Comparison
OCR turns text in an image into characters. Document AI can add extraction and retrieval on top, but for scanned files it still depends on OCR quality.
Definition
OCR is often an input to document AI, not an obsolete alternative to it.
The primary requirement is accurate text capture or a searchable scan, with downstream review handled elsewhere.
The workflow also needs structured extraction, semantic retrieval, source-linked answers or repeatable agent steps.
Pick a real piece of work, agree what a good answer looks like, then go through the results together.