Skip to content
KiweelyKiweelyKiweely

Stop or audit a Workday AI agent: an admin runbook

When an agent must stop, deactivate it in every tenant where it was set up. Hiding it is not enough if Production will refresh Sandbox. Use View AI Agent User Audit Trail and View Security for Agent Skill to see what it did and what it can still do.

  • Checked against Workday on
  • Applies to Workday Extend
  • 7 min read
  • Published
  • By Kiweely
On this page

Stopping a Workday-built agent is a per-tenant action in Agent Management Hub. Deactivate hides the agent from users and leaves it in the hub so you can activate it later. For Deployment Agent, Workday also says Flex Credit consumption stops in that tenant. If you configured the agent in Production so Sandbox refreshes keep the setup, you must deactivate Production too. Otherwise the next refresh turns the Sandbox agent back on.

You need Manage: Agents and Setup: Agents to deactivate Deployment Agent. Those domains are part of Agent System of Record. View AI Agent User Audit Trail is the report the ASOR setup topic names under Reports: AI Agent Security.

This runbook is the operational sequence. It does not invent a kill switch beyond deactivate, skill and tool configuration, and security activation.

Decide what “stop” means

Goal Do this What remains
GoalNobody can chat with Deployment Agent in this tenant Do thisDeactivate Agent on that tenant’s registry row What remainsThe agent stays in the hub. Chat history already stored follows the 14-day retention. Histories are per tenant.
GoalStop Flex Credits for Deployment Agent everywhere Do thisDeactivate every tenant you configured, including Production if it exists only for refresh What remainsA later activation starts a new period of use.
GoalStop one Self-Service action, keep the agent Do thisConfigure tools on that skill. Some tools cannot be disabled. The Enabled toggle is unavailable for those. What remainsThe skill’s other tools still run. Skill-level billing for tools you leave enabled does not drop just because you disabled one tool.
GoalStop a population from using a skill Do thisRemove that security group from Available To What remainsOther groups on the skill still can. Users need a new conversation before tool changes apply. The skills topic says skill and tool changes can take up to 10 minutes because of caching. The configure-tools topic says up to 5 minutes for new conversations, and an open conversation might not change until a new one starts. Plan on the longer window.
GoalSee what it already did Do thisView AI Agent User Audit Trail What remainsDeactivate does not delete the report.

Disabling a tool does not make the agent read your sentence verbatim. You can set a Tool Disabled Message, a Supplemental Message, or a Fallback Message. The agent uses that text to write its own reply. The configure-tools topic’s example: you disable a bonus-history tool and write that people should use the Bonus Management Portal, including a link. The user sees a generated reply based on that message, not the message itself.

Deactivate Deployment Agent

  1. Confirm you are in the tenant you mean to stop. Production and Sandbox are different stops.
  2. Open Agent Management Hub.
  3. Open Agent Registry.
  4. Open Deployment Agent.
  5. Choose Deactivate Agent.
  6. Confirm if prompted.

Result, in Workday’s words: the ASOR-enabled Deployment Agent is no longer available in that tenant, and Flex Credit consumption stops there.

Repeat for each tenant in the table. Customer Central is its own tenant. A connected Implementation tenant is another.

Tenant Why it might still be on Stop action
TenantImplementation Why it might still be onYou tested setup here Stop actionDeactivate here
TenantPreview Why it might still be onSeparate ASOR configuration Stop actionDeactivate here
TenantSandbox Why it might still be onActive for implementers Stop actionDeactivate here
TenantProduction Why it might still be onConfigured so refresh copies the agent into Sandbox. Not usable as Production chat, but still a configuration that reactivates Sandbox Stop actionDeactivate here or the next refresh undoes the Sandbox stop
TenantCustomer Central Why it might still be onSeparate setup Stop actionDeactivate here if it was registered

Write down the comment or the change ticket, the tenant, and who confirmed. The hub keeps the agent for a later activation. “Removed” is the wrong status word.

Audit after the stop

ASOR permissions for the audit report are View and Modify on Reports: AI Agent Security for an unconstrained group, plus the domain enabled in the System functional area. If the report is missing, fix that security before you conclude there is no history. The setup topic’s example report is View AI Agent User Audit Trail.

While you are in the registry, open the agent profile if your process includes the passport and stamps. A related walkthrough of passport stamps and the audit trail is in the Agent Passport guide. This runbook only depends on the tasks the admin guide names.

For “what could it still do if we turn it back on,” run View Security for Agent Skill. You need Manage: Agents. The report prompts are Registered Agent and Agent Skills.

Functional status Meaning What you do before any reactivation
Functional statusFull Match MeaningThe Available To group has all required policies for the tool. What you do before any reactivationAccept only if that group should have the tool.
Functional statusReview Required MeaningThe group has some required policies, not all. What you do before any reactivationReview the group before you activate.
Functional statusConfig Required MeaningA required policy has no security groups. What you do before any reactivationConfigure those policies first. Workday says to prioritize these.
Functional statusNot Evaluated MeaningAvailable To is empty, so there is no group to test. What you do before any reactivationDo not treat an empty skill as “safe and ready.” It is unevaluated.

Required Permissions are View or Modify. Recommended Security Policies and Assigned Security Group show how the tool is supposed to be reached. Additional Security (Optional) is field-level access that is not required for the tool to run. Test My Setup checks a security group and shows other policy paths. It needs Security Administration.

Clear Config Required rows before you activate again. Then Test My Setup for a real user in each Available To group.

Self-Service Agent: narrow it before you deactivate it

Deactivate is the full stop. If the problem is one action:

  1. Open Agent Management Hub, Agent Registry, Self-Service Agent.
  2. Configure the skill’s tools.
  3. Turn off only the tools whose Enabled control is available.
  4. Set a disabled message, and a fallback message for failures.
  5. Wait out the caching window and start a new conversation as a test user.
  6. If the skill itself should not exist for that population, clear Available To rather than hoping a disabled tool covers every action in the skill.

The Guide skill has to stay on for Self-Service Agent to operate. Turning off Guide is not a partial stop. It stops the agent from running. Policy Intelligence no longer requires the Workday Help SKU, so do not assume that skill is absent because you did not buy Help.

Runtime still checks the person’s access to the agent and their access to the tools’ domain or business process policies. Removing someone from Available To does not remove them from a domain. Do both if the domain was opened only for the experiment.

Reactivate only with a written reason

Activation is the same hub. Workday isolates Deployment Agent chat histories by tenant and deletes them after 14 days, so a thread from before the stop may already be gone. Reactivating does not restore that chat. It does make the agent available to the Available To groups again, and for Deployment Agent it is again a tenant that can consume Flex Credits until you deactivate it.

If Production was reactivated “for the next Sandbox refresh,” say that in the ticket. Otherwise the next incident will look like Sandbox turned itself back on.

Frequently asked questions

Does deactivate delete the agent?

No. The deactivate topic says Workday hides it from users and keeps it in Agent Management Hub for a later activation.

Will a Sandbox refresh turn it off?

A refresh copies configuration from the tenant you refresh from. If Production still has the agent active, Sandbox comes back active. Deactivate Production when the goal is a full stop.

Is the audit trail the same as thumbs up and thumbs down?

No. Thumbs rate a Deployment Agent or Self-Service answer. View AI Agent User Audit Trail is the security report named on the ASOR setup topic. Use the report for “what ran.” Use thumbs for “the guidance was wrong.”

Can I disable a single tool for one security group?

No. The configure-tools topic says you cannot assign tools to specific security groups. Tools inherit the skill’s groups. To split populations, use separate Available To lists at the skill level, or do not enable the skill for that group.

Sources checked

Frontmatter pages, fetched 2026-11-16.

Sources checked

Related guides

All guides

Let Workday check the work.

Kiweely writes WQL and builds orchestrations in your own tenant, and Workday, not Kiweely, says whether they passed.

Get the next guide in your inbox.

One email a month with new guides and one Extend tip. Nothing else.

We’ll email you once to confirm. Unsubscribe in one click from any issue.