4 sessions on pandas with a coding agent: the agent failed at least once in 3, and 0 finished without a human stepping in.
Takeaway: I'd start by pinning the Python/Jinja2/Starlette versions up front, since most of the debugging time went into a Python 3.14 + Jinja2 3.1.6 weakref bug and a Starlette TemplateResponse signature change that had nothing to do with my actual spec.
Takeaway: Next time I'd specify upfront that action plans must be pasted in manually and that nothing should call the model automatically — I went through three different generation strategies (UI export removal, background auto-generation via Qwen, then placeholder-only extraction) before landing on what I actually wanted.
Takeaway: I'd start by telling the agent upfront that the local venv is missing deployment-only packages (torch, transformers, spaces) so it doesn't waste time discovering that itself.
Takeaway: Next time I'd front-load the dataset ID and scope questions in the first prompt so the agent doesn't have to burn a dozen turns guessing at what I meant by 'seed_dict' vs full retrain.