There's a moment in Ravi Kumar's keynote at the Cognizant AI Forum 2026 that cuts through a lot of the noise around enterprise AI. He puts it simply: a trillion dollars has already been invested in AI infrastructure. Another six to seven trillion is coming by 2030. And yet, the production value enterprises are actually getting from AI is nowhere near where the capability is.
That gap — between what AI can do and what it's actually delivering for businesses — is the problem Cognizant is positioning itself to solve. And it's the same problem that's been at the center of most conversations I've had with IT leaders, developers, and architects over the past two years.
Kumar isn't wrong about the diagnosis. He cites MIT, McKinsey, and Bain, along with comments from Uber and Microsoft, all pointing to the same challenge: token consumption has been reckless, ROI linkage has been weak, and enterprises are now asking hard questions about where the return actually is. Chapter one of enterprise AI — broad, open-ended experimentation — is over. Chapter two is about specifics, cost, and accountability.
Cognizant's answer is to reframe what a services company does in this environment. Kumar's term for it is "AI builder." The shift is from system integrator — the role Cognizant and firms like it have played for the past 25 to 30 years — to something that spans systems, people, and digital labor together. The addressable market he's describing isn't the familiar trillion-dollar IT services space. It's a $6 trillion opportunity that includes business operations, agentic workflows, and the messy integration work between old and new systems.
For developers and architects, a few things from the forum are worth paying attention to.
Context Engineering Is the New Craft
Kumar has been saying this since 2024, and the live demonstration at the forum makes it concrete. Work Fabric co-founder Rohan showed a working system — not a pilot — running across Cognizant's global sales force. The system extracts tribal knowledge from emails, contracts, conversations, and distributed team interactions, synthesizes signals across geographies, and surfaces revenue opportunities that no individual human could have found on their own. The result: $200 million in incremental pipeline, with a goal of $1 billion by year end.
The underlying idea is important. Models don't understand organizations. They process what you give them. Context engineering is the discipline of deciding what to give them — the work graphs, guardrails, tribal knowledge, and organizational nuance that makes an agent actually useful in a specific enterprise environment. That's a real engineering problem, and it's going to require people who understand both the technical and operational sides.
Token Economics Is an Architectural Problem
Kumar frames token spend as something that needs to be engineered, not just managed. Model routing, context quality, harness design, continual learning — these are all levers for making agentic systems more efficient and more predictable. For teams building or integrating AI systems, this is a cost discipline that's going to matter as much as any functional requirement.
The Agentic Sprawl Into Operations Is Real
Cognizant isn't just talking about AI-assisted software development. The bigger bet is on business operations — the $4.5 trillion of labor in global enterprises that Kumar says is exposed to agentification. The technical work here involves integrating agents with existing SaaS platforms, building what he calls "headless" interfaces to legacy systems, and designing agent development lifecycles that are fundamentally different from classical software delivery.
The forum also featured a telling example from LPL Financial, where Niteshun Bhasa, firm-wide AI leader, described how they think about AI as a multiplier of advisor judgment — not a replacement for it. The framing is useful: AI removes friction from everything that isn't judgment and relationship. That's a practical design principle, not a philosophical one.
These themes aren't new to anyone covering enterprise AI closely. Agentic systems, context engineering, human-digital labor integration, and token economics have been showing up in conversations with IT leaders, developers, and architects for the past two years. What's different now is that companies like Cognizant are moving from talking about these ideas to deploying them at scale — with real numbers attached.
Kumar's $6 trillion framing is ambitious. But the underlying observation is sound: the lines between systems and people are blurring, and the companies that figure out how to operate in that blurred space are going to be the ones that matter in the next decade. For developers and architects building in this environment, that's not an abstract question. It shows up in every design decision about where agents fit, what context they operate in, and how outcomes get measured.
That's the work. And it's just getting started.