Good: extraction and structure
Pulling fields off a card, turning rambling notes into a summary with clear action items, spotting the sentence where you promised something — this is exactly the kind of work language models handle well.
It is high-volume, low-judgement and easy to verify at a glance, which is the ideal profile for automation.
Bad: writing the message for you
Generated outreach reads like generated outreach, and the people you most want to reach receive a great deal of it. The tell is not grammar; it is the absence of anything only you could have written.
Use AI to remind you what you discussed. Write the two sentences yourself.
Be careful with enrichment
Automatic enrichment can be wrong in ways that are invisible until you cite it back to someone. Confidently attaching the wrong company to a contact is worse than leaving the field empty.
Treat enriched data as a suggestion until something confirms it.