Replying to
@deprecated_ii@poa.st I'm doing some AI experiments with a local Qwen 3.6 for a pretty simple site. I already had a big spec document detailing all the flows, screens layout, error conditions, table schemas etc. I start a new chat and just keep getting it to read the spec doc and some ascii diagrams and saying 'find any inconsistencies or potential problems'. It's been pretty good at spotting gaps ngl.
Also been lurking the /g/ local LLM/slopcoding generals and they can't agree if it's better to write a big spec up front or start with a small idea and build up from there. Got it to spit out some wireframes and static simulations and its generally pretty good but constantly gets small things wrong like a closing brace or tag.


