Raw LLM Responses

I have been using AI like chat gpt 3.5, 4o, claude sonnet 3.7 as well as some open source like qwen 2.5 coder. They all are doing great job for writing initial code but… none of these AI models could do it job 100% without any flaws. Sometimes it is a simple optimisation, security issue, using deprecated external libraries or simple logic mistakes. You can of course spend time iterating with chats to refine the final solution but you have to be experienced developer to spot these issues. If you are noob, you will fix runtime errors eventually but running code in production will give you bad time sooner or later. I think AI can dramatically reduce the time to write the code, especially if we need to deal with some new area we do not work with often and AI can do 90% of the work for us. Or with writing boilerplate or mapping big json to a set of classes etc. Or if you are experience developer and learning new language. AI is a blessing. AI is also great companion for a code review. I recently had a case where chat gpt found a bug in my unit tested code. Conclusion: AI is like any other tool to improve our lives significantly. However I do not see any time soon replacing developers. I think the demand in developers can even increase as more software is produced and eventually someone will need to work together with AI to fix many issues.