We solved meeting transcription. Did we solve the meeting problem?

I had a meeting recently where the conversation itself was useful, but what happened afterward wasn't.

We had a recording. We had notes. We had messages exchanged afterward.

Yet there was still uncertainty around what had actually been decided, who owned which task and what needed to happen next.

This made me think about the way we approach meeting technology.

Transcribing a conversation tells us what was said. It doesn't necessarily tell us what matters.

A business meeting contains structured information hidden inside natural conversation:

decisions
commitments
action items
customer requirements
objections
urgency
buying signals

The interesting technical challenge, to me, is turning that unstructured conversation into something actionable.

I'd love to hear from developers working with LLMs or meeting intelligence: what information do you think is hardest to reliably extract from real conversations?