Quick note on using AI for dev work
Lately I’ve been testing different AI models for coding help, especially in AI-powered editors. I tried Claude 4 and Gemini 2.5 Pro which are considered as “advanced” models, but honestly they kept messing up even simple tasks. The generated code had many bugs, linter and tests failed, and many instructions were ignored. Today I switched back to Claude 3.5 (which I used a month ago), and suddenly everything works again. Same prompts and tasks, but much better results. So here’s my takeaway so far: - Use Claude 4 or Gemini for deeper discussions, brainstorming, or architectural decisions. - But for coding and implementation, Claude 3.5 is still the best in my experience. - Deepseek is a decent second choice for actual coding tasks. Curious if anyone else is seeing similar results with these models? Let me know what combo you’re using 👇