FWIW my experience is still that LLMs produce excellent code but always in need of a review:
- missing cases, values, features that were requested
- faulty code flow, not following the request made
- bloated code: hard to read and debug
Makes me very skeptical about vibe coding / agentic coding.
- missing cases, values, features that were requested
- faulty code flow, not following the request made
- bloated code: hard to read and debug
Makes me very skeptical about vibe coding / agentic coding.
Comments
- The quality of generated code depends on the language: SQL is OK as long as you accept complex joins instead of simple subqueries, C# is correct as long as you don't require a very strict coding style, PHP is always horrificly bad (no OO, no prep. stmts, no input checks, no MVC, just bad)