Workday AI Agent Testing
How Workday Tests Its Own AI Agents, and What You Still Need to Test
Introduction
Workday tests its agents hard, but it cannot test your tenant.
Workday has described how it checks its own agents, from large evaluation sets to Agent Passport, a verification record for agents. That is good news. It also raises a fair question about Workday AI agent testing: what is already covered, and what is left for you?
This post explains what Workday says it tests, where that testing stops, and a checklist for the parts only you can check. You will also get a simple split of who tests what.
Key takeaways
Workday reportedly uses an evaluation set of more than 2,000 questions for its Sana Self-Service Agent and runs it more than once a day.
Agent Passport is designed to test and verify agents against public security standards before production. It starts with early access customers in the second half of 2026.
Vendor testing covers the agent in general. It cannot cover your security roles, approval chains, data and integrations.
You own five checks: security roles, approvals, data, integrations and release changes.
Keep evidence of your own tests, because auditors ask what you verified.
Why Workday AI agent testing matters now
What Workday says it tests
According to ITWire, a Workday AI engineering leader said evaluations are central to how agents ship. For the Sana Self-Service Agent alone, the report says Workday runs an evaluation set of more than 2,000 questions, more than once a day. The tests check accuracy, truthfulness and whether the agent did what it claimed. The report adds that one customer’s own evaluation set was added to Workday’s runs.
<!-- SOURCE NEEDED: confirm ITWire details and publication date; the figures come from one trade report, not a Workday release -->
What Agent Passport adds
Workday announced Workday Agent Passport on 2 June 2026, according to Workday’s newsroom announcement. It tests and verifies Workday-built and third-party agents before production and monitors them afterwards. Results are tied to standards such as OWASP LLM Top 10, NIST AI RMF and MITRE ATLAS. Cisco is the launch partner.
<!-- SOURCE NEEDED: confirm the general availability date (projected in the release) and whether early access applies to your tenant -->
Why the next few weeks matter
Agents are moving from early access to wider use. Vendor evidence will start to arrive, and it will look reassuring. Decide now what you will accept as proof and what you will test yourself.
Example: splitting vendor evidence from your own tests
Illustrative, not a real customer.
A company receives vendor test evidence for a Workday agent and asks what is left to test. Its HRIS lead puts two columns on one page: what the vendor evidence covers, and what the company must verify in its own tenant. Under the first she lists general accuracy, attack resistance and safe behaviour. Under the second she lists security roles, approvals, data and integrations.
In the sandbox, the first test shows a manager role seeing a report meant for HR only. The cause is a custom security group the vendor never saw. The team fixes the group, reruns the case and files the result next to the vendor evidence. Auditors later ask for both, and the company has them in one place.
Who tests what
Area | Vendor evidence typically covers | You still test | Pass looks like |
|---|---|---|---|
Accuracy | General answer quality | Your policies, terms and job data | Answers match your policy |
Security | Attack resistance and data-leak checks | Workday security roles testing by persona, using your groups | Each persona sees only what it should |
Approvals | Standard routing | Workday approval workflow testing for your chains, delegation and reassignment | Correct routing and audit trail |
Integrations | Product-level behaviour | Workday integration testing for your downstream feeds and chat channels | Data arrives once and is correct |
Updates | The vendor’s own release checks | Workday release testing on your configuration after each update | Same results, or a change you understand |
How PeopleFlow helps
Vendor evidence shows an agent is sound; your own tests show it is sound here. PeopleFlow’s no-code Workday testing comes with 1000+ pre-built tests for HCM, Payroll, Finance and Integrations, which you can run against your own tenant. Run them before a pilot and after each release, and keep the results as your evidence. Our Workday AI agents include Configuration, Report, Support, Testing and Integration agents. See the Workday testing page and the AI agents page. PeopleFlow is a Workday Innovation Silver Partner.
Frequently asked questions
How does Workday test its own AI agents?
According to ITWire, Workday uses an evaluation set of more than 2,000 questions for its Sana Self-Service Agent and runs it more than once a day. The tests check accuracy, truthfulness and whether the agent did what it claimed. Workday also announced Agent Passport for independent verification before production.
What is Workday Agent Passport?
Workday says Agent Passport tests and verifies Workday-built and third-party agents before production and monitors them afterwards. Results are tied to public standards such as OWASP LLM Top 10, NIST AI RMF and MITRE ATLAS and signed by the testing partner, starting with Cisco. It starts with early access customers in the second half of 2026.
Do I still need to test Workday AI agents myself?
Yes. Vendor testing covers the agent in general, not your configuration. Your security roles, approval chains, data and integrations are unique, and so are your policies. Run your own tests in a sandbox, keep the results, and rerun them after each release so you have evidence of what you verified.
What should I test first on a Workday AI agent?
Start with security roles, because a wrong permission can do the most harm. Then test approvals, including delegation and reassignment. Next check data and integrations, and finish with release changes. Use a few named personas, write the expected result first, and save screenshots as evidence.
Conclusion
Workday AI agent testing is shared work. Workday tests the agent in general, and Agent Passport is designed to add independent verification. You test your own roles, approvals, data, integrations and releases. Ask vendors for their evidence, then add yours, and keep both on file. To build your side faster, try PeopleFlow yourself or book a short demo.
