Imagine Descript being able to identify likely proper nouns in transcripts and, instead of just trying once (and usually failing) to transcribe the noun accurately, doing an extra pass including web search to really try to figure out who or what is being talking about. It could pull in a few of its best guesses for each identified noun that the user could choose from to confirm/dismiss. Maybe it could even pull in URLs to wikipedia or something similar that get attached to the word or phrase as a comment.
Failed proper noun transcription is the biggest hurdle between AI-generation transcripts and useable transcripts for public consumption and SEO.