Companies have spent more on artificial intelligence over the last two years than on any technology in a generation. Most have almost nothing on their income statement to show for it. Roughly four in five organizations report that AI has delivered no measurable business impact. Around 85% are overrunning their AI budgets, undone by token consumption that no one forecasted. 

This is why I predict that a majority of AI projects will be abandoned before they reach production, defeated by data that was never made ready, workflows that never changed, and models that never integrated into a functioning system. 

Set that against the technology itself. A more capable frontier model arrives every few weeks, clearing benchmarks that once seemed insurmountable. We estimate that 93% of U.S. jobs are already exposed to AI-led change, a level we had forecast for 2032. 

The raw capability available to any company has multiplied many times over within a single budget cycle, yet the share of organizations converting it into value has barely moved. This gap, what I call “the AI velocity gap,” between how fast models advance and how slowly skills, organizational design, and governance adapt, is widening, not closing. 

Closing this gap is a trillion-dollar growth opportunity not just for companies in the IT services industry but for the entire economy: closing the gap between what AI can do and what it delivers.

The opportunity runs on two engines. The first engine runs on work companies already pay for, now done in entirely new ways, such as modernizing systems, writing software, and running operations with AI and agents alongside our people. Call it the old work, reinvented. Yes, it is deflationary by nature. AI compresses the hours and cost of work our industry has billed for decades, but that same compression is what lets us deliver more value, faster than ever before. At Cognizant, nearly 40% of our code is already written with machine assistance, freeing our teams to take on the higher-order work that clients value most. 

If that were the whole story, the skeptics would be right to wonder if AI is nothing more than a cost-cutting tool. But efficiency at enterprise scale is not achieved by simply consuming more AI, a vanity metric some now refer to as "tokenmaxxing." It is achieved by engineering the entire system around the model: the data, the context, the guardrails, and the orchestration that transform a capable model into a reliable, enterprise-grade solution. This will further unlock both more software consumption and new value pools for the industry.

This is not merely about deploying AI; it is about building with it, creating compounding systems that amplify the potential of both human and digital labor. Done properly, this foundational work does more than deliver efficiencies; it builds the capability that powers the second engine. 

This second engine, in my view, is where the real value lies: creating entirely new products, services, and business models that AI makes possible. Together, these two engines will redefine how organizations realize value in the AI era. The second engine is where the opportunity dwarfs the first engine's deflation. It is the work that did not exist three years ago: building and running agentic systems fit for production, encoding the context an enterprise has built over decades into models so they behave reliably, governing AI for trust and auditability, underwriting capital to fund the development of these high-velocity systems, and extending intelligence into the physical world. There is entirely new demand for this kind of work, and it is enormous. The second engine doesn’t offset the first; the second engine overshadows it.

I believe this opportunity can’t be seized by software alone. Agentic AI is not a product that an enterprise can buy and configure. It is a system built around the particulars of each enterprise: its reimagined workflows, its institutional knowledge, its data, its risk tolerance, its regulatory reality. The harder the problem, the more this is true. Consider what it takes to let AI agents into the prior authorization process of a healthcare payer serving hundreds of millions of members, where a wrong answer carries clinical and legal consequences. Or to map every downstream dependency buried in a telecom's legacy systems before a single change is made. or to run loan origination for a major bank across thousands of underwriters and compliance checks. None of this is a matter of plugging in a model. It is integration, orchestration, change management, and above all, the willingness to stand behind the result.

To accomplish all of this, we must make several shifts: from systems integrator to AI builder, from the talent pyramid to interdisciplinary skill, and from selling labor to underwriting outcomes. 

And we must reskill workers to meet the moment. New roles like frontier certified engineers and frontier business operators are emerging to create a new generation of talent at the forefront of enterprise AI transformation. The emergence of these roles reflects the broader flattening and widening of organizational structures. The traditional pyramid of work, with its heavily layered hierarchies, is giving way to a more interdisciplinary and agentic delivery model. 

This is why the argument that entry-level work is doomed is backward. AI can create more jobs, not fewer. By lowering the barrier to expertise, it has the potential to shorten the path to mastery and widen the base of people who can do valuable work. This is not automation erasing labor. It is AI widening who gets to contribute.

The firms turning AI into outcomes are already pulling away from those that can only deploy it, and that distinction, not the sector they sit in, is what will decide the next decade. The gap between what AI can do and the production value it delivers is a bridge, and it has to be built before the next wave of AI investment can pay off. The companies that commit to building it, engineering the data, the context, and the systems that turn capability into realized value, are the ones who will define what comes next.