The Overstated Threat of Software Replacement

A central concern among investors is that AI makes it easier to build software, and therefore existing platforms will be replaced. In practice, this risk is overstated. Enterprise software is deeply embedded in business operations, and the cost of failure from replacing core systems is far greater than the savings from rebuilding them.

Even if AI reduces development costs, organizations remain highly risk averse. The potential downside of disruption, compliance failure, or operational breakdown outweighs the theoretical upside of cheaper code. As a result, most foundational software will persist for the foreseeable future.

However, this does not mean pricing remains untouched. The perceived threat of replacement does have an impact. It subtly weakens pricing power, particularly in areas such as add-ons and premium features. Enterprises can increasingly build their own extensions or source them more cheaply, reducing vendors’ ability to command high margins in these areas.

The net effect is modest. Core pricing remains relatively stable, but the ability to extract premium growth diminishes.

AI’s Deflationary Pressure is Real, but Nuanced

AI is inherently deflationary, but not in the simplistic way many assume. Prices are not collapsing across the board. Instead, we are seeing a shift from value-based pricing toward cost-to-serve dynamics, a trend I have discussed previously.

As AI tools mature, competition increases, operating costs decline, and pricing models simplify. This creates downward pressure over time. However, vendors are also introducing higher-value capabilities, which allows them to maintain or even increase overall revenue per customer, albeit through different mechanisms.

In short, pricing is not deflating; instead, it is being restructured.

The Emergence of Agentic Native Pricing Models

The most significant disruption comes from a new category I describe as “agentic native” systems. These systems fundamentally change how software is consumed and, therefore, how it should be priced.

Traditional enterprise software relies heavily on per-seat pricing. This model breaks down in an agentic world. When autonomous systems perform tasks independently, the concept of a “seat” becomes irrelevant.

Three alternative pricing models are emerging:

Outcome-based pricing, such as per invoice or per transaction

Value-share pricing, where vendors participate in business outcomes

Usage-based pricing, tied to system consumption

In my view, usage-based pricing will dominate. Outcome-based models are difficult to scale and apply consistently. Value-share models, while attractive in theory, have proven challenging to operationalize at scale.

Usage-based pricing, by contrast, is transparent, flexible, and aligns incentives between provider and customer.

Why Agentic Systems Are Expensive Today

One of the biggest challenges with agentic AI today is cost variability. These systems rely on large context windows that tie up compute resources for extended periods. Unlike traditional systems that efficiently share processing power, agentic systems often monopolize it.

This creates three issues: first, costs are high and unpredictable. Second, latency can be significant. Third, providers face real financial risk if pricing does not adequately reflect usage.

This unpredictability is already creating resistance from CFOs, who are wary of open-ended cost structures. It is one of the key inhibitors to broader adoption.

Over time, I expect improvements at both the chip and architectural level to reduce these inefficiencies. However, in the near term, pricing must account for this variability. Again, usage-based models provide the most logical framework.

The Convergence of Software and Services

Agentic systems also blur the line between software and services. Many deployments require forward-deployed engineers who configure, manage, and optimize these systems in real time.

This creates a hybrid model where enterprises are effectively buying both software and embedded services. Pricing must reflect this dual nature.

A single pricing metric is unlikely to suffice. Instead, I expect a combination of usage-based software fees, and separately priced services or engineering support.

This represents a fundamental shift from the traditional separation between software licenses and services contracts.

AI’s Impact on Tech Services Economics

From a services perspective, AI introduces significant productivity gains across the software development lifecycle. Today, simply adding AI tools to existing teams yields around a 10 to 12 percent productivity improvement. With operating model changes, this can rise to 30 to 40 percent, and potentially reach 60 to 80 percent over the next few years.

This level of productivity inevitably compresses revenue. The same work requires fewer hours, and much of the value flows to the client.

However, this does not necessarily mean lower pricing per unit of labor. The opposite may occur. AI-enabled delivery requires higher-skilled talent, which commands higher wages. At the same time, the cost of tools is rising significantly.

We are moving from a 90 percent labor, 10 percent tools model to something closer to 50-50. As tools become a larger share of delivery, they can no longer be absorbed into labor rates.

The Rise of the Tools-Plus-Labor Model

This shift leads to a new pricing structure in services: a tools-plus-labor model. Instead of bundling everything into a single rate, providers will increasingly charge separately for human expertise, and for AI and software tools.

This approach is both more transparent and more sustainable. It reflects the true cost structure and aligns with how value is delivered.

While gainsharing and outcome-based pricing will continue to appear in niche cases, they are unlikely to become dominant. They are complex, difficult to govern, and often create misaligned incentives over time.

A Premium Future for Agentic Systems

Finally, it is important to recognize that agentic native environments are not just a cost story. They represent a step change in capability and, therefore, value.

These systems are complex, require specialized talent, and rely on expensive infrastructure. As a result, the cost to build and maintain them will be higher than traditional systems. However, the return on investment will also be higher.

Enterprises are likely to accept this premium because the productivity gains and business impact are significantly greater than what legacy systems can deliver.

The Bottom Line

AI is not eliminating software economics. It is redefining them. Pricing power is softening at the margins, not collapsing. Traditional models like per-seat licensing are giving way to usage-based approaches. The boundary between software and services is dissolving into a more integrated model.

For executives, the key takeaway is clear. The winners in this next phase will not be those who simply adopt AI, but those who understand how its economics reshape value, pricing, and competitive advantage.