Share with your CIO
Enterprises across India and the Asia-Pacific region are hitting a predictable wall: AI agents that performed well in controlled pilots are struggling to deliver consistent outcomes inside real business environments. Writing for Hindustan Times, Sanjay Rohatgi of Automation Anywhere argues that the gap isn’t a model problem, it’s an orchestration and governance problem. He cites McKinsey data showing that organizations generating the most AI value are nearly three times more likely to redesign workflows alongside deployment, not after it. KPMG’s progression, from one automated recruitment process to 150-plus automations before introducing agents, is the model he holds up.
What this means for your business
Your AI agent strategy is only as durable as the process infrastructure underneath it. Organizations that dropped agents into existing workflows without first cleaning up orchestration, the coordination layer that routes tasks across HR, IT, finance, and security systems, are the ones now reporting inconsistent outcomes and stalled scale. If your enterprise still treats automation and AI as separate budget lines owned by separate teams, you’ve already created the conditions for the failure mode this article is describing.
The KPMG example deserves more attention than it gets here. The sequence matters: structured automation at scale first, agents second, built on governance that already existed. That’s not a conservative approach, it’s the only approach that has actually worked at enterprise scale. The recurring failure mode looks like the reverse: an enthusiastic agent deployment that performs beautifully in demos but degrades the moment it hits a legacy approval chain, an undocumented policy exception, or a system that wasn’t in the pilot’s scope. The agent didn’t fail. The foundation wasn’t there.
Rohatgi is SVP at Automation Anywhere, a vendor that sells exactly the orchestration and automation platform he’s describing as the prerequisite for agent success, so the argument tilts toward “buy the foundation first” in a way that conveniently extends his company’s sales cycle. The core claim still holds, but the practical implication worth stress-testing is whether an organization with strong existing workflow tooling actually needs to rebuild on a unified platform, or whether connecting agents to what’s already in place is sufficient. That’s the renewal and architecture decision your next vendor conversation should force to the surface.
Concept deep-dive: Orchestration
In the context of enterprise AI, orchestration is the coordination layer that sequences tasks across multiple systems, people, and applications to complete a business process end to end. Think of it as the traffic controller that tells an AI agent which system to touch next, under what policy conditions, and when to hand off to a human. Without it, agents complete individual steps correctly but the overall process still breaks down at handoffs, which is exactly where enterprise work is most fragile.
Based on reporting from Real-world deployments are redefining the use of AI agents, originally published 2026-08-29 06:03:00.
