Working with agents generates a lot of markdown input and output. So I'm working on viva, a tool to help author and review docs. You tell it what kind of doc you want, attach any necessary links/files/repo paths/etc, then it asks you clarifying questions until it has the data it needs. Then it authors a doc and give you a review interface. You can approve, request more info, request changes, make direct edits, etc. T…
It feels like an impossible task. With little bips and bops being relevant across the whole project. One would have to scour through the entire document they are creating, for every bip and bop they want to write down.
Not like LLM could not do this. But, there always comes a point, where some important relation is skipped, or is just part of the tribal knowledge. Someone, somewhere just knows, that if X fails, he has to look in Z, but this fact is not noted in the knowledgebase. So, the re-write always falls apart.