SAP, Salesforce, ServiceNow. Even hearing the names can feel heavy and uncomfortable. Their screens are complex, they are hard to learn and even harder to change. Yet global large companies still use these systems.

There is a reason. Eric Zhou (에릭 조우) and Seema Amble (시마 앰블), partners at Andreessen Horowitz (a16z), a prominent Silicon Valley venture capital firm, point to data as the reason.

In a recent post on the a16z website, the two said legacy systems such as SAP contain core corporate data. They said business procedures and organizational structures accumulated over time are embedded in the systems, and breaking away requires enormous cost and time. They said German supermarket chain Lidl tried to switch away from SAP but gave up after losing $500 million. They also cited an estimate that upgrading from SAP ECC to S4HANA requires 3 years, $700 million and 50 Accenture employees.

The problem is that even after a transition is completed, the frustration does not go away. Office workers switch apps an average of 1,200 times a day as they work. They waste 4 hours a week moving between apps. As many as 47 percent of office workers say they cannot find the information they need. Another estimate puts the share of large-scale system transition projects that achieve their goals at just 30 percent.

But as AI advances, changes are also emerging in the ERP experience. According to the two, AI-driven ERP changes can be seen across 3 stages: implementation, use and expansion.

At the implementation stage, the most expensive and risky parts of a system transition are the initial requirements definition and configuration work. AI turns scattered information such as meeting minutes, documents and tickets into structured requirements, and automatically generates test scenarios and migration plans.

The two said startups such as Axiamatic, Conduct and Auctor are pushing into this market. They said these companies are creating value by reducing transition risk and shortening timelines.

AI also has a role when people use systems. AI can attach to Slack or a browser sidebar to answer questions such as "Where do I find this?" and "How do I do this task?" and support direct execution through APIs. AI agents can also handle screen-based tasks that previously had no way to be automated. This is done by reading and manipulating SAP screens directly and automating repetitive work. The two said companies such as Factor Labs and Sola are moving in this direction.

At the expansion stage, more interesting changes can be seen. The two said companies constantly change through new products, new regulations and mergers and acquisitions. They said it is wasteful to modify the system or build separate apps each time. They said AI quickly overlays a light, purpose-fit experience on top of existing systems, such as building a dedicated app that handles document collection, approval and registration at once instead of navigating 12 SAP screens to register a supplier.

AI is unlikely to make legacy systems themselves disappear. The two said SAP, Salesforce and ServiceNow are too deeply embedded. They said what changes is how people work, shifting to a structure in which AI becomes the interface and systems handle tasks automatically when users state the desired outcome. They added that the data and workflows accumulated in the process become assets that competitors cannot easily catch up with. They said successful work becomes a reusable pattern, and AI riding on old software is becoming the center of a new software competition.