Same here. I’ve had limited success getting AIs to do very simple stuff. Every one I’ve tried invents APIs that don’t exist and eventually get stuck in a circle where it tells me to try A. When that doesn’t work, try B. No luck? Try C. Hmmm my apologies, try A. Lather, rinse, repeat.
No, I haven’t tried that yet. I don’t really want to turn on auto mode when it’s iterating on my credit card and it looks like it’s in an infinite loop… Is that a silly thing to be worried about?
I work mostly in C++ (MFC applications on Windows) and assembly language (analyzing crash reports).
For the C++ work, the AIs do all kinds of unsafe things like casting away constness or doing hacks to expose private class internals. What they give me is sometimes enough to get unstuck though which is nice.
For crash reports (a disassembly around the crash site and a stack trace) they are pretty useless and that’s coming from someone who considers himself to be a total novice at assembly. (Looking to up my x64 / WinDbg game and any pointers to resources would be appreciated!)
I do prototyping in Python and Claude is excellent at that.
Can you not just use Claude Code with a plan and cap your spending that way? I am using just the £18 plan, and it is quite helpful already. If you haven't tried Claude Code yet, just do it, experience agentic coding, and afterwards we can talk.
>> No, I haven’t tried that yet. I don’t really want to turn on auto mode when it’s iterating on my credit card and it looks like it’s in an infinite loop… Is that a silly thing to be worried about?
Yeah. Most AIs today are pretty good at detecting that they're in a loop and aren't making progress. When that happens, they either take a different approach, or stop and say they are stuck. But, if you're really worried about it, you can cap monthly spend on the billing page of virtually every AI provider.
