The primary challenge in enterprise AI is not inventing new models but engineering them to perform reliably and correctly at massive scale.
The vast majority of enterprise AI agent tasks will be accomplished via API-based tool calling, with UI automation serving as a necessary but secondary fallback for legacy systems.
A 'right tool for the right job' approach is necessary for AI; LLMs are not suited for predictive tasks on structured data, where classical ML or specialized transformers remain superior.
AI is a catalyst for new paradigms in the software industry, forcing a revival of Test-Driven Development, a shift away from static user interfaces, and a transition to consumption-based business models.
A flexible, partnership-driven architecture is the wisest strategy for a large enterprise platform like SAP, avoiding over-investment in specific AI technologies that are likely to become commoditized.
1972
Herzig notes that SAP was founded by former IBM employees to create standardized finance software, establishing the company's foundation in core enterprise processes.
Pre-AI Era
As described by Herzig, SAP grew into the market leader for enterprise applications, serving 400,000 customers and building a massive platform with 20,000 APIs, which now presents a significant scaling challenge for AI.
Recent Past
Herzig highlights SAP's development of specialized AI, such as publishing research at NeurIPS on its RPT1 model, a transformer designed specifically for predictive tasks on structured, tabular data.
Present
Herzig reports that SAP is in a phase of broad AI implementation, deploying agentic coding for all its developers, launching AI agents in products like Concur, and seeing rapid growth in tools like 'Joule for Consulting'.
Near Future
Herzig's focus is on the next evolution of enterprise software, which includes developing 'agent mining' to improve AI, moving towards dynamically generated UIs, and managing the company-wide transition to a consumption-based business model driven by AI.
▶Pragmatic Enterprise AI ImplementationApr 2026
Herzig's discourse centers on a practical, engineering-led approach to AI. He emphasizes that the hardest problem is not building models, but making them perform reliably at enterprise scale (20,000 APIs), advocating for a flexible, partnership-based architecture to avoid betting on commoditizing tech and stressing the importance of security.
This suggests SAP's strategy is to be a stable, trusted AI integrator for its risk-averse enterprise clients, prioritizing reliability and security over being on the bleeding edge of model development, which could be a strong selling point for large corporations.
▶The Shift to Agent-Based Enterprise SoftwareApr 2026
Herzig envisions a future where AI agents are the primary interface for enterprise software, automating complex workflows. This is supported by SAP's internal use of 'agentic coding' for all developers, the launch of agents in products like Concur, and the development of concepts like 'agent mining' to improve performance.
Investors should monitor SAP's ability to effectively scale these agentic systems across its vast API landscape, as success would create a powerful competitive moat by deeply embedding AI into customers' core business processes.
▶AI as a Catalyst for Business Model TransformationApr 2026
According to Herzig, AI is fundamentally altering SAP's business model, driving a strategic shift away from traditional seat-based licensing towards consumption-based pricing. He acknowledges this is a gradual transition, with a hybrid model currently in place to accommodate enterprise customers who require budget predictability.
This transition to a consumption model introduces both opportunity and risk; while it can capture more value from high-usage clients, it also creates potential revenue volatility and requires SAP to continuously prove its value to prevent customers from optimizing away their usage.
▶Specialized AI for Structured Enterprise DataApr 2026
Herzig argues that general-purpose LLMs are not a panacea for all enterprise needs, particularly for predictive tasks on structured data. He highlights SAP's investment in specialized models like the RPT1 transformer and foundational layers like the SAP Knowledge Graph to bridge the gap between natural language and the company's core strength in structured business data.
SAP's focus on building specialized AI for its core domain could be a key differentiator against more generalized AI platforms, potentially allowing it to deliver more accurate and reliable outcomes for high-value business functions like demand forecasting and financial planning.