From onboarding to offboarding — what happens when your workforce isn't human?

AI agents are no longer experimental tools sitting beside human workers — they're becoming full participants in the enterprise workflow, executing tasks, making decisions, and accessing sensitive systems at scale. Yet most organizations are deploying these digital workers without the governance scaffolding that any human employee would require: no onboarding standards, no performance reviews, no access reviews, no clean offboarding.

This session reframes AI agent governance through a career path lifecycle — and walks through what a mature "Hire to Retire" model looks like for autonomous and semi-autonomous agents.

We'll cover how to vet and "hire" agents (capability assessment, identity provisioning, least-privilege access), how to onboard them into workflows with clear roles and accountability, how to monitor performance and behavior drift over time, how to manage role changes and promotions as their scope expands, and — critically — how to retire agents safely when they're decommissioned, replaced, or compromised.

Drawing from cybersecurity, identity governance, and operational excellence frameworks, this presentation gives leaders a practical mental model and a checklist they can apply Monday morning, whether they're managing two agents or two hundred.

Key Takeaways

1. Treat AI agents as a workforce, not as software. Every agent needs an identity, an owner, a scope of authority, and a documented purpose — the same accountability you'd require of any employee with equivalent system access.

2. Lifecycle governance is the missing control layer. Most AI risk today isn't from the model itself — it's from agents that were deployed quickly, accumulated permissions, and were never reviewed, rotated, or retired. A Hire-to-Retire framework closes that gap.

3. Human-agent collaboration requires designed handoffs. The highest-performing teams aren't human OR agent — they're human AND agent, with explicit decision rights, escalation paths, and review checkpoints engineered into the workflow