As enterprises across Asia Pacific accelerate investments in agentic AI, a widening gap is emerging between ambition and operational readiness. While many executives see autonomous AI systems as critical to future competitiveness, organisations are still struggling with foundational issues around governance, accountability, and risk management.
Vinod Anand Bijlani, AI practice leader, Asia Pacific, Hewlett Packard Enterprise (HPE) shares with iTNews Asia his perspective on why governance, not technology, is becoming the defining challenge in the next phase of AI adoption.
According to Bijlani, enterprises are eager to move beyond generative AI into systems capable of taking autonomous actions across workflows and business operations. However, the maturity needed to manage those systems responsibly is lagging behind.
“Ask any enterprise leader in APAC what’s top of their agenda, and agentic transformation will be near the top of the list. But when you ask them whether their governance, security, and controls are ready for it, the conversation gets uncomfortable very quickly. The ambition is agentic. The readiness is not,” Bijlani said.
What are the bottlenecks
Bijlani argued that most organisations are still operating with governance frameworks designed for “human-in-the-loop” workflows rather than autonomous systems capable of making decisions independently. This challenge is particularly pronounced in APAC, where competitive pressure is driving enterprises to accelerate deployments before governance structures are fully mature.
Despite growing optimism around agentic AI, executives remain deeply concerned about accountability, compliance readiness, and workforce implications.
The first concern is ownership and accountability. When an autonomous agent makes a consequential decision and something goes wrong, executives want to know who is responsible. Most do not yet have a clear answer.
- Vinod Anand Bijlani, AI practice leader, Asia Pacific, HPE
Among the most significant fears is the loss of visibility into how autonomous systems make decisions. Bijlani referenced a financial services deployment where an AI customer support agent began surfacing account numbers in outputs despite access controls appearing compliant. “The technology worked exactly as designed. The visibility didn’t exist,” he explained.
He added that this reflects a broader industry problem where traditional oversight models are not designed for AI systems capable of pursuing goals, adapting in real time, and acting independently across interconnected environments.
Shadow AI is escalating into “Shadow Agentic AI”
Bijlani warned that enterprises are entering a more dangerous phase of unsanctioned AI usage, moving beyond simple chatbot experimentation into autonomous AI agents operating inside enterprise environments without governance oversight.
“Shadow AI is ultimately a signal. It tells you that demand for AI capability has outrun the frameworks built to support it,” he said.
According to him, many organisations have unintentionally fuelled shadow AI adoption by moving too slowly to provide employees with secure and governed AI alternatives. He stressed that the rise of open-source agent orchestration platforms significantly raises the stakes because employees can now deploy autonomous agents capable of interacting directly with enterprise systems and sensitive data.
Responsible AI must move beyond policy documents
Bijlani argued that one of the biggest mistakes organisations make is treating AI governance primarily as a compliance exercise rather than an operational engineering requirement.
Organisations that succeed are those treating governance as a prerequisite to scale rather than an afterthought. He added that governance failures often emerge from the disconnect between written policies and real-world deployment practices.
He outlined five critical pillars for responsible AI deployment - strategy alignment, continuous risk management, strong data foundations, operational governance, and dedicated agent oversight mechanisms. He also stressed that governance must include clearly defined ownership structures, cross-functional accountability, real-time monitoring, and human oversight for high-stakes decisions.
“Responsible AI is not just about compliance. It is about building the operational discipline required to deploy AI safely, effectively, and at scale,” he added.
Productivity gains are real but trust determines long-term success
Despite governance concerns, organisations are already seeing measurable value from agentic AI deployments, particularly in customer service and voice-based automation.
However, he also cautioned against measuring AI success through productivity metrics alone. “Agentic AI is a different category of technology, and it deserves a different category of measurement,” he said. Instead, he argued enterprises should evaluate success across four dimensions including value creation, trust, resilience, and accountability. “Productivity that is not trusted does not scale. Efficiency that cannot be explained does not survive regulatory scrutiny.”
As enterprises push deeper into autonomous AI adoption, Bijlani believes trust will ultimately become the deciding factor between sustainable transformation and short-lived experimentation. “The organisations capturing the most value from agentic AI are not necessarily the fastest movers, but the most deliberate ones,” he said.
For APAC enterprises, the challenge ahead is no longer whether agentic AI can deliver value, but whether organisations can build the governance discipline required to deploy it responsibly at scale.
