Software, AI and Where the Value Moves

Software stocks have had a meaningful drawdown. Over the past three quarters, they’ve diverged from the broader tech sector, and in recent months the decline has accelerated. Software ETFs are roughly 30% below their highs, with some marquee enterprise software names down an eye-watering 40–50%.

That’s large enough to justify a closer look.

Not all of this is about AI. Some of it is cyclical. Software spending was elevated in the years following the pandemic and has since normalized, and valuations were high to begin with. But those factors don’t fully explain the magnitude of the losses. The more important shift is in how people think about software itself.

For the better part of two decades, the prevailing view was that software would expand into every industry, capturing more of the value created in those industries through subscription models and cloud delivery. As Marc Andreessen put it in 2011, software would “eat the world.” That prediction held up pretty well - until recently. 

Now the concern is that AI eats software.

The argument is straightforward. If users can interact with computers through large language model, via text or voice, and if AI agents can execute workflows on their behalf, then the role of traditional software begins to shrink. Why pay for rigid tools if a flexible interface can replicate much of what they do? The intuition is compelling, but it’s an oversimplification.

Traditional software and AI solve different problems. Most software systems are built around a comparatively simple structure: a user interface, a set of rules that process inputs into outputs, and a database where information is stored. These systems are designed to be precise, consistent, and efficient, but they are also narrow in scope.

AI, by contrast, is a general-purpose prediction system. It can handle a wide range of inputs and generate outputs dynamically. It can even generate new software. That flexibility is powerful, but it comes with real trade-offs. AI is less predictable, much less accurate, and typically much more resource-intensive.

Because of those differences AI and traditional software are better understood as complements. Tasks that are repetitive, rules-based, and require precision and auditability still favor software. Tasks that are unstructured or human-facing tend to favor AI. That distinction matters because the real question isn’t whether software disappears. It’s how its economics change.

AI lowers the cost of building software and, in some cases, shifts the interface layer away from the application itself. That has implications for pricing power. But the impact isn’t uniform.

At one end of the spectrum is “thin” software. These are simple tools that are often little more than interfaces over relatively light functionality. Thin software products are rarely mission-critical and are increasingly easy to replicate. AI agents can already perform many of these tasks, and the barrier to building similar tools has fallen sharply. Over time, much of the economic value in this category is likely to be competed away.

At the other end are so-called “systems of record and systems of execution” - software that underpins core business processes and is tied to structured, often proprietary data. These systems are deeply embedded in business operations and require a high degree of accuracy and reliability. They are unlikely to disappear. But their position may shift. In some cases, AI will take over the interface layer and sit on top, pushing the underlying software further into the background. Demand holds up, but pricing power may weaken.

Another category that looks resilient is software that acts as creative tools and collaborative workspaces. These software packages should benefit directly from AI integration; the technology enhances the product rather than replaces it.

For investors, these differences matter, but it’s not clear the market is discriminating amongst software companies very well.

Software isn’t going away. The locus of value is just shifting and the market is still working through what that means.