The Self-Learning Enterprise: When AI Creates Its Own Business Model

Executive Summary

Jensen Huang's CES 2025 keynote introduced a vision that extends far beyond enhancing existing enterprise applications with AI. His concept of "AI factories" hints at a profound shift where AI agents may eventually develop their own understanding of business operations, potentially inverting the traditional relationship between software and intelligence.

For technology leaders implementing COTS solutions like SAP S/4HANA, this raises transformative questions: What happens when AI agents evolve from using pre-defined business rules to creating their own understanding of how your business works? How might this change the fundamental value proposition of enterprise applications? And most critically, are we preparing for the right future when we implement today's enterprise systems?

Based on this emerging reality, two immediate actions that forward-thinking organizations should consider:

• Establish an AI Insight Lab: Create a small, cross-functional team tasked with identifying 2-3 high-value business processes that could benefit from AI-driven insights. This lab should operate outside normal IT governance, with direct reporting to executive leadership and a 90-day cycle for demonstrable outcomes. This approach creates a contained environment for exploration while generating quick wins that build organizational confidence.

• Engage with Advanced AI Vendors for Targeted Proofs of Concept: Select 1-2 vendors with advanced business AI capabilities (such as Microsoft's Copilot for SAP or SAP's Joule) to prototype complete, end-to-end processes using their most advanced frameworks. Focus on processes that cross traditional system boundaries, such as forecast-to-fulfill or design-to-manufacture, to demonstrate the power of AI agents that develop their own business understanding rather than following pre-programmed rules.

This article examines the revolutionary implications of Huang's AI factory vision, providing business leaders with a framework for understanding what may be the most significant shift in enterprise computing since the advent of digitization.

The Implied Revolution in Huang's "AI Factory" Vision

Jensen Huang's keynote presented the concept of "AI factories" as a fundamental shift in how computing operates:

"From retrieval-based computing to generative-based computing, from the old way of doing data centers to a new way of building this infrastructure, and I call them AI factories," Huang explained. "They're AI factories because [they have] one job and one job only: generating these incredible tokens that we then reconstitute into music, into words, into videos, into research, into chemicals, or proteins."

While Huang didn't explicitly state that AI would replace traditional applications, his vision implies a computing model where intelligence emerges from AI systems rather than being programmed into applications by humans.

From Programmed Rules to Self-Developed Business Understanding

The most revolutionary aspect of Huang's vision may be what he didn't directly state but strongly implied: the potential for AI agents to develop their own understanding of how business operations work rather than relying entirely on rules programmed by humans.

In technical terms, this is referred to as a "semantic layer" – the part of software that understands the relationships between business concepts and applies meaning to raw data. Traditionally, this understanding is programmed by humans into enterprise systems, such as SAP. Huang's vision suggests AI might eventually build this business understanding on its own.

Several elements from his presentation suggest this direction:

1. His description of NVIDIA Cosmos as a system that builds its own understanding of physical reality through training on "20 million hours of video" focusing on "physical dynamic things."

2. The emphasis on AI agents that can "perceive, reason about their environment, planning their next motion, and acting" without pre-programmed responses.

3. References to reinforcement learning and "AI feedback," where systems evolve their understanding through experience rather than fixed programming.

4. His statement that "the IT department of every company is going to be the HR department of AI agents" - suggesting AI systems with emergent capabilities rather than just programmed functions.

This points toward a future where, instead of enterprise applications defining how business processes work, we might see AI systems that build and maintain their own understanding of business processes through observation and interaction.

The Gap in Current Enterprise Thinking

Most discussions about AI in enterprise contexts assume a complementary model where:

1. Traditional enterprise applications like SAP and Oracle continue to define business rules and relationships.

2. AI capabilities enhance these applications with better interfaces and analytics.

3. The fundamental architecture of enterprise computing remains largely unchanged.

Huang's vision challenges this assumption by suggesting that AI systems may eventually build their own understanding of business concepts, relationships, and processes - potentially inverting the traditional hierarchy where applications define how business works.

The Three Phases of Enterprise AI Evolution

Based on this interpretation of Huang's vision, we might envision three phases of enterprise AI evolution:

Phase 1: AI as Application Enhancement (Current State)

• AI capabilities are added to existing enterprise applications.

• Business rules and relationships remain defined in traditional COTS systems.

• AI serves primarily as an improved interface and analytics layer.

Phase 2: AI as Business Interpreter (Emerging)

• AI agents act as intermediaries between humans and systems.

• They draw on data from traditional applications but create their own working models.

• Traditional applications remain the systems of record, but AI increasingly mediates interaction.

Phase 3: AI as Business Understanding Creator (Future State)

• AI systems develop and maintain their own understanding of business operations

• Traditional applications may be reduced to transaction processors and data stores

• Business rules and relationships emerge from AI models rather than being pre-programmed in applications

Timeline and Indicators: When Might This Evolution Unfold?

While precise predictions are impossible, the pace of AI advancement suggests a rapid evolution:

• Phase 1 (2023-2024): We're already concluding this phase. Major COTS vendors like SAP (with Joule), Microsoft (with Copilot), and Salesforce (with Einstein) have integrated foundation models into their platforms.

• Phase 2 (2024-2026): Cross-system AI agents will emerge much faster than many expect. Early indicators include AI assistants that operate across multiple enterprise systems, synthesizing information without pre-programmed integrations. Companies like ServiceNow and Microsoft are already moving in this direction with cross-platform agents.

• Phase 3 (2026-2028): The emergence of AI systems developing their own business understanding could begin within 2-3 years. Look for AI systems that identify and suggest process improvements without being explicitly programmed to do so. Research from organizations like Google DeepMind and early implementations from companies like Tesla (in manufacturing) already show initial capabilities in this direction.

Key technological milestones to watch include improvements in AI reasoning capabilities, advancements in systems that combine rules with learning, and the emergence of business-focused foundation models trained specifically on business processes.

Strategic Implications for Manufacturing and Supply Chain

For vertically integrated companies with complex operations, these changes will have particularly significant implications:

Production Planning: Today, production planning typically relies on highly configured ERP modules with complex rulesets. In phase 3, AI systems might develop more sophisticated models that balance constraints beyond what pre-programmed systems can consider.

Consider a manufacturing line producing custom industrial equipment with thousands of configuration options. Where current systems require explicit rules for each scenario, advanced AI agents might develop their own understanding of manufacturing constraints, intuitively balancing factors like material availability, labor allocation, and quality considerations without requiring explicit programming.

Product Lifecycle Management: Traditional PLM systems rely on structured workflows defined by humans. The evolution toward AI-created business understanding could enable AI systems to develop sophisticated insights into product relationships and dependencies that weren't explicitly modeled.

Supply Chain Resilience: The ability of AI to develop its own understanding of supplier relationships, logistics constraints, and demand patterns could transform how organizations manage supply chain disruptions - moving from rule-based responses to an intuitive understanding of systemic impacts.

The Competitive Advantage of Early Adoption

Organizations that successfully navigate this transition early will realize significant competitive advantages:

• Rapid Process Adaptation: Early adopters of AI business understanding will be able to modify operational processes 3-5x faster than competitors. When market conditions, supply chains, or customer expectations change, these organizations can realign processes within days rather than months, creating significant operational agility and customer responsiveness.

• Labor Market Advantage: As skilled labor becomes increasingly scarce, organizations with AI agents that understand business operations can onboard new employees 40-60% faster. Complex tribal knowledge becomes accessible to new employees through AI interfaces, significantly reducing the impact of turnover and enabling faster growth in emerging markets.

• Information Arbitrage Advantage: Organizations that successfully implement AI-driven business understanding gain access to internal information arbitrage—the ability to identify and act on operational insights that remain invisible to competitors. This creates compound advantages in bidding, pricing, customer service, and product development that grow over time.

The organizations that begin this journey now, even while maintaining their existing COTS investments, will establish advantages that become increasingly difficult for competitors to overcome as AI capabilities mature.

Investment Strategy and Financial Considerations

CFOs and boards face critical questions about how to allocate resources during this transition. A balanced investment strategy should consider:

• Protecting Current Investments: Current COTS implementations aren't becoming obsolete overnight. Their role may evolve, but the data foundations they provide will remain valuable as training grounds for AI systems. Organizations should continue planned implementations but with an eye toward data quality and system accessibility.

• The Reality of Enterprise ERP Transitions: In my experience at ABB and similar manufacturing enterprises, this challenge is particularly relevant. Many global organizations continue to operate core ERPs on older platforms, such as SAP ECC 6.0, with some systems still being depreciated. According to recent Gartner analysis, approximately two-thirds of SAP deployments haven't migrated to S/4HANA, and at least 50% won't make the transition by 2030.

For technology leaders, this creates a strategic dilemma. "Waiting" for S/4HANA or the next best thing isn't viable - decisions need to be made now about how to leverage existing systems while preparing for AI transformation. The organizations that successfully navigate this balance will create significant competitive advantage.

• Incremental Investment Approach: Rather than betting heavily on either traditional COTS or revolutionary AI, consider a portfolio approach:

o    70% maintaining and enhancing core systems

o    20% developing AI capabilities that work with existing systems

o    10% experimenting with more advanced agents that may eventually develop their own business understanding

This allocation might shift to 50/30/20 by 2026-2027 as the technology matures.

• ROI Measurement Evolution: As value shifts from process efficiency to business insight capabilities, CFOs should evolve their approach to measuring technology ROI. Traditional metrics, such as cost reduction and efficiency, should measure early AI investments. However, as capabilities advance, new metrics, including decision quality, innovation acceleration, and adaptability, become increasingly relevant.

Three Strategic Responses for Technology Leaders

Given this potential future, technology leaders have three possible strategic responses:

1. The Traditional Path

• Continue implementing COTS applications with a traditional focus on process standardization.

• Add AI capabilities incrementally as mature solutions emerge

• Assume traditional application architecture will remain dominant

• Best for: Organizations in highly regulated industries or with recent major COTS investments

2. The Hybrid Strategy

• Implement COTS applications but prioritize data foundation and system accessibility

• Create parallel AI agent capabilities that draw on application data

• Prepare for a gradual shift in where business intelligence resides

• Best for: Most organizations, especially those with complex operations and moderate risk tolerance

3. The AI-First Approach

• Minimize investment in traditional process-centric applications

• Focus on data platforms and AI infrastructure

• Accept higher short-term risk for potential long-term advantage

• Best for: Digital-native organizations or those facing disruption with an appetite for higher risk

The COTS Vendor Response

For traditional enterprise software vendors, Huang's vision presents both a threat and an opportunity. The danger is clear: if business understanding eventually emerges from AI rather than being pre-defined in applications, much of their traditional value proposition erodes.

However, vendors who recognize this shift could evolve toward providing:

1. Exceptionally clean, structured data foundations that serve as optimal training environments for AI

2. Rich business context models that accelerate initial AI understanding of business concepts

3. Governance frameworks that ensure AI evolution remains aligned with business need.

4. Verification systems that validate AI-developed business models against operational requirements.

Managing Risk and Governance

As AI systems begin developing their own business understanding, organizations face new governance challenges:

• Data Accuracy and Bias: Self-learning systems can only develop accurate business understanding if they learn from high-quality, representative data. Organizations must implement governance frameworks that ensure AI systems have appropriate training data.

• Decision Transparency: When AI agents create their own understanding of business operations, explaining their decisions becomes more challenging. Implementing explainability requirements from the outset will be crucial, particularly in regulated industries.

• Change Management: As systems transition from static, human-programmed rules to emergent business understanding, organizations require governance mechanisms to ensure that changes align with business objectives and compliance requirements.

Forward-thinking organizations, such as Capital One and JP Morgan, have already established AI governance frameworks that could serve as early models; however, these will need to evolve as capabilities advance.

Conclusion: Preparing for a New Kind of Business Intelligence

Jensen Huang's "AI factory" vision may have implications far beyond adding AI capabilities to existing software. If taken to its logical conclusion, it suggests a world where AI agents not only consume but eventually create their own understanding of how business works - a fundamental inversion of traditional enterprise architecture.

For technology leaders implementing systems today, this creates a dual imperative: continue delivering immediate business value through traditional approaches while preparing for a future where the center of business intelligence may shift dramatically from applications to AI systems.

To position your organization for this transition, consider these immediate actions:

1. Conduct an AI-Readiness Assessment: Within the next 60 days, evaluate your current enterprise applications against AI-readiness criteria: data quality, API accessibility, process documentation clarity, and semantic model completeness. This baseline assessment lays the foundation for strategic planning and identifies opportunities for immediate improvement.

2. Initiate Cross-Functional AI Governance: Establish an AI governance framework that extends beyond IT to include operations, legal, finance, and business units. Unlike traditional technology governance, which focuses on standardization, this framework should emphasize experimentation guardrails, ethical boundaries, and business outcome measurement for AI initiatives.

3. Develop an AI Talent Strategy: Create a deliberate talent strategy that strikes a balance between building internal capabilities and forming strategic partnerships. The most successful organizations are developing hybrid roles that combine domain expertise with AI understanding rather than treating AI as a purely technical specialty.

4. Schedule Quarterly Board-Level AI Strategy Reviews: Elevate AI strategy to a board-level discussion with quarterly reviews of progress, market developments, and strategic adjustments. This cadence reflects the accelerated timeline of AI development, ensuring organizational alignment around this transformative capability.

By acknowledging the potential for this shift without overcommitting to unproven approaches, technology leaders can navigate the complex transition period that lies ahead—maintaining operational excellence today while preparing for the emerging AI future that Huang's vision suggests is on the horizon.

Dmitriy Gerzon is a Technology Transformation Leader with extensive experience guiding organizations through complex technology transformations. He specializes in implementing Commercial Off-the-Shelf (COTS) solutions that enable business growth and acceleration, with a particular expertise in scaling operations and accelerating business value.

#AITransformation #EnterpriseAI #FutureOfERP #AIFactories #DigitalTransformation #TechLeadership #AIGovernance #EmergingTechnology #S4HANA #SupplyChainAI #ManufacturingInnovation #AIAgents