The discussion centers on a paradigm shift from static, human-operated user interfaces to proactive AI agents that automate complex business processes. This involves not just building models, but teaching them to use tools (APIs) reliably and verifiably at massive scale.
AI is not just a technological shift but a commercial one, forcing a transition from predictable seat-based licenses to value-aligned consumption-based pricing. The speaker notes this is a journey, with hybrid models serving as a bridge to manage customer concerns about cost predictability.
The speaker observes a growing disconnect between the rapid pace of AI innovation and its tangible adoption and impact within enterprises. SAP's strategy is to focus on delivering concrete business outcomes, like cost reduction and faster project delivery, to close this gap.
The primary engineering challenge is not model creation, but ensuring agents perform tasks correctly and reliably. This has led to a renewed focus on methodologies like Test-Driven Development (TDD) and new concepts like "agent mining" to capture decision context and improve verifiability.
A key assertion is that general-purpose LLMs are insufficient for core business predictive tasks like demand forecasting. SAP is developing specialized models (e.g., RPT1) designed for structured, tabular data to achieve the necessary accuracy and reliability for these critical functions.
Keep pulling the thread on Philipp Herzig.