For my pet project, I’d say 90% of the code. At work - I almost don't use vibe coding because not all tools are whitelisted, so I just use Copilot for autocomplete, tests, and documentation.
When this first wave of vibe started - I experimented and made 2 features like that without large editing. Despite my prompts, code was inconsistent. From naming and declarations to controls selection. On review (I work with other devs) it was clear that it’s written not by Senior dev🥲. Besides that awful moment, use only mock generations, boilerplate in fixed context and alternative version for hard tasks. Regular (in current moment) usage.
And btw, I’m using copilot as a code reviewer in PRs in my pet projects. That’s funny when it finds code improvements for the code that was implemented by AI 😂
My approach: try to do anything without writing code, even if writing code is faster than prompting. It’s not efficient, but you understand better how to write the right prompts. After a couple of screens, I’d say that prompting can be more efficient than coding itself. And I’m principal SE, so my coding skills are higher than Junior/Middle, but they can implement features with the same speed as I, using the Vibe Coding approach. When you use this approach, you describe logic using your language, instead of programming, so prompting isn’t about “do this and that”, it’s about programming without Swift.
Vlad, how much of your code is “vibe coded”?
It depends 😂
For my pet project, I’d say 90% of the code. At work - I almost don't use vibe coding because not all tools are whitelisted, so I just use Copilot for autocomplete, tests, and documentation.
What about you?
When this first wave of vibe started - I experimented and made 2 features like that without large editing. Despite my prompts, code was inconsistent. From naming and declarations to controls selection. On review (I work with other devs) it was clear that it’s written not by Senior dev🥲. Besides that awful moment, use only mock generations, boilerplate in fixed context and alternative version for hard tasks. Regular (in current moment) usage.
And btw, I’m using copilot as a code reviewer in PRs in my pet projects. That’s funny when it finds code improvements for the code that was implemented by AI 😂
Interesting concept
My approach: try to do anything without writing code, even if writing code is faster than prompting. It’s not efficient, but you understand better how to write the right prompts. After a couple of screens, I’d say that prompting can be more efficient than coding itself. And I’m principal SE, so my coding skills are higher than Junior/Middle, but they can implement features with the same speed as I, using the Vibe Coding approach. When you use this approach, you describe logic using your language, instead of programming, so prompting isn’t about “do this and that”, it’s about programming without Swift.