What I like best is being able to set up a separate agent for each workstream or client engagement instead of relying on one general-purpose assistant.
Each agent can be scoped to that engagement's specific metrics and definitions, so when I switch between project's the answers stay consistent with how that client actually defines their numbers, not a generic interpretation.
Setup for each one took about five minutes, which matters when you're juggling multiple engagements at once. Review collected by and hosted on G2.com.
The response time can slow down noticeably when I switch to Deep Dive mode on a more complex question, especially if it's cross-referencing multiple databases for different queries. For a quick look during a client call, this isn't the issue, but if I am building out something more layered for a deliverable, I have to plan for a bit of wait time rather than expect an instant answer. Review collected by and hosted on G2.com.