I know that agentic harnesses are popular in software development these days. But they are not the only way to generate code with AI. If I have a specific function in mind, I will sometimes just prompt for it in chat, and then paste it in.
I have found this works well. And it is worth trying different LLMs—I have found that occasionally a local model like Gemma 4 can generate a better function than a cloud-based model.
It does make me realize that:
• For string manipulation, having an LLM generate a function is way easier and faster than writing one by hand.
• String manipulation (and similar) tutorials are probably not going to be that useful in the future.
It is possible to piece together a program with a chat bot's output, and for some situations this probably works better than an agentic harness. It allows me to do more work locally—which means I cannot have my account canceled.