Enterprises across the globe are no longer asking whether to adopt AI — they’re asking why scaling it enterprise-wide remains so difficult. That’s the central finding of a recent Forrester Consulting study commissioned by FPT, which told a momentum colliding with maturity gaps. Only 26% of decision-makers consider their organisations advanced in AI operationalisation, even as more than half already commit at least 5% of their IT budgets to AI — a figure expected to jump to 87% within the next one to two years.
Agentic AI, in particular, is on a steep trajectory: respondents estimate AI agents currently execute 17% of their organisations’ core processes, rising to 26% within a year and 39% within two.
Yet the report is unambiguous about where the real challenge now lies. As adoption pushes from task-level automation toward full workflow orchestration, the barriers enterprises face are no longer about model capability — they’re about organisational readiness: fragmented data, legacy integration, unclear governance, and, above all, trust.
UpTech Media spoke with Frank Bignone, FPT Software Senior Vice President and Head of Corporate Strategy & Growth at FPT Corporation, to unpack what these findings mean specifically for enterprises across APAC, and what leaders should actually do about them.
Why APAC and Japan are betting on the full journey
One of the report’s more distinctive regional findings is that APAC and Japanese leaders place significantly more weight on end-to-end AI strategy and roadmap execution when selecting a partner — 37% and 43% respectively, compared to 32% globally.
It’s a marked contrast to North America (59%) and EMEA (54%), where enterprises more often prioritise a partner’s ability to simply engineer, deploy, and operate AI systems at scale.
Frank attributes this to both business culture and technical reality. “The business cultures in APAC and particularly Japan are deeply rooted in holistic process improvement. They understand that buying AI tools without a comprehensive operating model simply shifts the bottleneck from writing code to validating it,” he explained.
He added, “Given a significant legacy system footprint in Japan, scaling AI requires meticulous architectural design and a clear roadmap. In fact, 41% of organisations cite integration into legacy systems as the biggest operational challenge, and many allocate 60–80% of IT spending to maintaining their existing environments. They are looking for partners who can bridge strategy, governance, and operations to help them run and transform simultaneously.”
The real barrier to agentic AI isn’t technology — it’s trust
Among the seven biggest challenges enterprises face in developing agentic AI, the report’s top-ranked concern wasn’t technical at all: 40% of decision-makers cited limited organisational trust in agentic AI to make decisions or take action with reduced human oversight — ahead of integration difficulty (38%) and human-agent collaboration design (37%).
Asked what he sees as the single biggest barrier to expanding agentic AI across the region, Frank doesn’t hesitate.
“I believe the ultimate barrier to expanding agentic AI is organisational trust. Our study in collaboration with Forrester also revealed that 40% of decision-makers hesitate to let agentic AI make decisions with reduced human oversight.”
He goes on to mention that while every enterprise wants to accelerate with AI, they are rightly concerned about security incidents, data exfiltration, or losing control of their IP to obfuscated code.
“Trust cannot simply be bought; it must be engineered through audit-ready governance. We must ensure security, governance, and accountability are foundational and embedded from day one,” he said.
Frank also remarked that without such baseline governance and secure integration, one simply cannot build the trust required for autonomous AI operations.
Designing human-in-the-loop checkpoints without losing velocity
The trust gap has a practical downstream effect: 37% of leaders in the study say they struggle to design effective human-agent collaboration loops — the oversight, escalation, and shared-accountability mechanisms needed to keep agentic AI both safe and useful.
For Frank, the answer isn’t a single checkpoint, but a tiered system.
“Our study highlights that 37% of leaders struggle to design effective human-agent collaboration loops. The solution is to balance strict oversight with AI velocity by implementing a dual approach: Human-in-the-loop (HITL) and Human-on-the-loop (HOTL),” he said.
For him, HITL should act as a ‘hard gate’ requiring explicit human approval before an agent commits an action. This is reserved for security-sensitive changes, regulatory compliance, and architectural decisions with a material blast radius.
“Conversely, HOTL acts as a ‘soft gate.’ The agent executes routine, low-risk tasks autonomously, and a human reviews them within a defined service-level window. Supported by continuous audit trails, this dual approach allows organisations to maintain rigorous risk control without sacrificing AI-driven acceleration,” he explained.
The one recommendation that matters most right now
The report’s own recommendations converge on a similar point: enterprises are shifting away from traditional vendor and systems-integrator engagements (44%) toward platform partnerships (62%), managed services (58%), and co-innovation (50%) as their preferred models for scaling AI.
Asked which single action would move the needle fastest for an APAC executive over the next six months, Frank points to partner orchestration:
“My recommendation is to collaborate with a partner capable of orchestrating the entire lifecycle, from strategic advisory to execution and continuous optimisation. This will enable enterprises to benefit from accelerated time-to-market, access to a wider range of talent, resources, and perspectives, and reduced risk. The challenge lies in finding the right fit: partners that are too small lack governance, while those that are too large struggle to adapt quickly,” he said.
He also mentions how FPT operates in that balance, as they bring the global scale to deliver confidently, and the agility to move quickly, adapt to changing requirements, and work side-by-side with clients throughout their transformation journeys.
What separates the scalers from the stuck-in-pilot
Looking ahead, the report finds that enterprises expect accelerating AI initiatives to deliver enterprise-level impact within two years: 80% anticipate stronger governance and control, 80% expect stronger security and resilience, and 77% expect scalable growth without proportional headcount increases.
But Frank frames the dividing line less in terms of technology, and more in terms of integration and people.
“Over the next two years, the dividing line will be the ability to successfully weave technology, talent, governance, and execution into a cohesive transformation strategy. The enterprises that successfully scale will be those capable of elevating AI from isolated pilot projects into a sustainable engine for growth, resilience, and competitive advantage,” he said.
Most importantly, Frank believes successful transformation is driven by people, hence the reason why FPT is investing heavily in building one of the industry’s largest AI-enabled workforces with more than 30,000 AI-augmented engineers, over 5,000 AI engineers, and 3,000 data engineers.
This is continuously reinforced by extensive reskilling and certification programs driven by our strategic partnerships with global AI and tech leaders such as Microsoft, NVIDIA, SAP, and Anthropic.
“We believe successful enterprises of the future will be human-led, agent-operated organisations, and we ensure to develop the talent foundations needed to scale AI and realise lasting business value,” he added.
***
Taken together, the Forrester findings and Frank’s remarks point to the same conclusion: for APAC enterprises, the next phase of AI competition won’t be won by whoever deploys the most models, but by whoever builds the governance, integration, and talent foundations to trust — and scale — what those models can do. As the report’s title suggests, the real work now is turning pilots into platforms.
