LLMs have a hard time producing simple code. One of the reasons is that determining what is "simple code" requires a lot of context that lives *outside* the code & docs. It depends on your team dynamics, your company business strats, your suppliers or the latest trends. Peter Naur's paper "Programming as Theory-building" (1985) states this knowledge in devs minds **is** actually the program, the Theory w/ capital T as calls it. LLMs are increasingly solving & automatic code & docs, but that's just a small piece of the full cake. Being aware of it helps understand many LLMs failure modes and also how to more effectively leverage them. To learn more watch my talk @PyBCN last week! youtu.be/LwchtmM-d-s

Thanks to @PyBCN & @codurance_ES for organizing yesterday's Python Meetup and @JavierMarAlc for his opening talk! For those interested, here's a pre-recorded version of my talk and all the links to code & slides below 👇

Medium Blog post: - medium.com/@pol.avec/from-te… Slides: - gamma.app/docs/from-text-to-… Talk2APIs Demo: - github.com/kafkasl/talk2apis Replit Agent demo: - replit.com/t/fewsats/czjf93/…