Separating stable engineering rules from feature-specific instructions seems especially important here.
I’d probably keep the “best practices” document for invariants, then record architectural decisions separately with why a choice was made and what would justify changing it. Otherwise the master document can become another source of stale instructions as the product evolves.
Do you treat architecture decisions as part of the same document, or keep them as a separate history?
The interesting part is that AI reduced the cost of rebuilding enough for you to optimize for disposability rather than permanence.
Keeping posts as flat Blade files and removing the DB/CMS also gives the agent a much smaller system to reason about. I like the “manual first, automate later” choice too — it avoids adding deployment machinery before the workflow proves it needs it.
Do you think you’ll keep the architecture intentionally disposable even if the blog grows?
That separation makes sense: keep the product vision stable while allowing the implementation to evolve.
The tricky part seems to be preventing the master document from drifting away from the code.
Does your periodic review also check for that drift, or do you update the document as part of each architectural change?