01 What happened
SAP ran an experiment with SAP RPT (sap-rpt-1.6) as a few-shot decision model within Joule Studio 2.0's intent-based development flow, published via the SAP Community on 22 October 2026. RPT correctly predicted asset type, S/4HANA API, and HTTP method for two query intents in separate tests, each at confidence 1.0. The model required no training pipeline—only a small number of labelled rows with structured features such as complexity, reasoning need, and operation type. The authors stress this is an early first-look result; the test intents were deliberately at opposite ends of the spectrum, and broader coverage including ambiguous and cross-domain intents is needed for a realistic accuracy picture.
02 Key details
- SAP RPT (sap-rpt-1.6) correctly predicted two test intents' asset type (Agent or n8n Flow), S/4HANA API, and HTTP method at confidence 1.0 from a small number of labelled examples, with no training pipeline.
- The model is proposed as a complement to Joule Studio 2.0's existing agent routing, tool selection, and execution planning, handling bounded design-time decisions.
- Next steps include extending to integration pattern recommendations, intent readiness checks, and testing against a wider range of ambiguous and cross-domain intents.
- The experiment is an early first look; results are limited to two deliberately distinct intents per test and require validation before production use.
03 Why it matters
Design-time decisions in IBD—asset type, API selection, method—are currently handled by LLMs or manual rules. RPT offers a structured, inspectable alternative that grounds predictions in labelled examples rather than pre-training. If extended and validated, it could make those choices more tunable and transparent for Joule Studio 2.0 developers.
04 Who it matters to
Data Specialists, SAP developers, developers and SAP implementation teams.