Deployment Agent is for implementers and administrators in Implementation, Preview, Sandbox, and Customer Central connected to non-production. It is not the employee assistant. HCM prompts below follow Workday’s CRAFT method from the sample-prompts topic updated 2026-11-06: Context, Role, Action, Format, Tone. You may omit Role after the first turn, omit Format when default steps are fine, and omit Tone on ordinary technical work. Do not omit “in my tenant” when you need this tenant’s configuration. There is no badge that tells you Ask is on. In Customer Central, name the tenant.
Published means the sentence is on a Workday page fetched for this guide. Pattern means you keep CRAFT and swap in your business object. A pattern is not a second official sample.
Open Deployment Agent from global search only after the agent is registered and your group is on Deployment Agent Users and on Available To.
Published prompts you can send as written
Worker calculated field for management level
Published in the calculated-field example:
I am a Workday administrator. I need to create a calculated field on the Worker business object. The logic should be: if a worker’s ‘Management Level’ is ‘Manager’ or any level above that (like Director or VP), return the text ‘Manager’. Otherwise, return ‘Individual Contributor’. Can you provide the step-by-step configuration and the specific calculation type I should use?
Expect the example’s typical answer: Evaluate Expression, default Individual Contributor, Management Level with In Selection List, return Manager. Evaluate Expression stops at the first true condition. Prove your hierarchy before you save. Published follow-up if the labels differ:
My tenant uses ‘Lead’ instead of ‘Manager’. How should I update the condition to include ‘Lead’ and everything above it?
How a calculated field in this tenant works
Published sample:
In my tenant, how does this calculated field work: [Calculated Field]
Custom reports a group created
Published sample, Customer Central form:
In my [tenant name] tenant, show me all custom reports created by [User/Security Group].
Plain-language concept
Published sample. Swap only if you truly want a different concept. Top of Stack is the one Workday wrote:
As a Workday administrator, give me a layman’s terms explanation of the ‘Top of Stack’ concept. Provide the answer in bullet points.
Compensation grades and job requisitions
Published optimization sample. It is general practice, not an inspection, until you add “in my tenant”:
Suggest best practices for linking Compensation Grades to Job Requisitions to ensure data integrity during a high-volume hiring phase.
Patterns for HCM work the admin guide already describes
These use the same slots as the published prompts. Replace the bracketed words. Do not add tasks the reply cannot point at.
Employment status text, matching Evaluate Expression’s own example
Workday’s Evaluate Expression topic already specifies Worker, text, default Empty String, then Worker Active, On Leave, Worker Retired, and Worker is Terminated (but not Retired). Ask the agent to map that grid, not to invent statuses:
I am a Workday administrator. In my tenant, I need an Evaluate Expression calculated field on Worker. Field type Text. Default Value Empty String. Return Active, On Leave, Retired, or Terminated using the conditions Worker Active, On Leave, Worker Retired, and Worker is Terminated (but not Retired), in that order. Provide a numbered list of Create Calculated Field steps and tell me which condition must stay first because evaluation stops at the first true condition.
Condition rule that uses the field
The condition-rule topic is the behavior you want the answer to respect: And stops on the first false row, Or stops on the first true row, Initiator returns nothing on the initiation step, and a processing user who cannot see a secured field causes a skipped step.
I am a Workday administrator working on HCM. In my tenant, write the condition-rule rows so an extra approval runs only when the calculated field [field name] equals Manager. Use And and Or the way Create Condition Rule evaluates them. Number the steps. Call out if I must avoid Initiator on the initiation step.
You still create the rule from the business process step and run Rule Tester. Calculated fields cannot be eligibility rules. If the reply tells you to put the field on eligibility, discard that line.
Who created HCM reports
I am an implementer working on HCM. In my [tenant name] tenant, show me all custom reports created by [security group]. Provide the answer as a list I can open in Edit Custom Report.
Hiring-phase data integrity, inspected
Take the published compensation-grade sentence and add the tenant marker the inspection samples use:
I am an implementer working on HCM in a Phase X implementation. In my tenant, suggest the configuration checks for linking Compensation Grades to Job Requisitions during a high-volume hiring phase. Number the steps. Limit this to configuration I can open in this tenant, and say when you are giving general practice instead of an inspection.
Do not use Deployment Agent to design Self-Service skills
Self-Service Agent skills are a different product. The skills topic lists HCM base (retrieve profiles, compensation and benefits, team information) and HCM advanced (update personal data, reference letters), plus manager base and advanced skills. Guide must be enabled. Asking Deployment Agent to “enable ESS - Human Capital Management - Advanced for all employees” skips Agent Management Hub and Available To. If you need a plain-language reminder of that boundary:
As a Workday administrator, give me a layman’s terms explanation of the difference between Deployment Agent and Self-Service Agent. Provide the answer in bullet points. Do not inspect a tenant.
That Tone and Format are the published learning shape. The contrast itself is in the Workday-Built Agents descriptions: Deployment Agent is non-production implementation guidance and inspection. Self-Service Agent retrieves self-service data and takes common actions such as personal information and time off.
What every HCM reply must survive
| Check | Pass looks like |
|---|---|
| CheckBusiness object | Pass looks likeWorker, or another object you can see on Business Object Details. |
| CheckFunction | Pass looks likeEvaluate Expression when the design is ordered if/then text. First true row wins. |
| CheckLocal labels | Pass looks likeA follow-up names Lead or whatever your hierarchy uses. |
| CheckCondition rule | Pass looks likeNot eligibility. Not Initiator on initiation. And and Or short-circuit described correctly. |
| CheckTenant | Pass looks like“In my tenant” or the Customer Central tenant name was in the prompt when the answer cites your fields. |
| CheckSelf-service | Pass looks likeNo instruction to skip Guide, Available To, or security activation. |
Rate the reply with thumbs up or thumbs down after you try the steps. The rating does not create the field.
Frequently asked questions
Can I paste this whole library into one chat?
No. One Action per prompt. The CRAFT topic’s samples are single tasks. A second question belongs in a follow-up after you have reviewed the first answer.
Are the pattern prompts supported by Workday?
The method and the published sentences are. The pattern sentences are fill-ins for HCM objects those pages already use (Worker, Evaluate Expression, condition rules, custom reports, compensation grades). If a reply names a task you cannot find, do not run a lookalike task.
What if the agent describes Production employee chat?
Correct it. Deployment Agent’s reference row is non-production. Employee and manager chat skills are Self-Service Agent, configured in Agent Management Hub.
How do I inspect one supervisory organization without a published sample for that phrase?
Use the inspection sample’s shape and one object: “In my tenant, how does this calculated field work” or “show me all custom reports created by [group].” Do not ask for every HCM object in one message. Workday did not publish a sample that dumps a supervisory organization.
Sources checked
Frontmatter pages, fetched 2026-11-16. Published prompts are quoted from those pages. Pattern prompts are labeled in the text.