Bridging the Gap Between Frontier AI Research and Real-World Value: Key Lessons From D:NEX Asia
“Intelligence is becoming the defining currency of the modern economy.”
While the world is fixated on compute commitments, model breakthroughs, and the extraordinary pace of research, one truth remains stubbornly present: there is still a significant gap between frontier innovation and real-world adoption.
Businesses are excited, but also, understandably, uncertain. They want to use AI to transform their organisations, but don’t always know how to apply it effectively, responsibly, or at scale.
And that gap matters.
Because the companies who learn to bridge it, by turning cutting-edge AI into production systems that actually solve their biggest challenges, will define the next decade.
I had the privilege to join a conversation about capturing the enormous AI opportunity before us, joining Julian Low of Starstorm Ventures, who moderated brilliantly, and fellow speakers Dr Kelvin Chan from Google DeepMind, Featherless CEO Eugene Cheah and Saim Yeong Harng of Synvo AI - each bringing insights on where AI is heading next.
Below are my key takeaways from the session.
Reimagining, Not Incrementally Improving, Core Processes
A theme echoed across the panel: plugging AI into an existing workflow for greater efficiency won’t bring competitive advantage. It will come from reimagining the core engines and processes at play within your business.
Organisations still default to familiar patterns, adding AI to speed up an existing task or reduce cost in one part of the business. It’s understandable, we need to know this works and it can be done, and we need to prove ROI quickly
But the real opportunity lies in rethinking the entire system, not just the process. The companies that win will be those willing to ask: “If we redesigned this service today - knowing what AI capability we can use – what would it look like?”
All industries are shifting toward more specialised AI models focused on reliability and task completion rather than pursuing the smartest possible AI, with techniques like fine-tuning and reinforcement learning being used to ensure consistent performance.
Evals and ROI: Validating Your Way to Value
When executives talk about ROI, the conversation usually returns to efficiency. But as we discussed on stage, the most valuable use cases for AI are growth-oriented, not cost-oriented.
Take Virgin Atlantic. Their brand is built on warmth, service and human connection. Rather than automating a back-office function, we helped them scale that signature service through a virtual concierge powered by real-time voice technology. Customers can now get personalised travel planning and guidance, at scale, without losing the emotional tone that defines the Virgin experience.
To unlock new experiences, new revenue, or new value for customers, use cases have to be validated effectively. New AI models should be evaluated not just on benchmark performance but on what new capabilities they unlock.
Evaluations aren’t about finding the perfect model, but understanding the context of your solution, the data you’re inputting, how AI is being deployed and how to measure performance.
Strong evals frameworks help companies:
- generate confidence in their solution
- confirm solutions are safe, accurate and aligned
- validate value early, before scaling
- ensure consistent performance and continuous improvement
- build trust with stakeholders who must approve deployment
Many leaders ask: “Where do we even start?” With the customer or business problem, not the model. Then apply evals to measure whether AI is genuinely improving outcomes and not just generating outputs.
Ensuring Adoption Through Rapid Deployment of Production-Grade Solutions
One of the strongest messages from the session was the urgency for businesses to move past experimentation. Businesses are running pilots everywhere, but production-ready systems are what bring organisational change.
To navigate this shift, companies need two things: speed and structure. At Tomoro, we typically bring solutions into production within 12 weeks. That speed is only possible because we prioritise:
Production-ready solutions: Well-tested, secure, scalable systems that integrate cleanly into existing infrastructure and demonstrate value immediately.
Real feedback loops: Iterating with real users, real data and real operational constraints.
Credibility and belief-building: Rapid deployment builds momentum. When users experience working systems, adoption accelerates. Appetite grows. Executives unlock more ambitious use cases.
We agreed on the importance of customising technology based on specific use cases, the context of your industry and your long-term goals, instead of adopting one-size-fits-all solutions.
Singapore stands apart for its clarity and ambition. This was clear from the appetite from the audience and the achievements of my fellow speakers and their businesses. Its commitment to trustworthy, well-governed AI provides a strong foundation for scaling intelligence responsibly. The country’s regulatory environment aligns closely with the panel’s focus on both opportunity and ethics, an ideal setting for next-generation AI builders.
Particularly in finance, logistics and public-sector-adjacent industries, we see businesses move fast and have a strong appetite for deployment. It was clear from the session that what they’re often missing is deeper AI engineering capability to make the leap from experimentation to production, avoiding the familiar pitfalls of stalled pilots and the undesired impact on adoption across the business.
Applying technology the right way, early, removes friction later. The organisations that pair ambition with structured, production-minded engineering will be the ones who lead Singapore’s AI transformation.
How to Deploy the New Currency
As intelligence becomes the currency of the modern economy, the winners won’t be those who simply adopt AI. They will be those who architect it, deploy it responsibly, and build systems that reshape how their organisations operate.
Frontier research may capture the headlines, but how it's applied is what will define the next decade of enterprise transformation.
