From Better Models to Better Context
Enterprise AI governance is the discipline of controlling how AI systems use business data, execute processes, and remain accountable in production environments, placing context management, policy enforcement, and auditability ahead of raw model sophistication or isolated prototype performance. At SAP Sapphire in Orlando, SAP argued that this is where the real AI race is being run. The company’s view is blunt: large language models and coding assistants are becoming commodity layers. SAP CTO Philipp Herzig stated, “What’s not differentiating is the LLMs… Use OpenAI models, use Anthropic models, whatever you like.” What matters is whether agents know the company’s business entities, can reach the right data, and respect authorizations and compliance rules baked into ERP systems. By reframing the problem from “Which model?” to “Which context and controls?”, SAP is pushing enterprises to treat AI as part of their operational backbone, not an experimental sidecar.
Inside SAP’s Business AI Platform Strategy
SAP’s Business AI Platform pulls together Business Technology Platform, Business Data Cloud, the Autonomous Suite, and Business AI into a single context-first stack. The pitch: use partner LLMs from providers such as Anthropic or Mistral AI, but run them inside SAP’s process-aware, policy-aware environment. Context lives in familiar assets: SAP’s data models, role-based authorizations, compliance configurations, and customer-specific extensions. Over this, SAP is building Joule Assistants and more than 200 specialized agents that operate across finance, supply chain, procurement, HR, and customer experience. Joule Work aims to be the “front door” for these agents, spanning SAP and non-SAP systems. The platform also focuses on AI context management by auto-generating product requirements, technical specs, tests, security setups, and observability hooks. Enterprise AI governance here is less about picking a favorite model and more about turning years of ERP configuration into a reliable context layer for agentic AI.
The Autonomous Enterprise Meets Execution Reality
SAP’s autonomous enterprise strategy moves AI from recommendation tools to active operators in business processes. Agentic AI control, however, raises a new question: once AI knows enough to recommend actions, who guarantees that actions execute with the determinism, auditability, and control enterprises expect? This is the concern Charles Crouchman, Chief Product Officer at Redwood Software, centers on. Redwood comes from workload automation across financial close, MRP runs, billing cycles, and supply chain orchestration, where timing, sequence, and approvals are strictly defined. According to Redwood’s Crouchman, “the question has moved to, ‘Can AI actually execute inside my business?’” In production, thousands of interdependent steps stretch across ERP, cloud, and legacy systems. Probabilistic agents can introduce subtle inconsistencies that do not fail loudly but accumulate over time. That execution gap sits squarely in the domain of enterprise AI governance, demanding stronger oversight before AI agents touch systems of record.
Designing Agentic AI Control and Oversight
As SAP customers move toward multi-agent environments, the operating model for AI needs clear boundaries. SAP’s Joule agents and Redwood’s agentic orchestration work point to the same requirement: AI context management must extend beyond data into workflow logic and controls. Redwood is evolving from workload automation to an agentic orchestration platform, adding MCP server support, A2A multi-agent orchestration, and planned tools such as Agentic Studio and Agentic Workflows. The aim is to expose deterministic business logic to agents without ceding control of execution order, approvals, or exception handling. For CIOs, this means enterprise AI governance cannot be an afterthought layered over “smart” agents. Policies about who can trigger which workflows, what needs human approval, and where audit trails are stored must be codified directly into AI orchestration. The autonomous enterprise strategy only works when agentic AI control is as deliberate as the processes AI is meant to optimize.
Why Context Demands Stronger Data Foundations
Both SAP’s Business AI Platform and Redwood’s execution focus share an implicit message: enterprises need better data and process foundations before they deploy advanced AI agents. Context-driven AI depends on clean master data, consistent process definitions, and clear authorization models. Without that, agents either lack the context to act or operate in ways that are hard to audit and correct. SAP’s push to auto-generate tests, specifications, and governance scaffolding reflects how much enterprise AI success now hinges on disciplined data management and observability. It also suggests that organizations still “vibe checking” models with a few happy-path prompts are not production-ready. A credible autonomous enterprise strategy starts with mapping critical processes, clarifying ownership, and instrumenting systems for traceability. Only when those pieces are in place can a business AI platform safely turn generative models into accountable agents embedded in day-to-day operations.
