From Agent Experiments to Enterprise AI Architecture
Enterprise AI architecture refers to the coordinated design of models, data, infrastructure, and governance that allows AI systems, including agents, to work safely and at scale across business processes. In many organizations, the first wave of AI has centered on conversational assistants and task-specific agents that automate narrow workflows. That phase is now giving way to a broader architectural rethink. SAP’s Business AI strategy, for example, highlights an Autonomous Enterprise vision in which agents are embedded into core applications but supported by a shared platform rather than deployed as isolated pilots. This shift matters because agents alone do not create sustainable value; they must sit on top of model choices, data foundations, and governance rules that can endure multiple AI generations. As Yaad Oren of SAP Labs US notes, AI moves in phases, and the next disruption is already forming in research.
Post-Transformer Models: Preparing for AI’s Next Phase
Post-transformer models describe the emerging architectures that AI researchers expect to succeed or significantly extend today’s transformer-based systems. After transformers were introduced in the 2017 “Attention Is All You Need” paper, they enabled the generative AI wave that enterprises are now productizing. Oren explains that new architectures are already visible at AI research conferences, even if they are not yet actionable for customers. SAP is collaborating with universities such as Stanford and the Technical University of Munich to study these post-transformer architectures and how they could reshape enterprise software over the next five to ten years. For AI infrastructure planning, this means enterprises should avoid hard-coding solutions to a single model type and instead design for model diversity, swapping, and orchestration. The priority is building platforms that can bring new model families into production without rebuilding every agent or workflow from scratch.
AI Governance Frameworks: The New Control Plane for Agents
As agents spread across finance, supply chain, and HR, AI governance frameworks are turning into a control plane for the entire enterprise AI landscape. SAP’s Business AI Platform illustrates this direction by combining agent creation, context, reasoning, and governance as one stack. According to ERP Today’s interview with Yaad Oren, governance “does not always get enough attention because it’s not as visible as the application experience,” yet it is essential when anyone in any department can create an agent. Tools such as Agent Hub from LeanIX give organizations a registry of all agents, including non-SAP agents, while Signavio supports “agent mining” to trace behavior and understand outcomes. In practical terms, governance now includes agent cataloging, authentication, exception handling, monitoring of generated data, and auditable reporting. Enterprises will need a consistent governance “blanket” that covers both current agents and future post-transformer systems.
Agentic Data Platforms: Building for Synthetic and Agent-Generated Data
Agentic data platforms are data environments designed for a world where agents both consume and create data at scale. Oren describes the “future of data” as one of SAP’s six strategic research areas, noting that customers will need more foundational services than current industry platforms provide. That includes synthetic data generation to train agents where real data is sparse, better data quality tooling, richer metadata intelligence, and new ways to understand data generated by agents themselves. In an agentic environment, lineage, semantics, and policy must travel with the data so that multiple agents can coordinate without breaking compliance or producing conflicting outcomes. This has direct implications for AI infrastructure planning: enterprises should treat data platforms as first-class AI components, not separate analytics utilities, and ensure they can support multiagent orchestration, real-time context injection, and long-term storage of agent traces for analysis and debugging.
Beyond Software: Robotics, Quantum, and Cloud Architecture
The next wave of enterprise AI will not be limited to software agents. SAP Labs US is tracking six key areas: future AI, data, user experience, robotics and physical AI, quantum computing, and cloud architecture. Robotics and physical AI will require enterprise systems that can connect physical tasks to digital records so robots can execute work, report what they did, and remain auditable. Quantum computing is expected to play a role in large-scale optimization problems in domains such as supply chain and logistics, and SAP is already collaborating with partners like IBM. Meanwhile, cloud architecture itself is evolving as multiagent orchestration and low-latency requirements reshape how SaaS applications are built. For organizations, the message is clear: start modernizing infrastructure and governance models now so today’s agents can evolve smoothly into tomorrow’s post-transformer, robotic, and quantum-augmented ecosystems.
