The Real Reason Why AI Agents are Failing in Organisations

For the past three years, generative AI tools have helped employees write emails, summarise documents, generate code, and improve productivity. Yet when it comes to agentic AI systems capable of autonomously executing tasks and making decisions, most organisations are still struggling to move beyond experimentation.

According to a recent Genpact-HFS Research study surveying 545 senior executives across 11 industries, 92% believe agentic AI will fundamentally change how work gets done. However, only 6% have successfully scaled it across their organisations.

Speaking to AIM, Ajay Vasal, Senior Vice President, Data & AI, Genpact, believes the problem isn’t the technology itself. “The biggest gap we see is not about deploying AI models. It’s about deploying AI into workflows that were originally designed for humans,” he says.

https://www.youtube.com/watch?v=KzLjih6-GAA

The Real Failure Point

Many organisations assume agentic AI can simply be layered onto existing processes. According to Vasal, that assumption is precisely where projects begin to fail.

“Today’s workflows are designed around humans handling ambiguity, exceptions, and decision-making. If AI is expected to orchestrate an end-to-end workflow, organisations need to fundamentally rethink and redesign those workflows,” he explains.

Instead of treating Agentic AI as a bolt-on technology, companies need to rebuild processes with AI at the centre. Those that are successfully scaling AI are not merely adding agents to existing systems; they are redesigning workflows from the ground up.

The rapid adoption of ChatGPT and similar tools gave enterprises confidence that AI could drive productivity. However, Vasal argues that agentic AI represents a fundamentally different challenge.

“Generative AI helped individuals work faster and better. It augmented people. Agentic AI is expected to execute workflows, coordinate decisions, and manage handoffs across functions. That’s a completely different responsibility,” he says.

While generative AI improved individual productivity, agentic AI requires organisations to rethink how entire business processes operate. This transition, according to him, is not a linear evolution but a complete re-architecture of enterprise workflows.

The Accountability and Regulation Problem

One of the biggest obstacles preventing organisations from scaling agentic AI is trust.

The Genpact-HFS study found that while most executives believe AI will transform work, only a small percentage are comfortable with AI systems making decisions independently.

“The concern isn’t whether AI can generate an answer. The real question is whether organisations can stand behind the decisions AI makes,” notes Vasal.

He believes enterprises need clearly defined accountability frameworks, decision rights, and escalation mechanisms before they can confidently deploy autonomous agents at scale. Without those safeguards, organisations will remain hesitant to move beyond pilot projects.

Highly regulated industries such as banking, healthcare, and insurance often face additional concerns around compliance and reputational risks. However, Vasal believes regulation is not the primary reason enterprises struggle with agentic AI adoption.

“The biggest barrier isn’t regulation. It’s process readiness,” he says.

According to him, organisations frequently underestimate the amount of process redesign required before AI can be trusted with decision-making responsibilities. As a result, many companies end up stuck in what industry observers often describe as “death by a thousand pilots”—a cycle of experimentation without meaningful production deployment.

AI ROI Isn’t Headcount Reduction

As enterprises invest heavily in agentic AI, demonstrating return on investment remains a major challenge. Many organisations continue to evaluate AI through traditional productivity metrics such as reduced headcount. Vasal argues this approach misses the larger opportunity.

“AI is not about taking people out of the system. It’s about redirecting people towards higher-value work,” he insists.

Instead of focusing solely on labour savings, organisations should evaluate AI based on business outcomes. For example, in finance operations, AI can improve invoice processing accuracy, reduce payment errors, minimise cash leakage, and increase opportunities to capture discounts.

“The real value comes from improving business performance, not simply reducing workforce costs,” he explains.    

Process Intelligence Before AI

Drawing from Genpact’s own transformation journey, Vasal believes successful AI adoption depends on what he calls ‘Process Intelligence’. The methodology provides AI and ML algorithms with a data-informed understanding of an organisation’s processes.

According to Vasal, many organisations mistakenly view AI as a technology deployment exercise. However, successful adoption depends on several organisational factors working together. 

“The success of AI adoption is less about technology and more about accountability, measurement, people and process,” he explains.

This, he says, is where many organisations struggle. While AI capabilities continue to improve rapidly, businesses often fail to prepare the underlying processes and governance structures required to support autonomous systems. Without that foundation, even the most advanced AI deployments remain confined to pilot projects rather than delivering enterprise-wide impact.

This transition from experimentation to production is what separates successful AI implementations from those that fail to scale.

The Future Workforce

Looking ahead, Vasal expects AI to fundamentally reshape workforce roles. “We believe everybody in the future will either be an AI builder or an AI practitioner,” he says.

Some employees will design, build and manage AI systems, while others will work alongside AI agents as part of their daily responsibilities.

This shift, he believes, makes continuous upskilling essential. “Upskilling today means actively using AI, experimenting with it and learning how to work alongside it. Every professional, regardless of their role, will need to become comfortable operating in a human-plus-AI environment,” he explains.

Despite the challenges, of 92% vs 6% gap, he remains optimistic about the future of agentic AI. As organisations gain experience and develop clearer governance frameworks, he expects the gap between ambition and implementation to narrow significantly.

For enterprises hoping to move beyond pilots, the lesson appears clear—scaling agentic AI is less about deploying smarter models and more about redesigning how work itself gets done.