SAP Exec Irfan Khan: AI Won't Simply Replace Enterprise Software

SAP's President and Chief Product Officer of Data & Analytics, Irfan Khan, said in an interview with China's Yicai that AI's impact on enterprise software should not be simplified as replacement. He argued that AI companies are largely building runtime layers, which are one-time environments that cannot fully mirror the years of order, procurement, sales, supply chain and financial records stored in enterprise software. Replicating that data history would require huge investment and remain very difficult.

Khan noted that large language models are essentially pre-trained models. When a company generates a new order, sale or purchase, that information does not automatically enter the training data. To achieve business success with AI, he said, enterprises need a data foundation that connects business context, using knowledge graphs to attach business semantics to cross-system data. Otherwise, LLMs may take context out of proportion and cause hallucinations when handling complex business problems.

On intelligent agents, Khan said software may no longer be just an interface operated by humans; agents could click the buttons that people used to click. Platforms need to define clear boundaries for what agents can and cannot do, such as not violating commercial rules like supplier discount floors, to avoid compliance and business risks. However, he added that so far, no enterprise has truly transformed its actual operations because of agentic AI. That change may come, but it will take time.

In February 2026, U.S. large-cap software stocks saw sharp volatility, with the S&P 500 Software & Services Index falling about 17% in six trading days, as concerns about AI replacing traditional SaaS intensified.