Vijay Raina is a preeminent voice in SaaS strategy and enterprise architecture, known for bridging the gap between high-level technological vision and the gritty realities of corporate deployment. In this discussion, we explore the shift toward the “autonomous enterprise,” focusing on the friction between probabilistic AI and deterministic accounting, the shifting economics of software consumption, and the critical role of governance as organizations move away from ad-hoc assistance toward independent AI agents.
Since large-scale internal deployments at SAP have shown productivity gains of 20% in administration and logistics, how should a typical enterprise gauge the timeline for achieving similar results when transitioning to an autonomous suite?
Achieving that 20% productivity mark is not an overnight feat; it requires a deep alignment between the digital core and the AI interaction layer. SAP utilized its own 110,000 employees as a massive testing ground to prove that Joule Work could handle high-volume administrative tasks, but a typical enterprise must first reconcile their data quality. You have to look at the “time to value” through the lens of process maturity—if your data is siloed, those gains will remain out of reach. In my experience, companies that see these results quickly are the ones that have already moved to a standardized environment across their global operations, much like we see with Nestlé operating across 140 countries. They aren’t just plugging in an AI; they are re-engineering the workflow so that the agent can actually take the wheel without constant human course correction.
The shift from ad-hoc assistance to autonomous AI agents represents a fundamental change in responsibility. How can organizations manage the inherent tension between probabilistic AI models and the rigid, audit-proof requirements of a deterministic general ledger?
This is the central paradox of modern enterprise AI—you are trying to run a business on “maybe” when the auditors demand “exactly.” To bridge this gap, the architecture relies heavily on the SAP Knowledge Graph, which maps over seven million data fields and complex relationship networks to provide the necessary business context. By anchoring the probabilistic nature of Large Language Models to this deterministic data foundation, the system can suppress the “hallucinations” that would otherwise be catastrophic in a financial setting. It is a sensory experience for the user; they transition from typing complex codes to simply asking a question in natural language, but behind the scenes, the Knowledge Graph is doing the heavy lifting to ensure the answer matches the reality of the S/4 universe. We are moving toward a “Human in the Loop” model where the AI proposes and the human validates, ensuring that the ledger remains a source of truth while the agent handles the heavy lifting of data synthesis.
With the introduction of Joule Studio, business users can now choose from over 70 AI models and design complex workflows. What are the strategic implications of giving non-technical users this much power over multi-agent orchestration?
The democratization of AI through low-code and pro-code environments is a double-edged sword that requires a very sophisticated hand at the governance wheel. By offering 70 different language and AI models, SAP is essentially giving business users a massive toolbox, but without a visual orchestration platform like the one provided by partner n8n, that toolbox would likely lead to chaos. We are seeing a shift where a procurement lead at a company like Novartis can now pilot a Sourcing Assistant to handle labor-intensive manual tasks without waiting for a six-month IT development cycle. However, the emotional weight of this shift is significant; there is a palpable anxiety among IT departments about “shadow AI” and uncontrolled agent installations. This is why the Agent2Agent (A2A) protocol and the Model Context Protocol (MCP) are so critical—they act as the digital guardrails that allow these custom agents to talk to each other and third-party systems without breaking the enterprise architecture.
The financial shift from flat-rate subscriptions to usage-based billing with AI Units and token prices seems to be causing some friction. How should CFOs approach budgeting for these unpredictable costs, especially when credits expire at the end of the year?
This is where the marketing narrative of the “autonomous enterprise” hits the cold, hard wall of fiscal planning. CFOs are currently facing a reality where they must purchase credit points in advance—essentially prepaying for a future efficiency that might not materialize if productive use cases aren’t launched in time. It creates a “use it or lose it” pressure that can lead to rushed, low-value implementations just to avoid wasting the budget. I often see companies tying up significant portions of their IT spend in these AI Units, only to find that their actual daily consumption falls far short of the hype. To navigate this, leaders need to move away from the traditional software subscription mindset and start treating AI capacity like a utility, carefully monitoring usage through tools like SAP LeanIX to ensure they aren’t over-provisioning for a “ghost” workforce of agents.
Current data suggests a divide between corporate AI announcements and operational reality, with 77% of users still leaning toward non-SAP solutions like OpenAI or Microsoft. Why is there such hesitation to fully commit to the native SAP AI ecosystem?
The hesitation is rooted in a perceived breach of investment guarantees, particularly for the on-premises user base. Many companies feel they are being pushed into “Rise with SAP” cloud subscriptions as the only way to access these new AI innovations, which creates a sense of vendor lock-in that many are desperate to avoid. There is also the “Business Data Complexity” issue, where confusing license clauses and restrictions on OData transfers make it difficult to move data freely. When the SAP API Policy (Version 4.2026) acts as a digital customs barrier, regulating bulk data extraction and forcing workloads through the fee-based BTP infrastructure, customers naturally look for more open alternatives. They want the power of the SAP data, but they want the flexibility of the hyperscalers they are already using for other parts of their business.
Governance is a major theme, particularly with the EU AI Act now in full force. What specific steps must a company take to ensure they aren’t legally liable for an autonomous agent’s erroneous procurement or payment decisions?
Under the EU AI Act, the user company is ultimately and without exception liable for the decisions made by its AI, which means “my agent made a mistake” is no longer a valid legal defense. To mitigate this, you must adhere to a “Clean Core” strategy, ensuring that your core ERP system isn’t cluttered with customizations that could confuse an AI model. You need a centralized directory of every active agent, which is where the AI Agent Hub within LeanIX becomes indispensable for tracking who—or what—is doing what. It is about creating a governance structure certified to international standards like ISO/IEC 42001, combined with tools like Signavio to analyze process bottlenecks. If an agent at a firm like PwC or Morgan Foods makes a payment recommendation, there must be a clear, auditable trail of why that decision was made and a “Human in the Loop” to sign off on high-value transactions.
SAP Pay is a relatively new entry into the market, aiming to automate invoice processing and reconciliation. How does embedding payments directly into the ERP workflow change the day-to-day life of a finance team?
Embedding SAP Pay, powered by Tereina, directly into the workflow is about removing the friction of manual file exports and external processing dependencies that have plagued finance departments for decades. Imagine a finance team at a major manufacturer that used to spend hours reconciling bank statements against open invoices; now, the system handles the payment and the reconciliation in one seamless motion when the invoice is due. It changes the role of the accountant from a data entry clerk to a financial strategist who manages exceptions rather than the rule. By reducing manual file handling, you not only speed up the cash cycle but also significantly reduce the risk of human error or fraud. It provides a level of control and transparency that was previously impossible when you had to jump between the ERP and a separate banking portal.
What is your forecast for the evolution of the autonomous enterprise as we move toward 2030?
Moving forward from 2026, I expect we will see the total disappearance of the traditional “user interface” as we know it, replaced entirely by ambient, natural language interactions that happen across mobile and desktop seamlessly. We are entering an era where the “Autonomous Suite” will not just suggest actions but will proactively manage entire supply chains, adjusting procurement orders in real-time based on global logistics data without needing a human to click “approve” for every minor deviation. However, this will also lead to a massive consolidation of the AI agent market; we will move away from having dozens of specialized bots to a few highly integrated, “super-agents” that understand the full context of a business from sourcing to payment. The companies that thrive will be those that prioritize digital sovereignty and open standards, using the Model Context Protocol to ensure they can swap out AI models as the technology evolves without being trapped by a single vendor’s pricing whims.
