I’ve been experimenting with running LLMs locally, trying it with Opencode. (A sort of open source claude code like cli tool to run LLMs against your code base), because I refuse to give OpenAI or Anthropic my code.

I tend to test them with small tasks and then expand to bigger tasks.

Today I loaded the “Qwen3 coder 30B” model and asked it to convert a medium sized ReactJS project to Vue. It refused, because apparently it thinks it might allow me to create malicious code.

Yeah, I think programmers might be fine for the foreseeable future.