Custom agents are now generally available in Workday Extend AI, and Workday reports that more than 125 early access customers and partners built over 6,000 custom apps and agents before general availability. The examples Workday has shared include a reorganisation agent at a financial services customer, a benefits experience for field workers at GE Vernova and an internal mobility agent.

Those are ambitious. Most organisations should start smaller. A good first agent removes a repetitive step, works within one persona's security, and has few actions that change data. Here are five that fit.

Each is rated on three things:

  • Risk: what happens if the agent gets it wrong.
  • Credit profile: whether its work is mostly retrieval (cheaper) or task completion (more expensive). On Workday's published example, the Self-Service Agent uses 1 credit to retrieve information and 5 to complete a task.
  • Tenant dependencies: the configuration that must be right before the agent can be.

1. Manager approvals briefing

The job: When a manager opens their inbox, the agent summarises pending time-off, expense and job-change approvals with the context needed to decide: balances, team coverage, policy limits. The manager still approves in Workday.

  • Risk: Low. The agent reads and summarises; it does not act.
  • Credit profile: Retrieval only.
  • Tenant dependencies: Clean business process routing. If approvals stall in the wrong inbox today, the agent will faithfully summarise the wrong inbox. See why business processes stall.

2. Benefits plan question answerer

The job: Answers employees' questions about their own enrolled plans, such as coverage, dependants and deadlines, and links to the right Workday task for any change.

  • Risk: Low to medium. A wrong answer about coverage can mislead. Limit it to the worker's own data and the plan documents you provide.
  • Credit profile: Retrieval, high volume around open enrolment.
  • Tenant dependencies: Accurate benefit plan configuration and the plan documents themselves. Plan the open enrolment peak in your Flex Credits budget.

3. New-hire readiness checker

The job: For each upcoming hire, a recruiter or HR partner asks the agent what is still missing before the start date: incomplete onboarding steps, missing documents, unassigned equipment requests. It reports; people fix.

  • Risk: Low. Read-only.
  • Credit profile: Retrieval, predictable volume tied to hiring.
  • Tenant dependencies: Onboarding business processes that are actually used, and security that lets recruiters see only their own hires.

4. Supplier invoice exception explainer

The job: For an accounts payable specialist, explains why a supplier invoice is on hold: a match exception, a missing receipt, a supplier on hold. It drafts the follow-up but does not release the invoice.

  • Risk: Medium. Financial data, but no action is taken.
  • Credit profile: Mostly retrieval.
  • Tenant dependencies: Consistent matching rules and supplier data. The agent will explain exceptions only as well as the configuration records them.

5. Address and personal-data change assistant

The job: Helps a worker change their own address, emergency contact or similar personal data by conversation, then submits the change through the normal business process after the worker confirms.

  • Risk: Medium. This one changes data, so it needs a clear confirmation before submission.
  • Credit profile: Task completion; the most expensive of the five per request.
  • Tenant dependencies: The relevant business processes and their approvals, and the knowledge that agent submissions follow the same routing as manual ones.

This is the right second or third agent for most organisations: build it after you have seen how a read-only agent performs and is reviewed.

What to leave for later

Agents that change compensation, run payroll-affecting processes, act across many workers at once or create financial transactions are where the largest gains are, and where the cost of a mistake is highest. Build them once your organisation has an agent review routine, security that has been tested by persona and real consumption numbers from simpler agents.

Where kiweely fits

Each of these agents depends on tenant configuration being right first: routing, security, data models. kiweely is a desktop co-pilot that connects to a client's non-production tenants with consent, reads how they are actually configured, and helps a consultant fix what an agent will rely on. That covers business process routing, security groups and the Workday Extend apps and business objects that hold the data, all tested in the browser before anything reaches Production. Start with the plan in building your first custom agent.