> there's a strong benefit from a framework or system that has been pressure tested in the real world with a lot of eyes on it - especially open-source ones that are the result of the accumulated judgement of many users
I agree I'd rather trust a project that execute 90% of community reviewed and tested code, and 10% LLM output, rather than 100% LLMs output. But at the same time, LLMs have a strong tendency to code as if frameworks are constraints rather than tools. They will pretend they use them, and then rewrite the entire handling of a get request to check the length of a user input rather than using your framework validation tools, giving you a false sense of confidence.
I agree I'd rather trust a project that execute 90% of community reviewed and tested code, and 10% LLM output, rather than 100% LLMs output. But at the same time, LLMs have a strong tendency to code as if frameworks are constraints rather than tools. They will pretend they use them, and then rewrite the entire handling of a get request to check the length of a user input rather than using your framework validation tools, giving you a false sense of confidence.