The sharpest anxiety in software is no longer whether AI is impressive. That part has been settled. Now, the focus lies on what AI does to defensibility. If features are faster to build, code is cheaper to generate, and agents can perform more of the tasks that traditional applications were designed to organise, then the visible product becomes a less reliable source of advantage.
That is why the current repricing of software companies has cut so deeply. Investors are not merely asking which products have AI features attached. They are asking which businesses still control something difficult to copy once features themselves become easier to imitate.
The answer emerging from the sector is not brand alone, nor product breadth alone. It is data, and more specifically proprietary operating context. Oracle and Salesforce have spent the week pushing back against the idea that AI will simply wipe out software-as-a-service incumbents.
In Oracle’s case, the company has argued that the AI boom will support revenue growth for several quarters, while also saying that new AI code-generation tools are allowing it to reorganise product development into smaller, more productive groups. Salesforce has made a related argument from a different angle: the value is not only in applications, but in the data engine, governance, and workflow history that sit underneath them.
This is where the moat becomes easier to see. Oracle’s position is strengthened by deep enterprise data across finance, supply chain, and human resources, as well as the ability to run its database across major clouds. Salesforce says its Data Cloud has surpassed 50 trillion records ingested or connected to via Zero Copy, and positions that underlying data layer as the basis on which Agentforce can ground AI in proprietary customer information.
The strategic point is straightforward. AI can help generate outputs. It is much harder to recreate the structured records, permissions, process logic, and historical context that determine whether those outputs are useful inside a live enterprise environment.
That does not mean every incumbent is equally protected. Some data is richer than other data, and some workflows are more deeply embedded, more regulated, or more expensive to unwind. Standardised information in common formats offers a weaker defence than messy, longitudinal, business-specific context tied to real decisions and switching costs. That distinction matters because AI is likely to erode parts of the old application moat even while strengthening the value of certain data estates underneath.
The market is beginning to sort software companies accordingly. Businesses built mainly on packaging and interface are under more pressure than businesses that sit on difficult-to-recreate information environments.
There is a buyer-side implication here as well. Procurement teams have long focused on front-end capabilities: how quickly a tool can be deployed, how well it integrates, how intuitive the interface feels, and whether the road map looks credible. Those questions still matter, but AI has made the underlying data model far more strategic. How much unique operating context does the platform actually hold? How portable is that context if a customer wants to switch later? How much governance sits above it? Which parts of the product become commodity features if generative models improve again within a year? These questions are closer to the source of future value.
Software is not disappearing into AI. Rather, it is being sorted by what sits below the surface. Products that rely mainly on feature advantage are likely to feel more heat as coding accelerates and agents become more capable. Platforms that combine workflow authority, proprietary data, governance, and room to deploy AI across real business processes have a stronger claim to endurance.
The visible interface will still matter but the stronger moat may increasingly be invisible: the data estate, the process memory, and the operating context that make automation reliable rather than merely persuasive in a demo.
You must be logged in to post a comment.