Vijay Raina is a seasoned veteran in the enterprise software landscape, currently specializing in software architecture and private equity strategy. With a career that spans multiple market cycles, including the volatile dot-com era, he offers a grounded perspective on why the current market anxiety surrounding artificial intelligence might be misplaced. By drawing on historical precedents like the Jevons Paradox and the evolution of the datacenter industry, he explains how technological efficiency actually expands markets rather than shrinking them. In this discussion, we explore the transition from software as a tool to software as a deliverer of knowledge work outcomes, and why the “death” of SaaS is likely the beginning of its most significant expansion yet.
The current market narrative suggests that because AI can slash software production costs by 10x, the pricing power of SaaS companies will inevitably vanish; how does the Jevons Paradox challenge this assumption about market contraction?
The market is currently reacting with a significant amount of fear, evidenced by public software stocks trading down 20% through mid-May and software trading at a discount to the S&P 500 for the first time in history. However, this pessimistic outlook ignores the historical reality of the Jevons Paradox, which teaches us that when efficiency makes a resource cheaper, we don’t use less of it—we use exponentially more. We saw this in the 1860s when more efficient coal engines actually accelerated coal consumption in Britain because they unlocked entirely new industrial demands. AI is performing the same role for software today by removing the massive production barriers that previously kept solutions expensive and scarce. Instead of shrinking the revenue pool, this 10x decrease in production costs is the catalyst that will finally allow software to address the massive latent demand that has been building for decades.
Looking back at your experience with datacenter investments in 2004, what parallels do you see between the skepticism of that era—where experts thought Moore’s Law would empty out server rooms—and the fears we see in the software market today?
In 2004, the sentiment was incredibly grim; I remember seeing dot-com companies literally pulling racks of servers out of sites, leaving empty datacenter floors that felt like ghost towns. Conventional wisdom at the time argued that because a single rack would soon deliver what 10,000 racks did in 2005, the industry was destined to shrink. I recommended acquiring a troubled business out of its second bankruptcy for $200 million, a move that was so unpopular one investment committee member told me if the business failed again, I would be going down with it. Today, that one rack does indeed deliver 20,000 times the compute power of a 2005 rack, but rather than having too much space, we can’t build new capacity fast enough. That specific investment eventually sold for $3.2 billion because the increased efficiency didn’t kill the industry; it made computing so accessible that it integrated into every facet of human life.
You’ve highlighted Financial Engines as a precursor to today’s agentic AI; how did their shift from providing investment advice to directly managing $169 billion in assets illustrate the potential for software to overcome human supply constraints?
Financial Engines was a fascinating case study because it proved that knowledge work is almost always supply-constrained and rationed. Initially, when the company only offered advice, only about 20% of employees were willing to take the time to manage their own 401(k) positions, which left the majority of people with poorly allocated portfolios. For example, at JCPenney, thousands of young employees had roughly 40% of their money in cash and another 40% in company stock, which was a recipe for financial disaster. The breakthrough happened when they introduced the “do it for me” option, delegating the decision-making to a computer system that functioned much like the agentic AI we talk about today. By the time it was acquired for $3 billion in 2018, it had proved that people aren’t just looking for tools to help them work; they are looking for software that can deliver the final outcome directly and reliably.
With the U.S. knowledge work market valued at $10 trillion compared to a relatively small $0.5 trillion software market, what specific opportunities do you see for software companies to capture a larger share of that 20x differential?
Currently, we are only spending about 5% of the total cost of knowledge workers on the software tools intended to help them. There are 100 million knowledge workers in the U.S. who require years of expensive training, high wages, and significant management overhead, which has made expert services a luxury for the wealthy. AI allows software to pivot from being a mere tool—a $0.5 trillion market—to being the provider of the work itself, which is a $10 trillion market. This means for the first time, small businesses or individuals who could never afford a dedicated strategist, lawyer, or financial adviser will be able to access those outcomes through software. The “prize” for software companies is no longer just the IT budget; it is the massive portion of the economy currently tied up in the manual rationing of human expertise.
While the economic prize of unlocking latent demand is vast, you mentioned that managing machine knowledge workers is a significant hurdle; what are the practical challenges in ensuring these systems deliver results safely and reliably?
The transition from managing human knowledge workers to managing machine knowledge workers is going to be one of the most significant industrial shifts of our time. Humans are notoriously difficult to manage because they require specific environments and move between problems they find interesting, but we have spent a century developing systems to oversee them. Software’s primary job now is to replicate that reliability and safety within an automated framework, ensuring that the “outcomes” being sold are consistent and legally sound. It is no small task to ensure that an AI acting as a lawyer or an analyst doesn’t just produce work, but produces work that a business can stand behind. The companies that will win this era are the ones that move beyond the “chat” interface and build the robust architecture required to govern these digital workers.
How do you view the rise of in-house AI solutions and VC-funded startups as a threat to established SaaS players, and can incumbent firms maintain their pricing power in this new environment?
The influx of new competitors and in-house solutions is certainly creating a wave of pressure, but it doesn’t change the fundamental demand for high-quality outcomes. While it’s true that AI enables a 10x decrease in software production costs, which lowers the barrier to entry for everyone, the value will migrate from the “code” to the “delivery of the result.” Incumbents who successfully integrate AI to deliver knowledge work outcomes directly will find that their revenue actually increases because they are solving a larger, more expensive problem for the customer. We are moving away from a world where software is priced per seat and toward a world where it is priced based on the value of the knowledge work it replaces. The strength of a software business in this new age will be measured by its ability to reliably automate the $10 trillion of human labor that is currently supply-constrained.
What is your forecast for the SaaS industry over the next decade as it integrates these AI capabilities?
I believe the software market is on the verge of an explosion that will dwarf its current size, rather than the contraction the market currently fears. As we solve the challenges of safety and reliability, software will move from the periphery of business operations into the very core of knowledge delivery, effectively ending the era where expertise had to be rationed. We will see the $0.5 trillion spent on tools today begin to merge with the $10 trillion spent on human knowledge work, creating a massive new category of value. The businesses that survive this transition will be those that stop selling “features” and start selling “results,” ultimately proving that while the old model of SaaS might be changing, the industry itself has never been more alive. Long live software.
