I analysed 17 years of fast food and coffee spending using OpenAI Codex

I wish there was more detail on verifying things like "code parsed PDFs to CSV". LLMs have trouble with arithmetic, so some verification should take place, but isn't mentioned.