Skip to content
KiweelyKiweelyKiweely

What a person still approves when Workday AI is turned on

Workday’s Self-Service Agent runs only after an admin registers it, enables Guide, and sets Available To groups. Several advanced skills call existing business processes. Deactivate, tool messages, and thumbs-down do not replace those human steps.

  • By Kiweely
  • 5 min read
On this page

“The agent handles it” is too coarse for Workday’s own setup pages. Self-Service Agent can retrieve data and take common actions, such as personal information, time off, goals, team learning, and policy articles stored in Workday. It does that only for people in the security groups you put on each skill, and only when those people also pass the domain or business process security on the tools. An administrator registers the agent, enables skills, and activates it. That sequence is the human approval Workday actually documents. This article does not add an approval step the admin guide does not name.

The admin actions that stay human

Action Who What Workday requires
ActionOpt in to UMSA when you are still on MSA WhoThe organization What Workday requiresWorkday-built agents need UMSA in non-production and Production. Self-built agents do not need UMSA in non-production.
ActionEnable Agent System of Record and activate security WhoAn unconstrained security group What Workday requiresASOR does not migrate. Constrained groups cannot be placed on ASOR policies.
ActionRegister the agent WhoThat admin, and only if the SKU includes it What Workday requiresUnregistered Workday Agent, then Agent Registry.
ActionEnable Guide WhoThat admin What Workday requiresThe agent does not operate without the Guide skill.
ActionChoose every other skill WhoThat admin What Workday requiresIncluding Policy Intelligence, which no longer needs the Workday Help SKU.
ActionSet Available To WhoThat admin What Workday requiresThis is the interaction policy.
ActionActivate the agent WhoThat admin What Workday requiresUntil then, people do not get the skills.
ActionActivate pending security policy changes WhoA security admin What Workday requiresDomain edits the agent told you to make are pending until this task.
ActionDeactivate WhoAn admin with Manage: Agents and Setup: Agents What Workday requiresHides the agent. For Deployment Agent, Flex Credit consumption stops in that tenant. Production must be included if it was configured for Sandbox refresh.

Employees do not perform that table. They use skills you left on.

Skills that still land in a business process

The skills topic is specific about manager payroll. MSS - Payroll - Advanced calls the existing Request Payroll Inputs business process, uses your payroll and security configuration, and respects the pay component security you configured. It creates inputs only for workers on Workday Payroll. It blocks the request when the primary position is in an external pay group or the worker has no pay group. For a multi-job worker, if the manager does not name a position, it defaults to the primary position on the input start date.

So a manager can ask the agent to start payroll input, and the business process you already configured still runs. Approvers on that process are unchanged by the existence of the skill. Condition rules on that process still use And, Or, and the processing user’s access. If a condition skips a step because the processing user cannot see a field, that skip is the process, not the agent “deciding.”

MSS - Human Capital Management - Advanced includes initiating job changes. MSS - Workforce & Frontline Management - Advanced includes approving, sending back, or denying submitted time. Those are actions a manager takes through the agent, under the security you granted. They are not evidence that a second, undocumented AI approval was inserted.

Guide sends the person to the Workday task or report. Landing in the task means the task’s own security and, where it exists, the task’s business process still apply.

What is not a human approval gate

Mechanism What it actually does
MechanismTool Disabled Message, Supplemental Message, Fallback Message What it actually doesText the agent rewrites into its own reply. The user does not have to accept a quoted policy sentence. Some tools cannot be disabled at all.
MechanismThumbs down What it actually doesOn Self-Service Agent, required on the latest reply if you want to add detail. It does not roll back a submitted request. The skills topic also says some skills only respond to “retry” or “check your work” in the conversation itself.
MechanismCaching What it actually doesSkill and tool changes can take up to 10 minutes. An open conversation may keep the old tools until a new one starts. That delay is not an approval.
MechanismFlex Credits What it actually doesThe public Flex Credits page meters Self-Service retrieval at 1 credit per action and autonomous task completion at 5 credits per action. Metering is not a review of the request.
MechanismDeployment Agent in Production What it actually doesNot available. Configuring it in Production for Sandbox refresh is an admin choice, not an employee approval path.

Workday has not, on these pages, described an out-of-band confirmation card inside Self-Service Agent. Do not tell employees to look for one. If your design needs a person to see the current value and the proposed value outside the chat, that is a different product decision. It is not a checkbox on Configure Agent.

A short list you can put in an operating guide

People still have to:

  1. Approve the SKU, the agreement, and which skills are on.
  2. Approve which security groups are on Available To, and activate domain changes.
  3. Approve time, job changes, and payroll inputs through the business processes those skills call, when those processes have approvers.
  4. Deactivate or narrow tools when the skill should stop.
  5. Read View AI Agent User Audit Trail when you need an audit, which is an admin report, not a worker vote.

People do not have to re-approve every retrieved balance if you enabled a base skill whose job is retrieval. That is the point of the base skills. Say so, or staff will invent a manual check the product is not waiting for.

Frequently asked questions

Does the employee approve their own time-off request twice?

The skills list says the advanced workforce skill can submit time-off requests. Any extra approval is the time-off business process you configured, not a second AI prompt. This article cannot see your process. Read the process definition.

Can a manager approve time in the agent?

MSS - Workforce & Frontline Management - Advanced includes approving, sending back, or denying submitted time. That manager still has to be on Available To and on the security that allows the action.

If we disable a tool, is that the approval?

It is a block. The agent explains it in its own words from your disabled message. It is not a record of a person approving a transaction.

What about custom agents?

Workday’s custom-agent documentation says those agents are delegate-mode only and stay hidden until ASOR activation. A person invokes them. Ambient unattended custom agents are not what that page describes. The human steps in the table above are about Workday-built Self-Service Agent and the admin tasks that surround it.

Sources checked

Workday pages in the frontmatter, fetched 2026-11-16.

Your Workday work, done by AI.

Kiweely Vine is the AI agent that builds, deploys and tests Extend apps as real people, makes changes in any Workday task screen, answers from your real tenant and builds reports.

Get the next one in your inbox.

One email a month for Workday teams: what’s changing in Workday and what to retest, one tip tried in a real tenant, and what’s new at Kiweely.

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