Governing AI agents like teammates

The demo always looks flawless: an agent triages a ticket, updates the CRM, drafts a proposal, and routes it for approval, all in seconds. Then someone asks how fast it can roll out everywhere. That’s exactly when the risk begins.

You buy the licenses, flip the switch, and productivity follows: that’s the SaaS playbook enterprises have run for two decades. Agentic AI breaks it. Unlike a standard LLM, which only carries content risk (a bad draft, a hallucinated fact), an AI agent can take action inside your systems, updating records, issuing payments, and changing live data. Built around Harvard Business Review research, this session walks through real failures, including an AI coding agent that deleted a production database and a prompt-injection exploit that turned a web form into a data exfiltration channel, and what each one means for how agents must be identified, monitored, and governed.