coders need to know how to write tests?
Inasmuch as they need to know how to write code.
This has been my experience as well. I've gotten really good results from Claude Code, but sometimes I have to press esc when I see it doing something stupid and course correct. It really seems to struggle with recursive solutions to problems or deeply nested constructs. Because I have a programming background, I know better when to interrupt and provide different directions or know when to abandon contexts better than someone without "coding experience" ever could. Just the years and years of debugging experience on your own projects will position you infinitely better than someone who has never really had to dig into complex problems and figure out what's going wrong. Sometimes you have to tell the agent specifically which logs to look into or which likely problem spaces it's encountering and knowing what sort of logs exist on a system is important for that.