IME vibe coding works spectacularly for simple one-off scripts that do a fairly simple thing. I can describe the work in ~10 minutes and save 2-4 hours of scripting.
Just last weekend I wanted a script to process a csv file and produce some reports and graphs out of that. I think it would've taken me the 2-4 hours to write it myself. Instead, I had cursor write it while waiting for boarding at the airport, probably no more than 10 minutes.
For codebases anything more complex than that, it starts to fall apart pretty quickly.
In that scenario it works ok only if I do all the work of designing the system and the functions and only let it type in the code for individual strictly-defined functions. So it does save some work which is nice, but it's not a huge win.
Yeah, that matches my experience: LLMs are amazing “script interns” and shaky “systems engineers.” One trick that helps on bigger stuff: force it into a tight loop of (1) write a failing test for a single behavior, (2) implement smallest change, (3) run tests, (4) refactor. When you make the unit of work “one green test,” the model’s tendency to wander gets way less destructive.
Just last weekend I wanted a script to process a csv file and produce some reports and graphs out of that. I think it would've taken me the 2-4 hours to write it myself. Instead, I had cursor write it while waiting for boarding at the airport, probably no more than 10 minutes.
For codebases anything more complex than that, it starts to fall apart pretty quickly.
In that scenario it works ok only if I do all the work of designing the system and the functions and only let it type in the code for individual strictly-defined functions. So it does save some work which is nice, but it's not a huge win.