Can Claude 3.5 Sonnet Replace Coding?
I've been hearing a lot of buzz about Claude 3.5 Sonnet lately, so I gave it a try—and I was genuinely amazed. It wrote simple CRUD APIs almost perfectly. However, when it comes to complex business logic or legacy code analysis, it still seems to have limitations. Has anyone actually replaced their work with Claude in practice? I'm a junior developer, and I'm considering it because I feel like using this could double my productivity.
9 answers
I recently tried Claude 3.5 Sonnet too, and it's definitely convenient for CRUD-level tasks. But I think analyzing legacy code is better done manually. For juniors, building code comprehension skills is more important.
We are partially adopting it in practice. Simple API or test code writing is mostly delegated, but I believe business logic still needs to be designed by humans. Productivity definitely increases, but full replacement is not yet feasible.
I'm a junior too, and I've been using Claude to write code and then reviewing it, which has really helped me improve. I've learned a lot by analyzing the code AI generates. Saying productivity doubles doesn't seem like an exaggeration.
I tried using it for legacy analysis, but large-scale projects are difficult due to context length limits. Still, if you break it down into small module units and ask, it's quite helpful. Think of it as an auxiliary tool rather than a replacement.
Claude 3.5 is great. But if you're a junior, be careful not to become too dependent on it. I was excited at first and relied on it heavily, but later realized my fundamentals were lacking. The key seems to be using it in moderation.
Have you considered security or licensing issues when using Claude? It might be restricted depending on company policy. We can't input sensitive code, so we only use it for public API tasks.
I'm a full-stack developer, but I delegate most frontend component creation. However, I find it more reliable to write backend business logic myself. It feels like about a 1.5x boost in productivity. For juniors, it could be even higher.
I'm in my third year, and after adopting Claude, code review time has been cut in half. Since it writes clean base code, there's less to fix. However, it's still clearly limited when it comes to complex exception handling.
Can it be replaced by Claude? I actually found GPT-4 to be better for coding... Sonnet gives good explanations, but the code quality was a bit lacking. If you're a junior developer, I'd recommend comparing multiple models.