Research email and Slack history before drafting a customer reply

The author, a solutions architect at Databricks working with public sector customers, connected Databricks Genie One to their Gmail, Slack, Google Drive, calendar, Confluence and Jira over MCP, running under their own accounts. When a question arrives they paste it in with a one-line request to research it. In one case a federal customer asked three questions deep in a long thread. Genie One found the thread, read earlier messages including the author's own reply, searched product docs, the internal wiki and Slack, answered each question with a pros and cons table, then drafted a reply after rereading the author's earlier email for tone. They also used it to find a Slack request with a Friday deadline and to check old meeting notes.

The customer email took 14 lookups and a little under three minutes, and the draft about 40 seconds. The draft recommended an approach the author had already told the customer might not work across two clouds, so it had to be corrected before sending. The Slack task took about 80 seconds and caught that a second project did not fit the request at all. On a slide listing which agencies share data, it cited private chats between the author and a colleague as evidence and credited one initiative to the wrong vendor; a second pass with stricter sourcing gave a shorter list the author could defend.

The author works for Databricks, which sells Genie One, so this is a vendor employee's account. Tell it what counts as evidence, since it will treat your own messages as proof of your own claims, and read every draft against what you already promised. Answers take from under a minute to a few minutes, so it suits asking and switching tabs.