This is really a lot less necessary with GPT-4. What required careful prompting in 3.5 often you can give it something slapdash in 4 and it can do a great job figuring out intent.