Build an agentic AI workflow with Joule Studio and SAP Build
A pro-code AI agent for Joule, built live on screen. Now on demand, code included.
Speakers
Stefan Georgievski
Enterprise Account Executive, Semos Cloud
Martin Vasilkovski
Senior AI/ML Engineer, Semos Cloud
Marko Apostoloski
Senior AI/ML Engineer, Semos Cloud
Ilija Kiroski
Head of Partnerships, Semos Cloud
Two Semos Cloud engineers build a pro-code AI agent for SAP Joule, live and unedited, recorded as part of SAP's Build Discovery Days program. A manager asks a real question, who on her team moved most on engagement last quarter, and gets back a grounded answer with a chart, inside the Joule conversation itself. The recording walks through the full architecture behind that answer: an MCP server for scoped data access, a LangGraph agent for reasoning, A2A for connecting to Joule, and the two-artifact pattern that gets a real chart into the chat instead of a static screenshot, then opens the actual GitHub repository so you can rebuild it yourself.
Key Takeaways
- Pro-code wins once it matters. Low-code is fine for a demo, but the moment an agent touches data a manager actually relies on, grounding and guardrails need to live in real code, not a template.
- Access, reasoning, and rendering are three separate problems. The AI part turns out to be the easy one, getting real data in safely and a real chart out is where the actual engineering effort goes.
- A Joule Skill template can only return text. Getting an actual chart into the chat required a deliberate second step, not a workaround bolted on afterward.
- Guardrails work as hard rules, not prompts. Never guess a number, never expose an internal ID, never open a write path, each one closes a failure the team actually hit.
- Running it is simpler than it sounds. One process on Cloud Foundry, no separate database, no queue, nothing exotic to operate once it's built.
Share
Free Resources
The full deck from the workshop, the use case, the architecture, and the live build, chapter by chapter.
Session slides (PDF)
The actual code from the session. Clone it, follow the setup, and rebuild the same agent against your own data.
Joule Pro-Code Analytics Agent Repository (GitHub)
Other Events
Ready to unify your people programs?
Stop running recognition, rewards, communications, and development in silos. See how Semos Cloud brings it all together in one AI-powered platform, built for enterprise, certified for compliance, and proven at scale.