Academy · AI for SAP Testing & Quality · 300 Professional
Testing Joule Agents and AI Features in SAP: Evaluation Sets, Non-Determinism and Sign-Off
How to test and sign off Joule agents and AI features when the same input can produce different outputs: evaluation sets with must and must-not behaviors, scoring with deterministic checks, LLM judges and people, thresholds and human checkpoints agreed with the process owner, OWASP-based safety tests, regression when models and platforms change, and the evidence a process owner needs before turning an agent on.
- 69 minutes
- 6 lessons + quiz
- Test Lead / Test Manager, Tester / Test Analyst, Process Owner, Key User
What you’ll be able to do
- Explain why AI features need evaluation rather than exact-match test cases
- Build an evaluation set with expected behaviors and scoring rules
- Define acceptance thresholds and human checkpoints with the process owner
- Test for unsafe actions, data leakage and prompt injection
- Plan regression when prompts, models or SAP releases change
- Assemble go/no-go evidence for an AI feature or agent
Sources and further reading
- Testing in Joule Studio · SAP Help Portal
- Testing with an Agent Card · SAP Help Portal
- Best Practices for Creating Joule Agents (Joule Studio, classic edition) · SAP Help Portal
- Add a Tool (Joule Studio, classic edition) · SAP Help Portal
- System-Defined Evaluation Metrics · SAP Help Portal (SAP AI Core)
- Benchmark Evaluation Report · SAP Help Portal (Joule)
- New Joule Studio for enterprise-scale agentic development · SAP News Center
- OWASP Top 10 for LLM Applications 2025 · OWASP GenAI Security Project
- Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile (NIST AI 600-1) · NIST
- Certified Tester AI Testing (CT-AI) v2.0 · ISTQB