This may hold true for greenfields, a little harder to do with a million line, 15 year old legacy code base. Good luck with your credits watching your agent churn those those files over and over and over for each request.

And are you seriously suggesting you'll be able to do spec driven development via markdown files in that? If not, then you're missing the point entirely.

I very explicitly stated that if your codebase isn't written like a spec you cannot effectively use it for LLM driven development.

You can still edit the code via LLM, it's just always going to be a highly involved process with very careful reviews unless you're happy with regressions.