AI agents need an offboarding process. Workday's deal for C1.ai treats it as identity governance

“An agent that was retired last quarter shouldn’t still be able to get in.”

The sentence comes from Workday’s announcement, on October 13, of a definitive agreement to acquire C1.ai, an identity platform that connects people and AI agents to business systems such as email, CRM, code repositories, finance systems and cloud infrastructure, with scoped access, enforced policies, human approval and a record of every action. It reads like common sense. In most enterprises, it is not yet true.

The status, precisely

The deal is announced, not closed. Workday expects it to close in the fourth quarter of its fiscal year 2027, which ends January 31, 2027, subject to customary conditions. No price was disclosed. The release gives no launch date for combined capabilities and carries the usual warning that unreleased features may change or not be delivered. Nothing about this reaches existing Workday customers today.

What Workday describes is a direction. Its Agent System of Record already registers agents and shows them. C1.ai would extend that from knowing an agent exists to governing what it can reach across Workday and other systems, with access driven by events Workday already tracks: hires, job changes and exits. Workday calls the result “the identity governance layer for people, money, and agents.”

Why an inventory is not enough

Most agent governance programs started with a list: which agents exist, who built them, what they are for. The list is necessary and, on its own, inert. The failures that matter happen in what the list does not control:

• The owner moves. The person who sponsored an agent changes role or leaves. The agent keeps running on permissions granted for a job nobody now holds.
• The task ends. A migration agent finishes its project; its credentials to production systems do not expire.
• Scope creeps. An agent built to read invoices is later given write access “temporarily” for one exception.
• Delegation stacks. An agent acting for a manager calls another agent, which inherits more than either person intended.

Each of these has an equivalent in employee identity, and enterprises spent two decades building joiner, mover and leaver processes to handle them. The point of Workday’s move is that agents need the same lifecycle, triggered by the same business events, because an agent’s legitimate scope is defined by the people and work it serves.

Where agents differ from employees

Agent lifecycle is not just employee lifecycle with a different badge. Three differences change the design:

1. Agents have sponsors, not managers. Every agent needs a named human owner whose own lifecycle governs it. When the sponsor leaves, the agent should be reassigned or suspended automatically, not discovered months later in an access review.
2. Agents work in tasks, not tenures. Short-lived, task-scoped credentials fit agents far better than standing access. A good default is that an agent’s access ends when its task does.
3. Agents act on behalf of others. An agent may hold its own identity and also a delegated one from the person it serves. Revoking one without the other leaves a gap, the distinction we drew in every AI agent has three identities.

What the statistics say, and who says them

Workday’s release cites research that 95% of large U.S. companies already run autonomous agents (C1.ai’s own report) and that only 21% have a mature way to govern them (Deloitte’s 2026 State of AI research). The first figure comes from the company being acquired; treat it as directional. The gap it describes is consistent with what we saw when Google gave coworker agents directory accounts and left their lifecycle to the customer.

The acquisition also continues a trend we have followed since HENNGE built a company around agent identity as the next IAM problem: identity vendors and systems of record are converging on the agent as an object with a lifecycle.

What to do before any deal closes

You do not need Workday’s future product to fix the basics:

• Give every agent a sponsor recorded in the system you already use for employee identity, and make a sponsor’s exit or transfer an event that suspends or reassigns their agents.
• Expire by default. Prefer credentials bound to a task or a time window over standing access.
• Review agents in the same campaign as people. Quarterly access reviews that skip non-human identities miss the fastest-growing population in the directory.
• Test the leaver path. Retire a pilot agent and verify, system by system, that it can no longer get in.

The last step is the one Workday’s sentence describes. Most enterprises will learn something from running it.