Agentic AI has quietly reached a strange inflection point: roughly 80% of Fortune 500 companies are running pilots, yet almost none have moved to enterprise-wide deployment. That’s the gap Arun Chandra, chief operating officer at NiCE, spends his days thinking about. The problem isn’t the technology, he argues—it’s that organizations are experimenting without a clear business objective. “Everybody’s trying to figure out what can we do with this technology?” Chandra says. But that question, asked in isolation, leads to fragmented pilots that don’t connect to revenue, cost reduction, or any other strategic goal.
Chandra’s prescription is blunt: stop layering AI onto broken workflows. “The last thing you want to do is apply AI on an outdated or an inefficient workflow,” he says. Instead, organizations need to rethink processes from the ground up and treat agentic systems as exactly that—systems. Agents require access to data, knowledge, and context to make sound decisions, plus connections to back-end systems if they’re expected to act. Chandra puts it simply: “The efficacy of these AI agents is purely a function of the context, the knowledge, and the data they can ingest and use.” Fragmented information produces fragmented performance.
The organizational side is equally thorny. When teams build isolated agent systems that don’t talk to each other, scaling creates a new kind of silo problem—one that governance, privacy, and security frameworks weren’t designed to handle. Chandra’s answer is to hold AI agents to the same standards as human workers and think of the workforce as a blend of both. That’s a meaningful shift from the usual framing of AI as a tool. It implies performance reviews, accountability structures, and change management processes applied to software.
What’s striking here is how closely this mirrors the early cloud adoption curve. Companies spent years running isolated cloud experiments before anyone figured out how to operate at scale, and the winners weren’t the ones with the best technology—they were the ones who redesigned processes around it. Chandra’s advice to avoid “boiling the ocean” and focus on high-value use cases sounds conservative, but it’s probably the only realistic path. The 80% pilot statistic suggests the industry is still in the trough of disillusionment, and the gap between experimentation and deployment is where most AI initiatives will quietly die.
