AI that answers, acts, and stays governed — the moment you turn it on.

She stands up the configuration that puts AI to work — activating the shipped Now Assist skills your teams see, authoring custom skills and agents where you need them, wiring the search that grounds every answer, and setting the governance posture. The conversations and agent runs stay in your team's hands; Phyllis configures the capabilities that power them.

No — she activates first. Common asks like summarization, resolution notes, and conversational intake already ship as Now Assist skills, so she tells you which are present but switched off and turns them on. A custom skill or agent is proposed only for what nothing shipped covers, and she tells you what she checked.

By grounding them in your own content. Phyllis wires search across your surfaces to point at your live knowledge, catalog, and records, so answers reflect what's actually in your instance. She also flags the steps needed for it to take effect, so a surface never silently returns nothing.

Through the AI Control Tower. Phyllis sets your approval posture — whether new AI assets trigger a review and require sign-off before production — assigns the steward and workspace roles, classifies risk, and turns on the feeds that inventory AI running across your environment. Guardrails for offensive content and prompt injection are configured for every deployed capability.

Correct. Phyllis discovers, proposes a plan, and only builds on your explicit approval, staging every change in a reviewable package. Skills are turned on incrementally, agents default to human-supervised, and risk classification is always your call — she never guesses a severity.