IBM says ASEAN business leaders are rushing to deploy agentic AI, but gaps in technology integration and elusive ROI reveal a region still struggling with readiness. In its latest study, the company found that 57% of leaders are already adopting AI agents at scale, and 67% see the technology reshaping core aspects of their business.
More than half (51%) also admitted that the speed of their investments has left them with disconnected technology systems: a fragmentation that undermines returns and complicates integration.
Against this backdrop, IBM convened its “Supercharging ASEAN’s Growth with Agentic AI” virtual media roundtable, where experts from Singapore’s Infocomm Media Development Authority (IMDA), the Indonesia AI Society (IAIS), and the IT & Business Process Association of the Philippines (IBPAP) shared both challenges and emerging use cases.
Landscape so far
As a regional leader in AI, Singapore’s strategy is to combine innovation with governance. IMDA is testing agentic AI use cases within industries while also emphasising safety by design, observed Clifton Phua, the agency’s Director of Labs.
“For Singapore, I think we evolved early. We now focus on constrained, high-value workflows that can grow in complexity as user confidence increases,” he said.
Previously, workers feared they would soon lose their jobs to automation and AI. While this concern remains valid, Phua said AI is intended to relieve workers of manual, repetitive tasks so they can focus on more meaningful work.
“We recognise that the real power of agentic AI isn’t in replacing human decisions, but in enriching them. When AI systems can think independently and collaborate meaningfully, every business process becomes an opportunity for innovation,” he said.
Unlike Singapore, which is moving quickly but with safeguards for agentic AI, Indonesia is taking a more cautious approach, requiring consensus from all stakeholders before scaling, said Henke Yunkins, Director of Regulation and Ethics at IAIS.
“What’s great about ASEAN is creating an ecosystem of approaches towards agentic AI. Perhaps what we need to learn as a region is safeguarding against potential issues that may arise,” he said.
In the Philippines, agentic AI is gaining traction in the outsourcing industry, which employs 1.9 million human agents. According to Jack Madrid, President and CEO of IBPAP, AI agents have been deployed to handle tier-one support, which involves basic FAQs and other transactional queries for global clients across banking, healthcare, retail, and telecom accounts.
“We’re seeing productivity gains with reduced handle times, better first contact resolutions, and an increasingly seamless handoff from AI agent to human agent,” he said.
Agentic AI challenges
For Phua, there are two major challenges to implementing agentic AI in enterprises: technical constraints and ROI.
“Right now, AI agents still struggle with long-horizon tasks, reasoning, and when deployed in multiples, coordination,” he said.
Performance of agentic AI systems is still behind humans by 20% to 30% on complex benchmarks, Phua added.
On ROI, Phua noted that the high costs of AI usage often do not correspond to tangible returns.
Further, the integration of these AI pilots into existing legacy systems is proving difficult, Madrid noted.
“I think it’s premature to measure ROI, at least for our industry at this stage. Integration entails some cost, but what I really want the industry to focus on is a more carefully thought-out understanding of each and every business process, where agentic AI can augment the role of our human agents,” he said.
Defying the odds
Despite these barriers, Singapore is working on several proofs of concept, including one with Singapore Airlines.
“They have CRM agents now in PoC stage, and the agentic AI platform so far has been shown to enhance their current CRM system. It does work hand-in-hand, supporting the human agents with real-time suggestions and motivation,” Phua said.
IMDA is also developing a legal workflow agent, as well as HR and social media management agents.
For the IT and BPO industry, nuance is crucial in addressing the varying problems of different organisations with unique needs.
“When people call, text, or message with their problems, they want both resolution and empathy. This is where human touch remains vital. For now, the winning solution isn’t man versus machine, but man working with machine, combining judgment, decision-making, and nuance,” Madrid said.
According to him, some of the efficiencies being observed in healthcare involve voice-based and text-based AI agents being piloted for summarising doctor-to-patient conversations, a significant shift from traditional medical transcription.
For Yunkins, there is no way forward other than trial and error, because AI is a non-deterministic technology. He also reminded organisations not to treat guardrails as an afterthought.
“Each step when you build your AI use case, you should think about how it’s going to affect your users,” he said.
Trust, he added, must be embedded in any AI framework.
“Trust has to be built in the interaction itself, not just a policy document or just compliance, otherwise you will lose customers,” Yunkins said.
In the end, starting with small-scale projects instead of large, high-concept ones can minimise risk and losses.
“It’s always good to start small with well-scoped PoCs and projects. Another key consideration is not just hardware but also software, since these systems often become highly complex,” Phua recommended.
