>>105955503
I think there are two definitions of it currently. They both involve telling programming agents (Claud Code, Cursor, whatever) to write code with little to no oversight or review.
Type A vibe coding:
User is an experienced software engineer who would able to write the program themselves and/or would be able to carefully review and amend all AI-generated output. But instead of doing that, they decide to trust the agent and just auto-accept everything. They'll probably at least skim through the code they're generating, keeping an eye out for obvious issues. They'll set up tools like linters and test suites to give some level of reassurance.
Type B vibe coding:
User is not a programmer and would have no chance of writing the program without AI. They have little to no understanding of the generated code. Despite this they are able to build something functional, maybe even useful. High chance of security/correctess/performance issues.