Nimish Panchmatia is the Chief Data & Transformation Officer of DBS - the largest bank in Southeast Asia.

He joined DBS in 2010 and has been at the center of one of banking’s most ambitious AI transformations. Today, DBS runs over 2,000 AI models across 430 use cases - delivering about SGD $1 billion in economic value in 2025. In this post, Nimish shares how DBS built that foundation, how agentic AI is reshaping the way organizations work, and what it means to be an AI-enabled bank with a heart.

Q: DBS has been consistently recognized as a front‑runner in AI adoption among global banks. At a strategic level, what are the principles and frameworks that have guided your AI journey - from experimentation to enterprise‑wide deployment?

Nimish: We began experimenting with AI as early as 2014, and since then we have done the heavy lifting to clean our data, place it in data lakes, and build the right architecture for us to unlock the power of data.

First, we re‑architected our technology stack to be scalable, robust, and automation‑ready. This meant moving away from legacy systems to open‑source, and investing in a hybrid, multi‑cloud infrastructure to unlock compute power and flexibility. At the same time, we built in‑house capabilities like:

• ADA – our self‑service data platform and single source of truth, ensuring governance, discoverability, quality, and security.

• ALAN – our AI protocol and knowledge repository, creating a standardized, repeatable approach to deploying models across use cases.

We also centralized all 700 Data Professionals under a Data Chapter, to scale up value from Data Analytics and AI/ML by developing a pool of deep experts to deliver business outcomes at scale and at speed.

These efforts enabled us to move from experimentation to enterprise scale. Today, DBS runs over 2,000 models across 430 use cases, touching every part of the bank – from customer engagement and risk management to operations and employee enablement. This foundation also positions us well to rapidly scale Gen AI and Agentic AI use cases.

While technology and process were critical, the people element was equally important. We were, and continue to be committed to bringing employees along on this transformation – through upskilling, AI literacy programmes, and embedding data specialists into cross‑functional squads aligned to customer journeys.

Finally, trust underpins everything. Our Responsible Data Use (RDU) framework and PURE principles (Purposeful, Unsurprising, Respectful, Explainable) guide every initiative, supported by strong governance and human‑in‑the‑loop controls. We also work closely with regulators and industry bodies – leading MAS’ MindForge Data & AI workgroup, contributing to the AI Risk Management Executive Handbook, and shaping Generative AI guardrails for banking.

Our efforts have allowed us to scale AI responsibly and earn global recognition, including winning Global Finance’s “World’s Best AI Bank” in 2025.

Q: Could you share some of your AI use cases and their impact to the bank and its customers?

Nimish: Our approach to AI is centered on delivering measurable business value at scale, while ensuring strong governance and trust. Fundamentally, we organize our AI deployments across horizontal use cases, which provide enterprise‑wide capabilities for all employees, and vertical use cases, which are deeply embedded within specific businesses, customer segments and job roles. This allows us to industrialize common capabilities across the bank, while applying AI in targeted ways that enhance customer experience, reduce employee toil and augment human judgment.

At the horizontal level, DBS‑GPT, a secure internal assistant, helps employees accelerate the search, synthesis and application of bank policies, institutional knowledge and internal content.

We have also deployed a range of vertical, role‑specific AI solutions:

• DBS Joy, our Gen AI‑enabled banking assistant, was introduced in late 2025 to provide 24/7 support to corporate customers. Joy delivers dynamic, contextual responses through natural conversations, remembers past interactions and seamlessly escalates complex issues to live agents, contributing to higher customer satisfaction.

• Wealth Co‑Pilot consolidates insights from multiple sources to help relationship managers deliver more tailored, relevant recommendations, augmenting human advice and improving the quality of client conversations without replacing human judgment.

In 2021, we were one of the first banks to quantify the economic value of AI. Last year, our data analytics and AI/machine learning initiatives delivered approximately SGD 1 billion in economic value in 2025.

Q: With the rise of agentic workflows and AI copilots, how do you see the shape of organizations evolving? Will this lead to leaner teams, new roles, or entirely new ways of working?

Nimish: At DBS, we expect leaner toil, not leaner teams. The fundamental shift is not about reducing headcount, but about redesigning roles and adopting an AI‑augmented way of working. As AI agents become more capable, they will increasingly complement our human workforce by taking on time‑consuming and repetitive tasks. This allows employees to focus on higher‑value work where human qualities such as judgment, creativity, empathy and complex problem‑solving remain essential and irreplaceable.

This transition is already reshaping how work is organized across the bank. We are redesigning job roles and redeploying employees into new career pathways as AI changes the nature of frontline and specialist work. At the same time, AI adoption has created entirely new areas of work, including AI governance, risk management, agent oversight, and agentic AI development and engineering.

Upskilling is a critical enabler of this shift. At a horizontal level, all employees are equipped with a baseline understanding of AI and practical exposure to tools that reduce manual toil and enhance productivity. In parallel, we have identified more than 11,000 employees in AI‑impacted roles for deeper, role‑specific capability building through vertical use cases, ensuring they can work confidently and effectively alongside AI in their respective domains.

Q: Your CEO Su Shan has spoken of her vision of DBS as an AI‑enabled bank with a heart. What does that mean to you?

Nimish: As we continue to scale the use of AI across the bank to deliver a seamless One Bank proposition to our customers and supercharge our employees, we will do so while retaining the human elements of empathy, judgment, creativity and cultural nuance. In doing so, we want to elevate AI from just artificial intelligence to authentic intelligence, combining the best of machine and human qualities to deliver trusted banking solutions, while upskilling and reskilling our people to be future‑ready.

A 5-time consecutive Midas List investor, Nigel Morris is the Co-Founder & Managing Partner of QED Investors, which has backed 30+ fintech unicorns. Prior to QED, Nigel also cofounded Capital One, where he was the president and COO.

In this conversation, Nigel unpacks what he looks for in generational founders, why venture is not stock-picking, and where the next wave of fintech opportunity is emerging - especially as AI and stablecoins shift from hype to industrial strength.

Q: You were reflecting recently on “four chapters” in your career. What are they, and how did they shape the way you invest today?

Nigel: I feel incredibly lucky. I was reflecting - just yesterday - that it’s been 40 years since I graduated from London Business School. And 40 years is a blink of an eye.

I’ve had four amazing experiences in those 40 years, all of which triangulate into my perspective today.

Chapter one: Coming out of LBS, I had no business experience. I was an experimental psychologist and statistician. I went into strategy consulting - at a very quantitative BCG spin-off - because I thought I could hold my own with arithmetic even if I didn’t understand business.

Chapter two: I left consulting to join a client - Signet Bank in Richmond - where I learned to go from talking about stuff to actually being an operator. You learn so much. You also earn scars on your back from mistakes.

Chapter three: The momentum to take Capital One public and spin it out in 1994. That was a decade-plus of building out the model and running a public company with my partner, Rich Fairbank.

Then in my mid-40s, I took a year off. I thought I might be done. But I wasn’t.

Chapter four: Starting what became QED - with Frank Rotman and Caribou Honig - before fintech existed as a concept. We applied the same frameworks and heuristics that worked at Capital One to emerging technology businesses around financial services.

Q: QED has backed an unusually high number of iconic fintech companies. Why did you focus on building a fintech-only firm - and how did you make sure you “see the deals”?

Nigel: A critical tenet of the QED model is that you can only do the deals that you see.

If all the deals go to other marvelous firms - then you don’t get to see them, you can’t do them. Even if seeing them doesn’t guarantee success, not seeing them guarantees you won’t be successful.

So we’ve worked really hard to be approachable - to be “promiscuous” in the marketplace, to help, and to be very positive to young founders. We’ve made over 200 investments, and we’ve built a brand and experience base that we work hard to leverage.

And I also genuinely believe fintech is a force for economic and social good. It takes away friction, creates transparency, and challenges incumbents who often confuse loyalty with inertia. As we move toward open banking and agentic AI, barriers to switching keep diminishing - and that’s going to be a field day for fintechs.

Q: From your perspective, what defines a “generational founder”? Is there a recipe?

Nigel: I wish there were a formula, but there are patterns.

First: They’re massively driven. And because I resemble this myself, I have permission to say it - often there’s something wrong with them. They have a desire, passion, energy, resolve, tenacity that’s three standard deviations beyond normal. They’re driven to solve a problem and succeed. The reasons vary - childhood stories, chips on shoulders, who knows - but something ignites them.

Second: They have to be charismatic and passionate enough that people will follow. These companies aren’t built by individuals. They’re built by teams. And founders need to bring people along.

Third: I love when there’s a complementary partnership early on - often a visionary / communicator paired with someone more execution-oriented or technical. That dyadic relationship is powerful. Capital One wouldn’t have happened without Rich and me together. You see similar patterns across many great companies.

Finally: The best founders are the best listeners. Success is almost never linear. It’s non-linear and complicated. You have to know when you’re at a decision node - left or right - and when to pivot. The people who hold onto an idea after the data stops supporting it for too long often die.

There’s a delicate balance: don’t believe you know everything; think in decision trees; surround yourself with talent; but also be confident enough to lead.

Q: You mentioned founders are trained to “gild the lily.” What do you actually value in a pitch?

Nigel: We train founders to put their best foot forward and tell the most exciting story. And when did you ever see a pitch deck that didn’t show revenue going to infinity?

But we don’t always give founders enough credit for being balanced.

I love the founders who are constantly looking at what could go wrong, and protecting themselves against it - rather than being Pollyannish. The odds are against you building these companies. You need all the help you can get.

Q: You said something that really stood out: “venture is not stock picking.” What do you mean - and what does QED try to be for founders?

Nigel: Venture isn’t: “here’s a dollar, off you go, bring me back five dollars.” It’s not that.

Financial services is full of complex and nuanced risk - treasury risk, fraud risk, credit risk, compliance risk, regulatory risk. We understand those, because we’ve seen them over two lifetimes.

Our job is to subtly shift the probability odds of success by helping founders build, create, and scale.

Trust is everything. I call it the Ghostbusters effect.

When founders are doing well, they’ll call and tell you they blew their quarterly numbers away - great. But when they’re in a pickle - confused, angry, frustrated - who do they call?

We punch above our weight because founders call us then. We won’t overreact. We’ll be honest, practical, pragmatic. And if we don’t know the answer, we’ll find someone who does - through a deep network: ex–Capital One diaspora and now the talent spun out from 200+ companies we’ve backed.

Q: Looking ahead 5-10 years, where do you still see big opportunity in fintech, especially now with AI and stablecoins?

Nigel: If you asked me five years ago, I probably wouldn’t have said stablecoin, and I wouldn’t have said AI. And I would have rolled my eyes a year and a half ago about whether stablecoin had an industrial-strength application.

Things are changing really fast.

At the macro level: financial services revenue is about $14 trillion. Fintech - if you add it all up - is about $400 billion. That’s ~3% penetration.

Fintech grew ~21% last year in revenue. Incumbents grew ~6%. At that delta, in 10 years fintech is about 10% penetrated. So we’re still early.

Incumbents - banks and insurance companies - are great at protecting the corpus and avoiding mistakes. But what they don’t do is innovate. Fintech is like the Galápagos Islands - Darwinian survival of the fittest. Small companies are fighting for oxygen and trying to reproduce. Incumbents should watch that ecosystem closely, put tentacles deep into it - invest, partner, even buy. Don’t rely on slide decks from Goldman or consultants. Get video recordings, not photographs.

On AI

AI is alchemy - turning lead into gold.

In the next 12–24 months, most benefits will be automating processes and reducing cost, especially in call centers and manual operations. Places with large call-center workforces will come under real siege over the next few years.

Then AI starts making business models economically viable that previously didn’t work. You can tailor messaging and targeting in bespoke ways. In the Capital One days we used to say we delivered the right product to the right customer at the right time at the right price - but we did it in batches. In a world of LLMs, that becomes far more personalized and dynamic - lifting response rates, reducing CPAs. Fintechs will be early adopters because they have to be.

On stablecoins

Stablecoins are now exploding in a way I wouldn’t have predicted - especially for cross-border and protection against currency devaluation. That’s genuinely exciting.

On inclusion

Exponential inclusion is massive. I added up the customers across the businesses I’ve been involved with - Capital One, Klarna, Nubank, Credit Karma, ClearScore and others - and I get to about 600 million. I’ve set a goal to get to a billion.

Where will that come from? Indonesia, India, Nigeria, Latin America - huge populations that still have limited access to financial services. We’re going to see a Credit Karma in Africa. We’re going to see a Capital One / Nubank equivalent in India - variations on these themes in the developing world, leapfrogging the West.

And being global helps. We’ve seen models work in one geography and transport the IP elsewhere - like building ClearScore in the UK on the early framework of Credit Karma, and expanding from there.

Hussein Kanji is the Founding Partner of Hoxton Ventures, one of Europe’s most respected VCs, known for backing generational companies like Deliveroo and Darktrace at their earliest stages.

Q: Hussein, before we dive in, could you share a quick overview of your journey and what Hoxton Ventures is about?

Hussein: Hoxton started with what was considered a slightly rogue idea at the time: building a truly early-stage venture firm in Europe, based in London, but with the ambition and operating style of top West Coast firms. The goal was simple - invest early, be proactive, sit on boards, and help European companies win mindshare and customers in the U.S.

When we started around 2012–2013, almost no one in Europe was doing that at seed or Series A. Today, our latest announced fund is ~$200M, up from a ~$28M first fund. Total amount raised across multiple funds is ~$450m. Our portfolio is roughly 50% enterprise software, 30–35% deep tech, and the rest consumer. We invest across Europe - about 50% UK, 25% Western Europe, and 25% Eastern Europe, which we’re particularly bullish on due to the depth of engineering talent.

Q: You’ve backed companies like Deliveroo and Darktrace at very early stages. How do you assess founders when there’s little revenue and often no product-market fit yet?

Hussein: It depends on the category. In deep tech, the invention itself can be the signal - something genuinely novel that few others could build. That gives you a clear anchor.

In other cases, like Deliveroo, it was triggered by traction. When I first met Will Shu (Founder of Deliveroo), I was actually quite skeptical. At the time, food delivery didn’t obviously look like a massive market, and the version of the idea that existed then felt quite narrow. My initial instinct was that it might be too small to justify the kind of outcome venture investors look for.

But then the product launched - and the data changed everything.

Very quickly, Deliveroo was growing around 5–7% week-on-week, which is an extraordinary rate at that stage.

What stood out just as much was how Will was operating. He was doing deliveries himself, running the company out of a small apartment, deeply involved in every part of the business. There was no separation between “founder” and “operator.” He was living inside the problem.

At that point, the original concerns about market size mattered less. When you see:

• Strong, sustained early growth, and

• A founder who is completely immersed in execution

you pay attention. The combination is powerful.

That’s also why I don’t believe there’s a single template for early-stage investing. In Deliveroo’s case, it wasn’t deep tech or a breakthrough invention that drove the decision - it was traction and founder commitment showing up very clearly, very early.

At the end of the day, checks get written for different reasons: breakthrough innovation, early traction, or occasionally sheer founder quality. There’s no single formula.

Q: Many tech categories feel “winner-take-most.” How do you think about market structure?

Hussein: Most global technology markets are winner-take-most by default, unless there’s a strong reason they aren’t - things like regulation or geographic fragmentation.

Social networks are the classic example. There were many strong local players around the world, but the vast majority of the value ended up accruing to one global winner. You see similar dynamics in enterprise software. In CRM, for instance, there are lots of vendors, but most of the market cap sits with Salesforce.

That doesn’t mean you can’t build a good business as a number two or three. You can. Markets are often large enough for that. But if you’re aiming for truly outsized outcomes, you generally want to back the company that has a real shot at being number one.

The difference between a very large outcome and a good one can be enormous - orders of magnitude apart.

Q: For founders outside the U.S., how should they think about competing globally against Silicon Valley companies?

Hussein: In reality, you’re usually optimizing either for downside protection or for upside potential - and the two lead to very different outcomes.

If you optimize for downside, you focus on building a strong regional business where you have local advantages and less direct competition. That can work, and it can produce good companies. But it also tends to cap the size of the outcome, because it assumes global players won’t move in aggressively.

If you want to build a truly massive company, you eventually have to compete where the category is being defined. In most technology markets, that means the U.S. market. It’s not just about size - it’s where customer expectations, pricing, and narratives are set.

Stripe is a good example. The founders were Irish, but they moved early and oriented the company around the U.S. Had they stayed in Ireland and optimized locally first, the outcome would almost certainly have been smaller.

My bias is that if you know you’re playing for the biggest outcomes, it’s often better to make that decision early, before the company hardens around a more local optimum.

Q: In AI, many people argue that product moats no longer exist—that velocity matters more than defensibility. Do you agree?

Hussein: I actually think moats matter more than ever, not less.

Right now, the industry is very focused on short-term signals: fast revenue, fast markups, fast fundraising. AI makes it easier than ever to show early traction. But early traction doesn’t necessarily translate into durable value.

We’ve seen this before. In periods where technology is moving very quickly, entire business models can disappear almost overnight. Companies that looked incredibly strong early on simply didn’t persist.

If you don’t have a real moat - whether that’s data, distribution, deep integration, or something else that compounds over time - then your claim to terminal value is fragile. You might grow quickly, but that doesn’t mean the business will still matter in ten years.

History has a lot of examples of this, from the dot-com era through to more recent cycles.

Q: Given all that uncertainty, which areas are you most excited about right now?

Hussein: Deep tech remains very exciting for us. Areas like materials science, for example - companies working on recreating rare earth materials in labs - are genuinely interesting.

Robotics is another area where we’re seeing step-function improvements. It was relatively stagnant for a long time, but progress is accelerating now, with noticeable improvements every six months or so.

That said, as venture investors, we don’t pretend to know exactly where the next big winner will come from. Our job is to stay close to builders, build a portfolio that reflects uncertainty, and be ready to recognize real breakthroughs when they appear.

Rodrigo is currently President (Asia) at Microsoft. Previously, he served as President (Americas Enterprise) and President (LatAm), leading the company across some of the world’s most complex regions. In this conversation, we spoke about leading at scale across diverse regions and cultures, and how AI is reshaping enterprise workflows.

Q: Rodrigo, could you start by sharing your journey - and how you ended up leading Microsoft in Asia?

Rodrigo: I never really planned my career in a linear way. I grew up in Brazil and originally thought I’d become a medical doctor, but at the last minute switched to engineering. I studied industrial and mechanical engineering, and somewhat unexpectedly began my career in finance at IBM.

Jobs were scarce at the time, so I leaned in and learned finance properly, even completing a master’s in the field. I spent more than a decade in finance roles, eventually becoming CFO for Brazil and later Latin America.

That background became a real differentiator. I’ve always been passionate about technology, but finance gave me a structural lens on how businesses actually work - allowing me to bridge both worlds.

In 2016, I moved to the U.S. with my family, looking for a new challenge. Not long after, Microsoft approached me with an opportunity to lead in LatAm: a region they felt held untapped potential. Given its complexity and pace, it felt like the right next chapter.

Q: You’ve led Microsoft across Latin America, the U.S. enterprise business, and now Asia. How does the nature of operating and leading at scale differ across these regions?

Rodrigo: There are both similarities and very real differences.

In Latin America, you’re dealing with over 30 countries - but mostly just two languages: Portuguese and Spanish. That reduces complexity significantly. Asia is fundamentally different. Languages, cultures, and business norms vary dramatically between Japan, China, Southeast Asia, and India. The cultural intersections are much smaller, which means you have to connect locally in very different ways.

Another major difference is economic dependency. Historically, Latin America has been closely tied to the U.S. and Europe. Asia is far more independent - economically and in terms of identity. If the U.S. and Europe slow down, Latin America feels it immediately. The world, however, cannot simply “shut down” Asia.

What’s exciting in Asia is that innovation is increasingly local. Companies are being built here and scaling globally. That fundamentally changes the GTM dynamic.

Q: How do you get highly diverse teams across cultures and time zones to truly work as one?

Rodrigo: The most important thing I’ve learned is to listen first.

People often over-index on differences - but in reality, we’re far more similar than we think. We all have families, ambitions, and pride in our work. When I brought my Asia leadership team to Seattle recently, people at HQ were surprised by how close-knit we were — how much we trusted and defended each other.

That trust comes from curiosity and humility. When you land in a new market, you must assume that what worked for you over the past ten years may not be the best answer anymore. That’s not a natural muscle for leaders - but it’s essential.

Q: Microsoft is at the forefront of AI. Where do you see enterprise AI adoption taking hold first, and why?

Rodrigo: AI is fundamentally different from past technology waves. The PC, internet, and cloud were largely about democratizing technology - making powerful tools cheaper and more accessible. AI, however, is about democratizing knowledge.

For the first time, expertise that once sat with a small group of specialists can be made available, on demand, to anyone inside an organization. That shift has far-reaching implications for how work gets done, how decisions are made, and how value is created.

In the near term, adoption is happening fastest where there’s a clear combination of three factors: labor‑intensive processes, large volumes of data or documents, and well‑defined rules or frameworks. That’s why areas like customer support, compliance, KYC, and financial crime are seeing rapid traction - the problems are well scoped, the data already exists, and the impact is easy to measure.

The ROI in these use cases is very clear. Machines already outperform humans in many of these workflows - not because people aren’t capable, but because AI doesn’t get tired, doesn’t miss details, and can operate consistently at massive scale. Once organizations see that impact in one function, it becomes much easier to expand AI into adjacent workflows.

Q: What do you think is the biggest challenge enterprises face in adopting AI meaningfully?

Rodrigo: The challenge isn’t experimentation - it’s integration.

Many enterprises are running dozens, sometimes hundreds, of AI pilots. The problem is that these pilots often live on the edges of the organization, disconnected from core systems, incentives, and decision-making. As a result, they generate excitement, but very little sustained impact.

AI’s real value emerges only when it is deeply embedded into day-to-day workflows - not as a standalone tool, but as an invisible layer that augments how people actually work. Over time, every employee will have an AI companion, much like everyone has Excel today. People will build agents the way they once built spreadsheets - to manage information, automate repetitive tasks, and support better decisions in real time.

This shift requires enterprises to rethink processes end to end. It’s not about adding AI on top of existing workflows, but redesigning those workflows with AI at the core - redefining roles, changing handoffs, and updating how performance is measured.

The companies that ultimately win won’t be the ones running the most pilots. They’ll be the ones willing to do the harder work of re‑architecting their operating models around AI, so that productivity gains and decision quality compound across the organization.

Before transitioning from a Board Member role to become President (Revenue) at Cloudflare, Mark Anderson served as CEO at Alteryx (acquired by Insight Partners for US$4B+) and President at Palo Alto Networks.

Few companies sit at the intersection of security, performance, and AI quite like Cloudflare. From protecting over 7M websites to powering ~20% of the world’s internet traffic, its influence is felt every time a page loads faster - or an attack is blocked before it begins.

In this conversation, Mark shares what it takes to evolve a GTM engine at scale, the qualities he looks for in great sales leaders, and why the future of cybersecurity - and AI - will separate the prepared from the unprepared.

Q: You were a Cloudflare board member before stepping in as President of Revenue. What drew you to make that transition?

Mark: I joined Cloudflare’s board back in 2019 - the year we went public. It was a much smaller company then, maybe doing around US$170M a year. Over my five years on the board, the business grew nearly tenfold.

I fell in love with what the company does. Cloudflare democratizes access to a faster, safer internet. Today, over 7M customers and developers use us, and we add roughly thousands of new users every month. That kind of reach matters.

When I left Alteryx, a few Cloudflare board members asked if I’d consider coming off the board to help scale operations. It felt like the right time to take what I’ve learned and help the company serve not just startups or digital natives, but enterprises and governments at global scale.

Q: Even at nearly US$2B in revenue, Cloudflare continues to grow its million-dollar customer segment. What’s the secret behind that GTM momentum?

Mark: I’ve always been drawn to companies at inflection points - where the industry is shifting, and the company has a better way to solve the problem.

At F5, we built a better mousetrap - our ROI was nine months, so adoption took off. At Palo Alto Networks, I joined when it was a US$100M business and left when it hit US$3.5B.

Across all those journeys, one thing stays constant: you need the right people, with the right stage experience, pointed in the right direction. The people who got you to US$1B aren’t necessarily the same ones who’ll get you to US$5B.

For Cloudflare, that means hiring talent who’ve sold to large enterprises - banks, governments, global Fortune 500s - people who’ve lived through hypergrowth, made mistakes, and learned how to operate at escape velocity.

You can’t teach that on the job. You have to have lived it.

Q: What do you look for in great sales leaders?

Mark: Character and stage experience.

You can teach how the technology fits together, but you can’t teach judgment, curiosity, or integrity.

When I’m hiring for a role - say, an account manager for DBS Bank in Singapore - I’m looking for someone who’s earned the permission to be a trusted advisor. Someone with intellectual curiosity, strong ethics, and who’s navigated long-term growth - 30%+ year over year for three to five years.

Hypergrowth changes everything. By the end of each year, you’re running a very different company than you were at the start. That means you need a culture of non-stop professional development - people who want to learn every day and sharpen their craft continuously.

Q: Large sales organizations often struggle to adopt AI. How is Cloudflare approaching this?

Mark: You have to - because everything is moving so fast.

Look at how quickly the AI landscape evolved. A year ago, nobody was talking about agents; now, they’re redefining workflows. In two or three years, we’ll look back and realize we’re only at Act One.

At Cloudflare, we use AI both internally and across our network. Internally, our GTM teams leverage internal models that analyze how customers use our portfolio of 80+ services - surfacing patterns that sharpen value propositions, uncover expansion signals, and customer journeys.

Externally, AI helps us detect threats faster across the significant share of global internet traffic we see each day. We see about 25% of the world’s internet traffic daily. By analyzing patterns in that traffic, we can identify indicators of compromise and predict attacks before they happen. That’s the power of scale combined with intelligence.

Q: AI is reshaping both offense and defense in cybersecurity. How do you see this playing out?

Mark: The challenge isn’t just new data - it’s legacy data.

Many large organizations still run mission-critical systems on decades-old infrastructure - mainframes that can’t support modern AI workflows. You can’t run agents off a legacy system sitting in a basement data center.

So the real race now is modernization. Companies have to rebuild their infrastructure to even begin benefiting from AI. Every CIO, CTO, and CISO I speak with feels the pressure - they know that what they’ve built over the last 30 years will matter far less in the next five.

Modernization isn’t optional anymore. It’s survival.

Q: Some say AI will make companies more efficient - but with fewer people. How do you view that shift?

Mark: I’m an optimist. I’ve lived through multiple technological revolutions - the rise of the internet, the birth of cloud computing - and every time, the world adapted.

What’s different now is the speed. The gap between winners and laggards will widen because AI is not a cheap game - companies are spending tens of billions on GPUs and infrastructure.

But AI isn’t just replacing work; it’s amplifying it. It’s making high-impact engineers even better, and it’s making the other 90% more productive.

That same dynamic will extend to knowledge workers. The companies that thrive will be the ones that reimagine roles, retrain talent, and harness AI as a multiplier, not a replacement.

Fresh off an IPO in July 2025, Figma has big plans for global expansion. Scott Pugh, Figma’s GM (APAC), joined as the company’s first employee in the region three years ago. Since then, he’s built Figma’s go-to-market organization from scratch - localizing playbooks, cultivating community, and helping the world’s biggest brands reimagine collaboration.

In this conversation, Scott reflects on the evolution of design in APAC, the interplay between product-led and community-led growth, and how Figma balances speed, scale, and creativity across one of its fastest-growing regions.

Q: Before we dive in, could you give us a quick introduction - who you are, how you came to join Figma, and what you do today?

Scott: Sure. I’ve been at Figma for three years now - in fact, it was my three-year anniversary just last week. It’s been a phenomenal journey, both personally and for the company.

I joined as Figma’s first employee in Asia-Pacific, so I’ve had the opportunity to build the entire go-to-market function here from scratch - hiring the team, shaping our regional strategy, and helping some of the most innovative organizations transform how they build products.

What’s been incredible to see is not just how Figma has evolved as a company, but how the entire product development lifecycle itself has transformed. The way teams design, collaborate, and ship products today looks completely different from a few years ago - and Figma’s platform has evolved in lockstep to support that innovation.

Before Figma, I was at Mixpanel, where I built and scaled the APAC business for their product analytics platform. Prior to that, I spent seven years at LinkedIn - I was their first employee in Hong Kong before moving to Singapore, where I’ve now been based for about a decade.

Q: What were some of the key inflection points you experienced while building Figma’s presence across Asia-Pacific?

Scott: When you’re scaling a business in a region as diverse as APAC, the first thing is understanding where you already have product–market fit - where the early signals show you’ve found strong traction, a loyal user base, and those initial “beachhead” customers you can build on.

Figma has been product-led from day one. Our platform was designed for collaboration - people bring others in to view, comment, and co-create - so virality is built into the product itself. That natural network effect fits perfectly with a product-led growth motion.

When I joined three years ago, Figma was already nine years old and had a sizable user base in Asia-Pacific, but most of that demand was being served from North America. To truly scale, we needed to be closer to our customers - meeting them face-to-face, understanding local workflows, and building real relationships in-market.

Since then, we’ve not only grown a strong customer base but also an incredible community. We often talk about “community-led growth” alongside product-led growth, because the design community in this region is so vibrant, generous, and collaborative. Designers here love to share what they’ve learned, and Figma has been able to create the platform and environment to bring that energy together.

So, if I had to pinpoint the inflection points, they’re really threefold:

1. Recognizing where product–market fit already existed and doubling down there.

2. Localizing presence to engage customers directly across key markets.

3. Empowering community-led momentum to scale organically beyond what any traditional GTM engine could achieve.

Q: Figma has strong product-led and community-led roots, but enterprise is often seen as a very different game. How do you reconcile that DNA with scaling in the enterprise space - especially across markets with different levels of tech maturity?

Scott: I’d actually say we’re not moving into the enterprise space - we’re already deeply embedded there. Figma today is a foundational tool for roughly 95% of the Fortune 500 and 78% of the Forbes Global 2000. We’ve achieved that scale through a product-led motion, not by abandoning it.

It started organically. Designers within enterprises began using Figma alongside other tools. They loved the experience, started inviting teammates, and soon, entire product and design orgs were collaborating on Figma. That bottom-up momentum created internal champions - heads of design, principal designers, or product leaders - who drove wider adoption across their companies. Often, it began with a simple self-serve account before formalizing into enterprise contracts.

The beauty of Figma’s model is that it reflects how modern work actually spreads - one enthusiastic user at a time. But at scale, enterprise adoption naturally brings more complexity. You have multiple stakeholders - legal, compliance, security, procurement - all part of the buying decision. At that point, you need an enterprise GTM motion layered on top of the product-led foundation.

So for us, product-led growth remains the entry point, but sales becomes the multiplier. As I like to put it: product-led growth opens doors, but sales teams close them. It’s that combination that allows Figma to remain true to its collaborative roots while scaling as a critical platform across the world’s largest organizations.

Q: From your perspective, how is AI transforming the way people design and create - particularly across such a diverse region like Asia-Pacific?

Scott: The shift has been dramatic. In just the last six months, the way people build digital products has completely changed. The barrier to entry is almost gone - anyone with an idea can now prototype a product in days and pitch it with something tangible. It’s the most exciting time I’ve seen in tech in my 20 years.

Figma is uniquely positioned to support this new wave of AI creativity. We started as a UI/UX design tool, but the platform has evolved fast - with FigJam for ideation, Figma Design for high-fidelity work, Dev Mode for seamless handoff, and now Figma Make, which lets users go from prompt to product directly in code.

What makes it powerful is context. Companies already build their design systems in Figma, so when they use Make, the AI generates prototypes that are on-brand and consistent with their existing design language.

AI will make it easier than ever to build - but with that, quantity will explode, and quality could suffer. That’s why we believe design becomes the ultimate differentiator. In a world where AI can generate anything, the best products will still stand out by design.

Q: Across such a diverse region, do you see differences in how countries or industries are adopting AI and Figma’s tools, or is usage relatively consistent?

Scott: There are definitely differences - both by country and by industry. Markets vary in their maturity around AI adoption, design systems, and how they build products. But more than geography, we actually see the biggest contrasts by vertical.

Digital natives and startups tend to move fast - they experiment, iterate, and adopt new tools like Figma’s AI features almost immediately. In contrast, sectors like banking and financial services are more cautious. Some, like Commonwealth Bank of Australia, are leading the way and deeply engaged with our AI capabilities, but others are still in early stages - it’s a bit like turning a cruise ship; it just takes more time.

Regionally, it’s impossible to treat APAC as one market. It’s really five or six distinct subregions with different levels of maturity. Take India, for example - we’re opening our first office there in November. It’s a developer-led market with incredibly sophisticated teams who push Figma’s boundaries, building custom plugins and workflows that often inspire what we later bring into the native platform.

So yes, adoption patterns differ - but across all markets, the direction is the same. Whether it’s startups in Jakarta or banks in Tokyo, everyone’s moving toward faster, more collaborative, and increasingly AI-powered ways of designing and building products.

Genspark is one of the fastest-scaling AI startups of 2025, hitting US$50M ARR in just five months! Co‑founder and COO Wen Sang talks about how Genspark evolved from a search startup into one of the most talked‑about agentic AI companies today - and what it takes to build something enduring in one of tech’s fastest‑moving markets.

Q: Wen, could you start by sharing your journey before Genspark - and how the company came about?

Wen: I’m an engineer by training - I got my PhD from MIT back in 2013. After that, I started my first company, an enterprise SaaS platform that helped municipalities and commercial real estate owners manage their parking assets. The company backed by Y Combinator, Khosla Ventures, and a few others, before it was acquired. I stayed on for about a year, then left in 2023 - around the time I met my co-founder Eric Jing.

Eric had just moved to Silicon Valley, and we got introduced by a mutual friend. We became friends first, then co-founders. I was actually working on building a car at the time, but after a few conversations with Eric, it was clear there was a much bigger wave happening in AI. That’s how Genspark began.

Q: What was the early vision - and how did it evolve into the super agent suite?

Wen: Our first product was actually focused on search. Eric had deep experience in that space, and we wanted to build the next generation of search technology.

We reached about five million users, but we quickly realized that the future of search wouldn’t be about search itself. Access to information was becoming so cheap and abundant that it would be hard to build lasting value around it.

So we asked ourselves a simple question: Why do people seek information in the first place? Most of the time, it’s to get something done. That led to the idea of going a step further - rather than just giving people information, could we help them actually complete the task?

That’s how our super agent suite was born earlier this year. Within five months of launch, we hit US$50M in ARR - a sign that the world was ready for this shift.

Q: That’s remarkable growth. How do you think about standing out from so many AI agents launching globally?

Wen: Honestly, we don’t think much about competition right now. The majority of the world is still catching up to what AI truly is - most people are just starting to experience its potential.

For us, the focus has always been the product. What really differentiates us is the quality of the output. Users are smart - they know if an AI tool actually saves time or just creates more work.

We design for what I call the “one-shot outcome.” People should feel like they can issue a command, get the result they need, and move on. That reliability and accuracy are what drive adoption. Within minutes, users see value - and that’s what keeps them coming back.

We also ship new products every week. Constant innovation is the only way forward in this space.

Q: Some people say in AI, the “loudest” players often win. How do you approach go-to-market and visibility?

Wen: Our growth so far has been entirely organic, fueled by genuine word of mouth. When something truly delivers value, people naturally share it - that’s the most authentic marketing signal you can get.

In the early days, we kept marketing intentionally light to test if product-led demand would hold on its own. When we saw that it did, we started introducing community programs, creator partnerships, and selective awareness campaigns to amplify the buzz.

Rather than chasing attention, our philosophy has been simple: build something people love so much they tell others about it.

Q: Churn and retention are common questions for consumer-facing AI products. How do you think about that?

Wen: It’s something we watch very closely. Our recent monthly paid retention has held steady at 88–92%, which is strong for an individual-user product - especially one that began fully consumer-focused. But retention isn’t just about percentages; it reflects how essential the product becomes in people’s daily routines.

Still, the core driver remains simple: product quality. If the output is fast, reliable, and genuinely saves time, users stay. Retention can’t be “fixed” through marketing; it’s earned through building consistent, dependable value over time.

Q: Margins in AI are a big talking point. How do you view the economics of running agentic products?

Wen: It’s fair to say margins in AI are messy right now - it’s still the Wild West.

On one hand, inference costs have dropped dramatically and we expect them to continue trending down, eventually close to electricity costs.

On the other hand, we run a hybrid architecture that balances closed-source and open-source models. We built a model-orchestration layer that intelligently routes different tasks to different models.

That, combined with our scale - we’re processing over a trillion tokens a month - gives us preferred rates with providers like OpenAI and Anthropic.

So while we’re not optimizing purely for margins right now, efficiency improves naturally with volume. For us, the priority remains the same: optimize for quality first.

Q: When you think about the future of work through this lens - what does it look like?

For most of history, humans were the “production machines.” We learned to use tools, then did the work ourselves.

Now, with super agents, the workflow flips. You delegate, the agent does 80–90% of the work, and you add the final touch. It’s like having a team of analysts in your pocket.

No one complains that washing machines took laundry jobs away - they just made life better. AI will be the same. It’s not about replacing people; it’s about elevating everyone’s potential.

From US$0 to US$50M ARR in just five months (and cashflow positive), Higgsfield AI hit numbers rarely seen, even in the AI era. For CEO and Founder Alex Mashrabov, this breakout success was years in the making.

A former Snap Inc. product leader in Generative AI and co-founder of AI Factory (acquired by Snap in 2019), Alex helped create the viral face filters and Cameos that defined a generation of digital expression.

Today, he’s pushing boundaries again - this time, turning generative video into the next creative medium.

In this conversation, Alex shares how he thinks about product readiness, differentiation in a crowded field, and the discipline behind scaling one of the fastest-growing AI companies in the world.

Q: You’ve shared that the name “Higgsfield” comes from physics. What’s the story behind it?

Alex: We named the company Higgsfield because, in physics, particles that pass through the Higgs field gain mass.

If you think about today’s internet - we live in an attention economy. Videos are what have mass. They’ve become the language of the internet.

So, the idea is: ideas that pass through the Higgsfield platform gain mass - they transform into videos that capture attention. It’s both a metaphor and a mission statement.

And it works on another level too. The best AI companies attract polymaths - physicists, engineers, storytellers. We’ve built a team united by that idea of transforming creative energy into impact.

Q: Tell us about your journey leading up to Higgsfield - what were the experiences that shaped you as a builder?

Alex: Both my parents are engineers - my mother designed rockets for the Soviet space program - so I grew up around engineering and started programming at 10.

At 20, I was ranked among the top three in the world in competitive programming.

I co-founded AI Factory, where we built neural networks that could run directly on mobile devices - something no one else had at the time. Snap acquired us, and that tech became the backbone of Snapchat’s face filters.

That experience taught me how AI, creativity, and distribution intersect. It’s not about the fanciest algorithm; it’s about delivering magic instantly, at scale. That philosophy directly informs Higgsfield.

Q: Higgsfield hit 11M users and US$50M ARR within only five months. How did you decide the product was ready - and what drove such rapid adoption?

Alex: We started in 2023. The first year was pure iteration - testing, failing, re-architecting. Our investors were patient; we kept asking one question: what truly works for users?

We began on mobile, but realized generative AI doesn’t change the economics of mobile apps - high churn, small ticket sizes, limited expansion.

The real inflection point came when we shifted to desktop and targeted creative professionals - social-media managers, content teams, ad creators. We also built over 100 cinematic templates with pre-defined camera controls.

That made videos not just realistic, but cinematic. Once that went live, traction exploded.

We also built a culture of velocity. We ship fast - sometimes four or more feature releases per week, occasionally even one per day. Whenever a new state-of-the-art model dropped, we integrated it within 24 hours. Customers learned they could trust Higgsfield to always have the latest, best-performing AI. That consistency and shipping rhythm compounded growth.

Q: The AI video and content generation space is getting crowded - from model players to application-layer startups. In such a fast-moving landscape, what’s a durable way to differentiate Higgsfield? And do you think success depends on being the loudest player?

Alex: We have to start with an acknowledgment that the video AI space is still early. If you look at more mature domains like search or coding, we already see clear examples of winners that succeeded by quickly implementing the best models into user workflows. That’s the same strategy we follow at Higgsfield, and I hope we execute it even better.

At the same time, we recognize that the economics of generative AI companies are challenging. Gross margins aren’t like traditional SaaS - they’re well below 80–90%. That’s why growth and efficient user acquisition are critical to offset this profile. Cheap acquisition only happens when you stay on top of people’s minds - when you’re one of the loudest players.

So yes, product quality comes first, but you can’t ignore GTM. Being visible matters. It’s why we’ve focused on pairing strong distribution with relentless product velocity. I’m also proud that Higgsfield is already cashflow positive, proving that it’s possible to grow fast, stay disciplined, and build a durable business even in a noisy market.

Q: Higgsfield is growing fast yet already cashflow positive. How is that possible, and how do you think about sustaining growth while managing churn?

Alex: It surprises many people, but our cashflow positivity is a function of both timing and efficiency. The number of people who need generative video tools will likely grow a hundredfold from where it is today. Organically, people are searching for ways to make videos with AI because it’s 100x cheaper and faster than physical production.

Higgsfield today has more reach across social media than all our competitors combined. That reach compounds - it strengthens our brand, draws in new users, and lowers acquisition costs. We also work closely with AI educators and creators, who act as natural distribution channels for us.

Of course, this is still an experimentation stage. Some users subscribe, cancel, and then return months later as product quality improves. Larger enterprises still lack clear policies for video AI use, so many professionals buy out of pocket.

But the pace of progress is extraordinary - we’ve made more in the last nine months than the previous nine years of AI video combined. And as quality and utility improve, churn will normalize. We’re building for a future where most videos on social media are AI-generated, and that expanding market keeps us focused and profitable.

Q: With such fast growth, how do you think about team size and structure as you scale?

Alex: First and foremost, our team is highly differentiated. We’ve built on the “996” culture - intense, fast, and deeply committed. Generative AI is creating entirely new go-to-market models and unit economics, and the only way to stay ahead is to ship constantly.

We release product updates almost every day - typically between four and seven feature releases a week, with one new release on most weekdays. This rhythm keeps us learning faster than anyone else in the space, and that’s unlikely to change.

As we mature, we’re expanding beyond engineering and product into operations, finance, accounting, legal, and sales. Our first salesperson actually started earlier today - a milestone for us as we evolve from pure product velocity toward structured growth.

Are stablecoins cheaper for cross-border payments? What are the promising use-cases for crypto? Coinbase's MD (Head of APAC), John O’Loghlen, sees crypto adoption in APAC accelerating. APAC is both a major opportunity and a tough proving ground. In this conversation, John shares his view on the region’s future - from stablecoins and trade finance to competition and what success could look like five years from now.

Q: Before we dive into crypto, could you share a bit about your journey and how you came to lead Coinbase in APAC?

John: I’m originally from New Zealand. After university in the U.S., I started my career in finance at Goldman Sachs, working through the dot-com bust and spending time in Beijing. Later, I pivoted into the restaurant industry across Asia, building Domino’s and our own ventures in China during the rise of platforms like Meituan and Dianping. That experience taught me about mobile-first consumer behavior and the importance of seamless payments - early lessons in digital adoption.

I later joined Alibaba, helping build its international business in Australia and New Zealand, scaling exports into China through platforms like T-Mall and Alipay. When that wound down, many of my old friends from finance and tech had moved into crypto. I went down the rabbit hole about four years ago, and Coinbase was expanding in Australia at the time. Their focus on compliance and transparency resonated with me, and that’s how I joined to lead APAC.

Q: Coinbase has expanded rapidly in APAC - from Australia to Singapore, Japan, and now India. How do you see the region’s complexity shaping your strategy?

John: APAC is not one homogenous region; it’s dozens of markets with vastly different levels of maturity, regulatory attitudes, and consumer behaviors. In Singapore, MAS has been proactive in building frameworks that balance innovation and investor protection, which creates fertile ground for new products. In Australia, regulators like ASIC and Austrac have become increasingly active, requiring companies to engage deeply on compliance, treasury, and custody. India, on the other hand, is a huge opportunity but also a market where we’ve had to learn humility - launching once, stepping back after regulatory challenges, and then relaunching with FIU registration under the Ministry of Finance.

Our strategy is to take the long-term view. That means building trust first: investing in compliance, setting up insurance and custody arrangements on par with global banks, and working directly with law enforcement to tackle scams and fraud. We also know customer experience must be localized. Users in APAC expect quick responses in their language and 24/7 support, which is why we’re scaling a center of excellence in Singapore. The region is complex, but if you do the hard work, it rewards you with scale and growth.

Q: Let’s talk about stablecoins. Some argue that on- and off-ramping costs reduce their benefits for cross-border payments. What’s your perspective?

John: At the moment, the costs of moving in and out of stablecoins can erode some of the savings people expect. But we’re still early. If you look at Ethereum’s evolution, fees have dropped dramatically with Layer 2 scaling solutions. Stablecoins will follow a similar trajectory as liquidity improves and adoption grows.

In APAC, remittances are one of the clearest use cases. Migrant workers sending money from Singapore to the Philippines, or from the Middle East to South Asia, face high costs through traditional channels. Early adopters are already using wallets and peer networks to reduce those costs significantly, often cutting hundreds of basis points compared to traditional fintech players. But to go mainstream, stablecoins need to be denominated in local currencies - SGD, AUD, JPY - so that they mirror the way people think about money day-to-day. That’s why Coinbase is supporting new local stablecoin projects. Once market makers and sovereign initiatives add liquidity, costs will fall further, and that’s when the real FX disruption happens.

Q: Beyond trading on crypto exchange, what use cases for crypto feel most promising in APAC? And what needs to happen for them to go mainstream?

John: Trade finance stands out. Right now, if you’re an oil trader in the Middle East trying to settle with a partner in Singapore, the transaction may take a week, involve up to 14 intermediary banks, and cost 400-500 basis points. That’s incredibly inefficient. Stablecoins can compress settlement times to minutes and reduce costs to a fraction. For industries like mining, energy, and commodities - where cross-border transactions are constant - the potential impact is massive.

But to enable that at scale, we need regulatory certainty at both ends of the transaction. If you’re sending from Singapore to Hong Kong, both sides need frameworks that give clarity to issuers and protect consumers. Encouragingly, regulators are moving. Just this week, ASIC announced relief for stablecoin providers, removing some of the uncertainty that stalls product launches. This matters because innovators need confidence they won’t be penalized for moving too quickly. With frameworks in place, investors keep funding, corporates gain confidence, and consumers know they have redress if something goes wrong. That’s how you unlock mainstream adoption.

Q: Coinbase is a U.S. company entering a region with strong local and global competitors. How do you view Coinbase’s role in APAC’s competitive landscape?

John: The first thing I’d say is respect. Local exchanges built these markets long before Coinbase arrived. They have many users and deep local roots. Our role is not to simply compete but to raise the bar - by bringing in the governance, compliance, and institutional-grade infrastructure that institutions and regulators demand.

We’ve made heavy investments in trust: crypto insurance through Lloyd’s, custodian licenses in New York that only a handful of global banks hold, and dedicated Trust & Risk teams that handle everything from compliance and litigation to law enforcement requests. These are costly, but they differentiate us. Unlike offshore exchanges, when we enter a market, we commit to being onshore - hiring experienced compliance leaders from TradFi and engaging regulators directly. That’s harder and more expensive, but it builds durable trust.

We’re also localizing our products. APAC users are far more yield-hungry than Western users, and we’re adapting to that. We’re investing in local CX and scaling a Singapore hub that will support the region broadly. The combination of global best practices with local relevance - that’s how Coinbase creates value in APAC.

Q: Finally, if we fast-forward five years, what does success look like for Coinbase in APAC?

John: For Coinbase, success means being the trusted bridge. If in five years a student in Manila can send money home instantly, a miner in Australia can settle with partners in Singapore at a fraction of today’s cost, and institutions can move billions securely on-chain - all with Coinbase as the trusted partner - that’s the vision. It’s about enabling prosperity, securely and compliantly, for millions across APAC.

Salesforce is making a major push into AI with Agentforce - and Kaylin Voss is at the helm of its go-to-market globally. From Slack to Salesforce, she has built a career at the crossroads of enterprise software and groundbreaking innovation. Today, she leads Agentforce and Data Cloud in Salesforce, tasked with shaping how AI agents and trusted data transform the enterprise.

In this conversation, Kaylin shares how Salesforce decides which AI use cases matter, how enterprises are adopting Agentforce at speed, why CRM is evolving from management to orchestration, and how the GTM playbook for AI looks very different from traditional SaaS.

Q: Kaylin, thanks for joining us. To start, could you share a bit about your role at Salesforce today?

Kaylin: I currently lead our global go-to-market for Agentforce and Data Cloud at Salesforce. Think of it as a startup within Salesforce - fully focused on data and AI. Nine months ago, this team didn’t exist. After Agentforce launched, Marc [Benioff] quickly realized we needed a dedicated go-to-market team. My role is to build and lead that.

I bring a mix of experiences to this role: CRO at Slack, where I focused on bringing consumer-grade technology into the enterprise; senior leadership at Tableau and Salesforce core. Together, that background helps me drive Salesforce’s push into AI and data with Agentforce and Data Cloud.

Q: Many companies debate whether to build AI for the sake of AI. How do you decide which use cases are truly worth pursuing?

Kaylin: My job is to identify which AI use cases truly move the needle. For companies still debating, my advice is simple: just start. Even failed experiments create valuable learning.

At Salesforce, we filter everything through “ROI-in versus agent-out.” Instead of building features because they look cool, we start by asking: what customer pain point are we solving? Contact centers that frustrate consumers, healthcare systems that create friction, financial services that are slow to resolve issues - these are the places where AI can create real impact.

Most companies begin with “efficiency AI” - reducing cycle times by 20–30%. But the more exciting shift is to “growth AI.” That’s when AI enables work that humans couldn’t cover before. At Salesforce, our digital SDR agents now nurture leads that previously went untouched. That’s entirely new revenue. Customers like ADECCO and Indeed are seeing similar results: digital labor capturing signals 24/7, driving growth instead of just efficiency.

Q: Enterprises often move cautiously with new technologies. How do you convince them to adopt AI, given the speed and risks involved?

Kaylin: Trust is the foundation. Enterprises demand compliance, governance, and security guardrails - and that’s exactly why they adopt AI with Salesforce. Their workflows, data, and applications are already embedded in our trusted platform.

We see enterprises start with assistive, lower-risk use cases: internal knowledge management, basic automation, agent copilots. Once they see value and trust the safeguards, they expand toward more autonomous use cases. And while enterprises are careful, the pace of adoption has been faster than many expect. Once they feel secure, they scale quickly.

Q: What will CRM look like five years from now in the age of AI?

Kaylin: CRM will evolve from customer relationship management to customer relationship orchestration. Instead of static databases, AI agents will be proactive partners.

In the old world, reps looked up records, typed in notes, and manually updated pipelines. In the new world, agents capture insights automatically, log data in real time, and proactively recommend next best actions - from highlighting churn risks to suggesting upsell opportunities. The interface will be conversational and multimodal, guiding users instead of burdening them with data entry. Paradoxically, people will spend more time in CRM, but in a more productive, value-adding way.

DIY solutions often fall short because they can’t provide observability or measurability at scale. Companies need to know how their human reps and AI agents are performing side by side. That’s where we see Agentforce winning: deep workflow integration plus enterprise-grade observability.

Q: You’ve led GTM across Slack, Data Cloud, and now Agentforce. How does the go-to-market motion differ for AI products?

Kaylin: It’s a fundamentally different motion. Traditional SaaS sales is linear: pre-sales, close the deal, hand off to customer success. With AI, it’s fluid and ongoing - you never “win” or “lose” a customer, because there’s always another workload or use case to unlock.

We call it the “accordion principle.” You start with a big, strategic vision, then narrow to a simple, practical use case. Once that lands, you expand again. Success begets success.

This also demands new talent. Seventy percent of my top performers today come from engineering or architecture backgrounds. They’re not just sellers - they’re co-builders. Workshops are the new demo. Instead of showing slides, we sit with customers, deploy agents into production, and refine them week by week.

And the buyer set is broader. In the past, you sold CRM to business owners and IT. Now, it’s everyone: COO, CMO, CTO, CDO, Chief Product Officer. AI touches every function, so we call it an “all-motion” go-to-market.

Philipp Kandal is the Chief Product Officer of Grab, Southeast Asia’s largest superapp and one of the region’s most iconic tech companies - serving 46M monthly transacting users. Yet as he points out, that’s still just 6% of the region’s population. In his view, Grab’s journey is only getting started.

In this conversation with Evolving Edge, Philipp shares how Grab balances hyper-local nuances with regional scale, why empathy and accessibility drive its product philosophy, and what “AI-first with heart” means in practice.

Q: Let’s start with your journey. How did you end up in Singapore and with Grab?

Philipp: I previously founded a startup in Europe focused on maps, which was later acquired by a Silicon Valley company. While on a stopover in Singapore, I met a friend working at Grab. Over a beer, he told me about the huge challenges they were facing with maps. I wasn’t looking for a job - I was actually about to start another startup - but decided to extend my stay for a day to see if I could help.

Long story short: six months later, I joined Grab, moved from San Francisco to Singapore, and never looked back. The problems here were so unique and impactful, it was impossible to resist.

Q: Grab is one of the few platforms that has truly scaled across such a fragmented region. From a product perspective, what’s the framework you use to navigate that complexity?

Philipp: We often talk about our 80–20 approach. About 80% of our platform is consistent across the region, but the remaining 20% is tailored to local nuances - from payment preferences to cultural habits.

It always comes back to our mission. We’re here to uplift people - whether that’s helping a driver earn more, enabling a small merchant’s business to grow, or giving a family better mobility options. So, our product philosophy is simple: make technology accessible and sustainable. That means building products people love, while ensuring the unit economics make sense too.

We want to take technology we think is transformative and make it more accessible because - for example - drivers in Jakarta may not have the same access to expensive smartphones as drivers in the U.S. It’s important that our products suit and are designed for our users’ practical needs and circumstances.

Generally, if a product is accessible, sticky, and economically viable, we know we have a winner. We consider this through two lenses: adoption (do people love and find value in the products we build?) and sustainability (do the unit economics work well enough for us to scale this?). We track engagement through adoption curves, repeat usage, and word-of-mouth virality, while sustainability requires that unit economics make sense for both Grab and our partners.

New features like Family Accounts - designed for Southeast Asia’s broader family structures - have scaled quickly, for example, because these conditions are being met.

Q: You’ve led product and engineering teams globally. What’s different about building in Southeast Asia versus the Valley?

Philipp: The talent here is outstanding - the bar we’ve set is no different from somewhere like Silicon Valley. We have R&D centres in 4 Southeast Asian cities and the vast majority of our tech talent are locals based region-wide, so they have lived experience and understand the region deeply. Our teams are international, so perspectives are diverse. We also have talent based in hubs like Romania, Bangalore and China, so we can recruit the best talent available, wherever they are based.

Empathy is especially critical in Southeast Asia. In contrast, in the Valley, teams often model products after themselves. That makes sense if you’re building, say, an AI coding assistant. For us, it’s crucial that we immerse ourselves in what our users experience, in order to build products that really serve their needs. It’s why, when I visited Vietnam a few weeks ago, my team and I spent hours with drivers to understand how our products can serve them better.

We spend a lot of time on the ground - riding with drivers, talking to merchants - to really understand their challenges. That level of immersion is a core skill for product builders in this region.

Q: Grab talks about being “AI-first with heart.” What does that mean in practice?

Philipp: “AI-first” reflects our deep belief that AI is hugely transformative. With every product we build, we must consider how it fits within an AI-powered world. We even run AI sprints internally so that teams can focus solely on experimenting with and exploring AI use cases for a dedicated period, like we did for a few weeks last month.

The “with heart” part is just as important. It means solving real user problems and making AI accessible to people who otherwise might never be able to use advanced technology.

Two examples I love, which solve real pain points for our merchant- and driver-partners:

• Merchant AI Assistant: A “virtual COO” for small businesses on our platform. It analyzes the merchant data on our platform, from sales data to customer reviews and gives merchants tailored advice, like which menu items to add or promos to run. It offers advice based on contextual information from data within our ecosystem, as opposed to generic advice.

• Driver Voice Reporting: Drivers in Southeast Asia are very community-oriented. They like to share information with each other about road conditions. So, we developed a Voice Reporting which allows drivers to share road updates about things like flooding or road blockages. This has massively increased feedback sharing, improving traffic predictions and ETAs across the platform, improving the experience for all our drivers.

Q: Beyond AI, how do you approach frontier technologies like autonomous vehicles that require ecosystem readiness?

Philipp: We’re very excited about AVs, but our guiding question is always: how do we make sure technology is accessible, safe and inclusive in Southeast Asia?

First, we think about what obstacles must be addressed in order to bring new technology to customers. That’s why we’ve signed partnerships with players including WeRide to look closely at what things like Southeast Asia’s traffic conditions, regulations, and infrastructure could mean for a more autonomous future. Ensuring driver and passenger safety will be paramount, as well as how we can support drivers in a rapidly changing environment.

I always think that constraints force us to be more creative. Take how we set out to map Southeast Asia, for example. The expensive cost of mapping rigs was holding us back from scaling up our mapping efforts. So, we built our own low-cost mapping hardware which we distributed to our drivers, who could help us map the region while on the road. This let us map Southeast Asia at a fraction of the cost, turning unit economic constraints into an opportunity.

Ultimately, whether it’s AVs, AI, or maps, our goal is the same: improve as many lives as possible by making technology usable, affordable, and human-centered. We never develop technology just for its own sake.

With more than US$300M in annualized revenue and a valuation of US$7.5B, Gong is the leading revenue AI platform globally. Co-Founder & CPO, Eilon Reshef, has scaled Gong into one of the most impactful SaaS businesses, redefining how sales teams understand and act on customer interactions.

In this conversation, Eilon reflects on Gong’s journey - from the founding insight to building early product-market fit, from shaping the future of AI in revenue operations to how he thinks about hiring in a world where technology changes week by week.

Q: Let’s start with your journey. How did Gong come about?

Eilon: I’d just come off a sabbatical after selling my previous company when I met Amit Bendov, my co-founder at Gong. Amit had just gone through a poor sales quarter at his previous company and realized something fundamental: CRM systems weren’t telling him why things were happening. Manually entered CRM data - often subjective and incomplete - rarely reflects the true story of what’s happening in a deal, leaving leaders blind to what’s actually going wrong in customer interactions.

The real insight was that the answers were hidden in conversations - calls, emails, chats, even Slack and WhatsApp messages. The problem was that listening manually wasn’t scalable. I was already exploring machine learning in 2015 - before “AI” became the buzzword it is today. The question we asked was: what if you could run a revenue organization based on meaningful data - actual conversations, emails and other engagements - analyzed intelligently? That was the genesis of Gong.

We imagined a layer of intelligence that could automatically capture, analyze, and interpret every customer interaction, and then surface insights that help companies improve forecasting, coach their teams, and manage pipelines more effectively. Over the last 10 years, that vision has grown into Gong as you see it today: a high-growth company tackling critical revenue challenges, one use case at a time.

Q: Gong is often cited as a masterclass in product-market fit. What were the signals you had found it?

Eilon: At the time, it wasn’t obvious that we had achieved product-market fit. When we launched in 2016, we worked with 12 design partners - companies willing to use our very early software and give us candid feedback. After a few months, we went back to them and said: “We’re officially launching, and we’re going to start charging.” Out of the 12, 11 converted into paying customers.

We were actually disappointed - why not all 12? In hindsight, that was an incredible signal. Typically, you’d expect maybe 20–40% conversion. Achieving 90% showed we were solving something urgent. The 12th company also became a customer later, and still is a customer today!

Another strong signal was customer complaints. When customers use your product and care enough to complain - “Why didn’t this call record?” “Why didn’t these action items show up?” - it means the product has become critical to them. Indifference is a far worse sign. Looking back, conversion from pilots to paying customers, coupled with high engagement and feedback intensity, were clear markers that we had real traction.

Q: Many founders worry about overfitting to design partners. How did you avoid that trap?

Eilon: It’s an art, not a science. The key is to work with multiple design partners rather than just one or two. That way, you can identify which requests are generalizable. If a customer says, “Please integrate into my unique internal system,” that’s not broadly applicable. But if they ask for dashboards or reporting, that’s likely valuable for many.

There’s a common fear of building something too custom. But in my view, more companies fail because they build generic solutions that no one truly needs, rather than because they built something too specific. I’d rather make one customer wildly successful than make 100 customers lukewarm. That success often creates the foundation for broader adoption.

Q: Gong helped define “Revenue AI”. Where do you see the category going over the next five years, especially with AI agents emerging?

Eilon: We believe every profession will eventually have an AI-powered system tailored to it. For revenue organizations, Gong aims to be that system: an end-to-end, AI-first platform that becomes the single source of truth and continuously optimizes how companies run their go-to-market operations.

This doesn’t mean traditional software goes away. Instead, we’ll see a hybrid: some tasks done through dashboards and reports, others through conversational AI. The promise of AI lies in two dimensions: efficiency and effectiveness.

• Efficiency: Automating manual, time-consuming work. Instead of spending two hours writing an email, you’ll spend two minutes with AI drafting it for you. Instead of filling CRM data manually, it will be auto-populated.

• Effectiveness: Helping people get better at their craft. For example, we recently launched an AI-powered training tool. It doesn’t save time - it actually requires reps to invest more time - but it makes them better. That’s something you can’t achieve otherwise.

Over the next five years, I expect “AI for revenue” to become indispensable, just as GitHub Copilot is becoming essential for developers. Gong wants to lead that transformation for go-to-market teams.

Q: Many companies are rushing to add AI features. How do you avoid building “AI for AI’s sake”?

Eilon: Interestingly, sometimes building AI “for show” isn’t a bad thing. If your company wasn’t born AI-first, you still need to get your teams - engineers, PMs, designers - comfortable with the technology. Launching a handful of AI features, even if they’re redundant, creates feedback loops. Customers give input, and teams gain firsthand experience. That sets the stage for more systematic, high-value AI.

At Gong, iteration has been crucial. For example, we launched account briefs that summarize all customer conversations into a single snapshot. Customers loved it - and then asked us to also pull in public company data from the web. We hadn’t thought of that initially, but version one sparked new ideas. You can’t get to the breakthrough features without shipping early versions. It’s about creating momentum and avoiding stagnation in a “local maximum.”

In other words: start fast, iterate relentlessly, and let both customers and your own team guide where to double down.

Q: Finally, how do you think about hiring in the AI age?

Eilon: I think about two profiles. First, you have executors - PMs and engineers who can absorb what’s happening in the field and execute features quickly. That’s relatively straightforward. They need technical skills and the ability to stay up to date, but the role resembles traditional product and engineering positions.

The harder profile is the innovator - people who are hyper-curious and stay at the cutting edge. These are the ones who know what’s happening in AI every week, who read the newsletters, experiment with new tools, and bring fresh ideas to the table. They combine curiosity with enough technical understanding to separate hype from real opportunity. In conceptual or leadership roles, that curiosity is the most important trait. Without it, you stop growing. And in AI, if you stop growing, you fall behind quickly.

The pace is relentless. One day Microsoft releases a new text-to-speech model. The next day it’s a new agent framework. None of these are singular breakthroughs, but if you don’t follow them, you fall out of the loop within months. Hiring people who thrive in that environment is critical.

Fresh off a US$200M Series E at US$4.5B valuation, Verkada is a leader in cloud-based physical security systems globally. VP & MD (JAPAC), Jonathon Dixon, spent decades leading cloud-native tech firms, and now brings that experience to one of the most traditional industries of all: physical security.

In this conversation, Jonathon shares how Verkada is disrupting incumbents, scaling across Asia Pacific, and unlocking new possibilities with AI.

Q: How did you come to join Verkada, and for readers less familiar, what does the company actually do?

Jonathon: I’ve had some great mentors in my career and a set of guiding principles that I’ve carried with me. Verkada ticked all those boxes: an incredible product, a disruptive vision, and a strong track record since inception. The company is almost nine years old now, and made up of great people. I joined ten months ago to lead our go-to-market and operations across Japan and Asia Pacific.

Verkada is a disruptor in physical security. We bring together six product lines - cameras, access control, intercom, sensors, workplace, and alarms - on a simple, cloud-native platform. Traditionally, physical security has been fragmented and stuck on-premise. Our integrated solutions and open API architecture change that. Today, we have 33,000 customers globally, including 96 of the Fortune 500.

Q: Expansion in APAC is never one-size-fits-all. How do you prioritize which markets to enter and where to double down?

Jonathon: This is my third APAC leadership role with cloud-native companies, so I’ve seen both successes and lessons learned. When prioritizing markets, we look at economic growth, multinational presence, and patterns of adoption - our sweet spot is organizations with distributed sites where a cloud-native platform can scale and modernize quickly.

Winning iconic brands is also key - they validate the model and wake up markets. In APAC, our current focus areas are Australia, New Zealand, Japan, Korea, and Singapore. But we’re also seeing strong signals in emerging markets like the Philippines, Thailand, and India, where multinational demand is now matched by growing local interest. Importantly, we take a country-first approach, tailoring strategy to each market rather than forcing a single playbook across the region.

Q: Security is often hyper-local. How do you balance global standards with local needs?

Jonathon: It’s a challenge, but one we’ve built for from the start. Our platform is highly configurable - you can turn features on and off to align with local regulations and operational requirements. Because it’s open and API-driven, customers can integrate with other systems or build new solutions for their specific needs. Even if something isn’t on the roadmap today, our product team moves fast when we see repeatable value across the region. Customer feedback is one of Verkada’s core tenets, right up to our CEO.

Q: This is a sector long dominated by legacy vendors. How does Verkada compete against incumbents with entrenched partnerships?

Jonathon: We’re disrupting the model by bringing a cloud-first approach. Everything we do goes 100% through the channel, which is critical in APAC given the local languages, currencies, and partner dynamics. Our products are designed to be both powerful and simple - integrated out of the box, easy for partners to install, and intuitive for end users.

We’re also attracting modern SIs - companies like NTT, Mitsubishi, LG CNS - who have cloud in their DNA. They see Verkada as part of a broader modernization journey for their enterprise customers. By making the partner experience as strong as the customer experience, we’ve lowered the barriers to adoption.

Q: Do you face resistance from enterprises that are traditionally conservative in adopting new solutions?

Jonathon: In fact, we’re seeing momentum around network modernization. If a company has a cloud-first or modernization policy, Verkada becomes a natural fit - we’re simply another endpoint on the network. That’s helped us work with both IT-focused integrators and traditional physical security providers. When customers see that installation is fast, management is simple, and the platform scales easily, adoption follows quickly.

Q: Everyone is talking about AI. How do you see it changing physical security in the near and mid-term?

Jonathon: AI’s biggest impact today is speed. It allows teams to respond in real time rather than hours - whether that’s initiating a lockdown or pulling up footage of an incident. That shift alone dramatically improves security outcomes. But it’s not just reactive; AI enables proactive security. Real-time alerts can flag prohibited individuals, identify specific vehicles, or notify teams when a room nears capacity.

Beyond protection, AI is turning security systems into sources of operational intelligence. For example, Mondial VGL, a logistics provider in Australia and New Zealand, uses our AI analytics for license plate automation, safety alerts, and yard optimization. Trucks entering their Perth facility are automatically identified, gates open without driver intervention, and health-and-safety compliance checks that used to take hours now take minutes. We’re only scratching the surface of what AI can do here.

Q: Looking ahead, what excites you most about building Verkada in JAPAC?

Jonathon: JPAC is one of Verkada’s fastest-growing regions - we grew more than 95% year-over-year in the first half alone. We’re attracting some of the best talent in the industry, building strong partnerships, and helping customers modernize and scale quickly. For me, it’s been 10 months of fun so far, and I’m excited about the road ahead.

Galbot reached US$1B valuation just two years after its founding, and is now one of the leading humanoid robotics companies globally. Chief Strategy Officer, Yuli Zhao, focuses on building humanoid robots deployed in real-world scenarios - from pharmacies and malls to industrial lines.

Galbot’s story is about taking robotics out of the lab and into daily life, and this conversation explores how the company is commercializing humanoids at scale, why retail was chosen first, and how data - not hardware alone - will define defensibility in this new era.

Q: You started your career in consulting and gaming. How did that journey lead you to humanoid robotics?

Yuli: I began at McKinsey almost 20 years ago, where I learned the discipline of problem-solving and long-term strategy design. But there’s a saying in consulting: you are often the “kingmaker,” not the king. I wanted to be directly responsible for execution. That led me to Qihoo 360, where I founded the strategy department and worked closely with the CEO to develop monetization plans for a user base of over 600 million. Out of that work came a bold idea: building a gaming business. Despite initial skepticism, we launched Qihoo’s mobile gaming unit and pioneered publishing models that generated huge growth.

Later, at Yoozoo Interactive, I scaled their corporate development and expanded globally — opening teams in Japan, Europe, and China. I saw firsthand how marketing, branding, and user engagement could make or break new products. This shaped my belief that technology alone isn’t enough. For robots to enter daily life, they must deliver clear value, be easy to adopt, and connect with end-users in meaningful ways. That blend of strategy and execution is what I bring now to Galbot.

Q: Galbot has made headlines for opening humanoid-run pharmacies in Beijing. How are you approaching commercialization?

Yuli: Commercialization has to start with solving real problems. In Beijing, we now operate over 10 intelligent pharmacies that run entirely on our humanoid Gilbert G1. These stores are open 24/7, with uptime above 97%, and require minimal human oversight - one technician can remotely monitor three stores simultaneously. For consumers, this means they can get urgent medicines overnight without waiting for a store to open or depending on overworked staff. For store owners, it means lower labor costs and a sustainable 24-hour business model.

We are expanding beyond pharmacies into malls in South China, with plans to reach 100 stores across 10 cities this year. Each deployment gives us not just revenue but also invaluable operational data. It’s proof that humanoids are not just for demos - they can deliver ROI in everyday commercial life.

Q: Why start with retail, rather than industrial or logistics use cases?

Yuli: We looked at two things: demand and foundational skills. Demand-wise, retail - especially pharmacies - is a sector where human limitations are clear. Running overnight shifts in tiny, crowded spaces is costly and unattractive for employees. A humanoid can operate continuously in spaces with 5,000–6,000 SKUs packed into 30 square meters, where it would be overwhelming for a person.

On skills, grasping is fundamental. Around 80% of useful work in the real world comes down to picking up and placing objects. By focusing on retail, we built and trained our first foundation model, GraspVLA, which uses billions of simulated and real-world datasets. This gave us a platform capability - grasping, sorting, stocking - that can later extend into logistics and industrial use cases. Retail is not just a business choice, it’s a proving ground for humanoid fundamentals.

It’s also important to emphasize that we are not limited to retail. Galbot has established deep collaborations with approximately two-thirds of the leading automotive enterprises in China. These partnerships cover SPS sorting, automated quality inspection, and tote transportation. In this sense, retail was only part of our proving ground, with automotive and broader industrial applications are already a major part of our commercialization journey.

Q: Hardware often gets commoditized over time. How do you think about defensibility for humanoids?

Yuli: Our moat will not come from the physical form of the robot alone. Defensibility comes from data and embodied intelligence. Unlike language models, which can train on billions of free internet tokens, humanoid AI requires 3D spatial and trajectory data. That data is expensive, slow to collect, and critical for performance.

If you hire 40 operators to telecontrol robots for a month, you may only gather 100,000 data samples - far short of the 10 billion needed for robust intelligence. Galbot solved this by building a proprietary simulator that models thousands of objects, human-like movements, and interactions. We have already accumulated over 10 billion synthetic datasets, which we then fine-tune with real-world deployment data from pharmacies and manufacturing sites. This hybrid approach gives us scale and accuracy that few others can replicate, creating a durable moat.

In addition, we have formed a strategic joint venture with Bosch Group and are industrializing embodied AI within SPS production lines, further strengthening our defensibility and accelerating real-world adoption.

Q: Many robotics companies build purpose-specific machines. Why did Galbot focus on humanoids?

Yuli: Purpose-built robots dominate today’s factories, but they are rigid and costly to update. If a product line changes, companies must rewrite code and reconfigure machines - a huge reinvestment. Humanoids, by contrast, are flexible. Updating the AI model is far simpler than redesigning hardware. That flexibility is the future of manufacturing.

There’s also a practical reason: workspaces are designed for humans. A humanoid fits into the same environment without costly retrofitting. Our Gilbert G1 is 1.8 meters tall, with arms that extend to 2.4 meters, allowing it to operate in human-designed environments. It runs for 8 hours per charge, compared to many humanoids that last only one hour before overheating. We intentionally designed it with wheels instead of legs for stability and endurance, but we are also developing a dual-leg model for the longer-term vision of AGI-level robots.

In short, humanoids offer immediate practicality and long-term adaptability. That’s why Galbot is investing here rather than in single-purpose machines.

With 2M+ developers on its platform and a rumored US$6B valuation, Groq is emerging as a key Nvidia rival. Groq is building AI-native chips with ultra-low latency architecture for inference, enabling new real-time LLMs, embedded AI agents, and latency-sensitive workflows.

GM (APAC), Scott Albin, is at the center of Groq's global ambition.

In this chapter of Evolving Edge, Scott shares why the future of AI in APAC will be driven by inference, how Groq’s cost and performance advantages are winning over early AI adopters, and what makes their go-to-market so unique.

Q: Let’s start with your journey - how did you end up at Groq?

Scott: My career has always been in software and services. First in large-scale enterprise IT systems, then moving into analytics, machine learning, and now AI. The core principles are similar, even if the labels have changed.

I came to Groq initially as an investor in the Series D round. In discussions with the team, I learned Groq already had a meaningful base of early adopters in Asia. Having spent nearly nine years in Singapore and four in Australia, I understood the regional landscape and began suggesting how to capture that demand. That evolved into leading the APAC business from January this year.

Q: For readers who may not know - what exactly does Groq do that’s so different?

Scott: Most AI infrastructure today runs on GPUs (especially NVIDIA’s), which were never purpose-built for AI but adapted well over the past decade. Groq took a clean-sheet approach: Design a chip and stack from the ground up for inference.

The reality is most enterprises don’t train models - they consume them. And in APAC, the demand is overwhelmingly for inference. Our technology delivers significantly faster throughput, lower cost per token, and much greater energy efficiency than GPUs. Those three advantages combined are extremely compelling.

Q: How do you structure go-to-market in such a diverse and regulated region, and how do you win over cautious enterprise buyers?

Scott: We are building a broad GTM capability: direct sales, channels, OEM partners, cloud alliances, and resellers, each with its own strategy, targets, and integration points. In APAC, beyond US and China hyperscalers, you have neo-clouds, telcos, technology platforms, and others who want to embed inference into their offerings to differentiate or create new revenue streams. Each motion is tuned to local market conditions, regulatory requirements, buyer maturity, and ecosystem dynamics, often involving co-selling, proof-of-concepts, and deep performance benchmarking versus the incumbents.

Developers have driven early adoption (over 2 million have signed up for GroqCloud), but for enterprise CTOs/CIOs, the conversation quickly shifts to total cost of ownership, performance benchmarks, compliance obligations, and data sovereignty.

Once an AI product moves from MVP to millions of users, latency and cost are not just important - they’re mission-critical. For example, we can offer Kimi K2 - a trillion-parameter open-source model - at US$3 per million tokens vs. US$75 for Claude Opus, delivering a 20x cost advantage at similar quality. As the region’s experimentation phase gives way to scaled deployments, decision-making is becoming more pragmatic and ROI-driven, and that’s where we consistently win.

Q: How does GTM differ between APAC, the U.S., and the Middle East?

Scott: Use cases are consistent globally, but execution varies significantly. In APAC, language diversity, cultural nuance, and infrastructure preferences shape adoption. Global models often miss local context - whether that’s colloquial phrasing, domain-specific terminology, or cultural references - creating demand for localized models, fine-tuning, and infrastructure with in-region data residency and sovereignty.

On-prem remains common in Asia due to compliance requirements, data control concerns, and integration with legacy systems. Our GTM in the region will need to reflect these realities, from different deployment options and language support to security certifications, procurement processes, and compliance posture tailored for each market.

Q: Will localized LLMs emerge outside China in APAC?

Scott: They already have. Singapore’s SEA-LION model, Malaysia’s ILMU model, and ongoing initiatives in Japan and Korea are clear proof. Some of these efforts involve building foundational models from scratch, while others focus on fine-tuning open-source or existing models for local needs. AI capability - spanning talent, model development, and supporting infrastructure - is now regarded at the same level as national security. Governments are investing heavily in these areas to avoid dependence on external model access, ensuring they have the skills, data pipelines, and compute resources to sustain their own AI ecosystems.

Hong Kong Science and Technology Parks (HKSTP) is one of APAC’s most prominent tech ecosystems - home to thousands of startups, deep tech researchers, and venture-backed scaleups.

Albert Wong was CEO of HKSTP for almost a decade, before retiring recently. A GE veteran with roots in engineering, M&A, and global operations, Albert brought a commercial edge to what was once a government real estate operation - and became a key architect behind HKSTP’s transformation.

In this Evolving Edge conversation, Albert reflects on the mindset shifts, execution pivots, and platform bets that helped redefine Hong Kong’s innovation identity - and why he believes the most important breakthroughs are still ahead.

Q: For those who may not know your background, can you share a bit about your journey, from engineering and GE to leading HKSTP?

Albert: I was born and educated in Hong Kong and spent most of my career with multinational companies, particularly GE. I held roles across M&A, commercial leadership, and operations. Before joining HKSTP, I was CEO of GE Oil & Gas in China. When a headhunter first approached me about HKSTP, I laughed and said, "You must be joking - Hong Kong doing tech innovation?" But I took the leap in 2016. It turned out to be one of the most fulfilling chapters of my career.

Q: You’ve led HKSTP through a transformative chapter in the past decade. What are you most proud of?

Albert: We made two key changes: one external, one internal. Externally, we turned a real estate project into a thriving innovation and technology ecosystem, now supporting 25,000 people. By the time I left, our venture capital fund reached US$300M, with investments in over 100 companies in deep tech, AI, healthtech, and advanced manufacturing. Every dollar we invested drew 15 times that in external funds, shifting our role from landlords to key ecosystem investors.

Internally, we introduced variable pay to spark an entrepreneurial mindset. By tying compensation to results, we encouraged initiative and innovation, moving away from a fixed, employee-focused perspective. Though initially debated, this shift moved us from routine tasks to measurable outcomes, fostering a proactive, results-driven culture like that of entrepreneurs focused on value and growth.

Q: Looking back on the last decade, how would you describe the evolution of Hong Kong’s tech ecosystem?

Albert: The biggest shift is mindset. Back in the '90s, we used to joke - low tech, low risk; high tech, high risk. That sentiment persisted when I joined in 2016. People were still skeptical. But today, most of society understands: technology is not optional - it’s existential.

Government funding has improved, the startup ecosystem has matured, and most importantly, we’re finally building belief. We’ve gone from why tech? to how fast can we scale? That’s a big change.

The I&T sector is establishing itself as a cornerstone of Hong Kong’s economy, with strong ties to Mainland China’s tech ecosystem. We’re upgrading to rival global hubs like Silicon Valley. Recent startup competitions show our startups have world-class potential. That’s why we have launced the Booster Programmes, offering intensive training, pitching, and business-matching to help them succeed globally.

Q: Hong Kong talent has always been strong, but is risk appetite still a blocker?

Albert: I don’t think Hong Kong people are uniquely risk-averse. Everyone should be. The real issue was: for a long time, it was too easy to make money here through traditional means like property.

What we lacked were visible tech success stories - founders and engineers young people could aspire to. But now we’re seeing more of that. And I always remind people: you don’t need to be a founder to be in tech. If you’re passionate about AI, robotics, or genomics, dive in fully. Research and development offers a deeply fulfilling path, where you can tackle cutting-edge challenges, advancing AI for societal good, designing robots to revolutionize industries, or unlocking genomic breakthroughs for better health. By immersing yourself, you’re not just innovating; you’re shaping a brighter future for humanity.

Q: HKSTP has supported thousands of companies - including multiple unicorns. What separates the breakout successes?

Albert: People love to say: "Just be persistent." Honestly? That advice can be dangerous. I’ve seen too many founders burn a decade chasing a product that wasn’t working.

The reality is: 10 out of 10 startups will need to pivot. If you can’t pivot, you’re finished.

The best founders combine deep conviction with the willingness to adapt. They believe in something others don’t - but when the signals shift, they adjust. That combination of stubborn belief and strategic flexibility is rare, but essential.

And of course, execution matters. Startups have to juggle everything: manufacturing, tech, legal, hiring, fundraising. Those who can manage that chaos while staying focused? They have a real shot.

Q: What's next for Hong Kong innovation? Which sectors do you think will define Hong Kong’s next wave?

Albert: AI, healthtech, genomics, and Web3 - they’ll all play a role. But more importantly, they’re converging. Blockchain meets AI. Genomics meets cloud. These intersections will spark the most innovation.

That’s why I’ve stopped trying to predict the next big thing. Three years ago everyone was obsessed with the metaverse and NFTs. Now, that hype has cooled, but the foundational tech still matters. Trends change. Platforms endure.

At HKSTP, we focused on building the platform - so we’re ready when the next thing hits. As someone from Google once said: "We don’t predict the future. We build the platform so we can catch it."

Q: On retirement, what’s next for you?

Albert: It’s only been two weeks. I’m traveling, spending time with family, and taking it easy. I’ve spent the last nine years running full-speed - so I’m giving myself space to decompress. That said, I’ve already had a few conversations about joining boards, advising funds, and continuing to support the ecosystem in new ways.

I’m also staying in touch with some of the startups I worked closely with at HKSTP - not in any official role, but just to offer guidance. I don’t think I’ll ever fully “retire” in the traditional sense. I’ll always want to be close to innovation, and help the next generation of builders wherever I can.

Having raised US$1B+ and with current ARR at US$300M+, Snyk is a global leader in DevSecOps.

Chief Innovation Officer & Chief Marketing Officer, Manoj Nair, discusses how Snyk is embedding security into AI-coding. He explains how the company balances enterprise trust with innovation speed, and where he sees the developer security stack heading in the next five years.

Q: Let’s start with your role — what do you focus on at Snyk?

Manoj: I run Snyk’s innovation group and marketing team. On the innovation side, think of it like an internal VC incubator - building next-gen products, sometimes from scratch, sometimes through acquisitions, often co-built with strategic partners like Google, Atlassian, and ServiceNow. We also manage our most strategic design partner customers directly.

On the marketing side, I’ve been leading the transformation of Snyk’s voice in the market - and using AI not just in what we say, but in how we run marketing itself.

I’m an engineer by training, an “accidental” product manager, and a former entrepreneur. After selling my startup, I joined Snyk because it felt like a generational company opportunity - one my entire career had prepared me for.

Q: AI coding tools like Cursor, Replit, and GitHub Copilot are everywhere. How does Snyk fit into this world?

Manoj: Snyk was founded on a simple premise: give developers the tools and context to write secure code from the start, rather than relying solely on detection and prevention after deployment. This philosophy helped propel the “shift left” movement - if you’re a fan of it, you can thank us; if not, you can blame us.

Our early use of AI focused on supercharging DevSecOps. We acquired a spin-out from ETH Zurich, one of the world’s top machine learning research hubs, that had built a hybrid AI engine capable of scanning millions of lines of code in under a minute. Legacy tools often took six hours or more to process similar workloads. That leap in speed and accuracy enabled us to integrate security checks seamlessly into live developer workflows, reducing friction and increasing adoption.

When AI coding assistants like GitHub Copilot began gaining traction, enterprise customers wanted to know if Snyk could operate in parallel with these tools, scanning code as it was generated. This evolved into full integrations (e.g. with Gemini Code Assist) where Snyk’s security layer evaluates code before it even reaches the developer, preventing insecure patterns from ever entering the repository.

Today, our aim is what we call “secure at inception.” With estimates suggesting that 50%–95% of all code could be AI-generated in the near future, embedding security context at the moment of creation is no longer optional - it’s essential.

Q: What’s been the biggest challenge in building for this AI-assisted development world?

Manoj: Speed is critical. AI is evolving so quickly that keeping pace isn’t enough. We need to stay ahead. That’s why we’ve invested heavily in talent, hiring and acquiring top AI security researchers, such as those from our Invariant Labs acquisition. These PhDs have uncovered vulnerabilities in major platforms like GitHub’s MCP server and contributed tools to the open-source security community.

We pair this deep research capability with agile engineering squads that can quickly turn findings into real-world protections. These teams work on everything from scanning MCP servers for malicious components, to building detection for AI model and agent risks, to hardening security in an “AI-native” development lifecycle where code generation, design, and inference are tightly integrated.

Q: How do you balance innovation speed with enterprise requirements like compliance and trust?

Manoj: Many large enterprises want to move fast, but with trusted vendors. Our platform advantage is that we’re already deployed in thousands of environments, so customers don’t have to re-integrate identity, security scanning, or compliance frameworks.

We use feature flags and public “labs” experiments so customers can try new AI security capabilities in controlled environments. They get the speed of innovation without losing the compliance and governance they need.

Q: Looking ahead 3–5 years, how do you see the developer security stack evolving with AI agents embedded in workflows?

Manoj: I think there will be four pillars:

1. Build-time AI security - Guardrails around all usage of models and agents, independent of the model providers. You don’t want the fox guarding the henhouse.

2. Identity & authorization - Evolving to handle fine-grained, ephemeral permissions for AI agents as well as humans.

3. Runtime “safety net” - A consolidation of CNAP, EDR, NDR, and XDR into a unified real-time safety fabric.

4. Governance, risk, and compliance - A horizontal layer ensuring everything meets security and regulatory requirements.

As Marc Andreessen once said, every company became a software company. Now, every company will be an AI company — and AI is eating software. Securing that shift will be one of the most important challenges of our time.

With over 240M MAU and ARR growing to US$3.3B, Canva has become one of the most iconic design platforms globally. Chief Customer Officer, Rob Giglio, is scaling Canva's GTM machine by layering an enterprise engine onto its PLG foundation - all while embracing AI to further democratize design.

In this conversation, Rob unpacks how that transformation is unfolding - from his path to Canva to the inner workings of feedback loops, PLG evolution, and embedding AI into the creative experience.

Q: You didn’t join Canva immediately after meeting Cliff, Co-Founder of Canva. What happened?

Rob: I first met Cliff in late 2020. We had a great chat about go-to-market challenges in selling to businesses.

We spoke again briefly in 2021. Then in 2023, I got a text from an unknown number: "Hey, it’s Cliff. Can we chat?" I had to double-check who it was!

At the time, I was loving my role at HubSpot. But Cliff framed the opportunity in a way that was compelling not because of financials, but because of mission. Canva was at an inflection point. They needed someone who had seen what great looked like, but who wasn’t wedded to repeating the same old playbook.

He wanted a partner to co-create a new GTM engine for taking Canva into the workplace. I said yes and I am so glad I did because working with him directly and the great folks we have at Canva has been incredible.

Q: Let’s talk customer feedback. What does an effective loop look like between sales and product?

Rob: It starts with a mental model. At Canva, we use something called "The Bowtie", named for its shape - wide at both ends and narrow in the middle. It reflects the journey from broad top-of-funnel awareness, narrowing through conversion, and expanding again through customer success and advocacy. The Bowtie maps the full customer journey:

Each phase has measurable signals. For example, are we creating awareness? Are users adopting quickly? If we see drop-offs, we investigate what’s creating friction.

Second, there’s full commitment across GTM and product to improve the journey. It’s not just sales shouting at product. Everyone's responsible for unblocking the flow.

We quantify this in two ways:

1. Closing the Loop system: Every year, Canva receives millions of user suggestions. We route them into a structured system, tag them by priority, and allocate engineering capacity to respond. When a loop is closed, we notify the user directly. It blows people’s minds.

2. Voice of Sales via third-party tooling: Sales calls surface objections or missing features. "I'd buy if you supported this integration." We log the frequency and revenue potential. That goes straight to product.

This eliminates conflict. It’s not subjective. It’s measurable. And product teams are actively listening - especially as we expand into enterprise where needs are more complex.

Q: Canva started as a PLG rocket ship. How has that evolved?

Rob: The core of Canva's growth is still organic. But it's not PLG in the typical "try-and-buy" sense.

Most PLG companies target a persona, get them to try the product, and convert them. Canva is more network-driven:

• Anyone can use it.

• They invite collaborators.

• Usage spreads naturally inside teams and organizations.

Collaboration fuels growth. That’s the multiplier.

What’s changed is we now layer a sales motion on top. When a company has 500 or 1,000 users, someone from procurement eventually calls us: "What is Canva, and why are we paying for so many licenses?"

Sometimes, we proactively identify large usage clusters and reach out to consolidate into enterprise deals. It’s very similar to what Dropbox or DocuSign did.

So PLG is still our base motion. But now we have:

• Inbound sales triggered by user footprint

• Proactive outbound based on usage signals

• Account management for large teams and departments

Q: Everyone's talking about AI. What does it mean for Canva?

Rob: If Canva was known for democratizing design, our next act is democratizing AI.

That doesn’t mean building the most powerful model. It means embedding AI in everyday workflows so users don’t have to hop between tools.

Take image generation, for example. Most AI tools today require you to:

• Prompt an image

• Export it

• Edit elsewhere

• Upload it to another tool

• Then publish it

That’s digital hopscotch.

What we’re doing at Canva is collapsing that entire chain. We embed AI inside our platform:

• Easy to use

• Context-aware

• Publish-ready

We already have the distribution: 240M monthly users. It’s easier to bring AI to them than ask them to leave Canva and come back.

Q: Sounds like distribution > models?

Rob: 100%. Models are commoditizing. The real battleground is UX and distribution. Whoever embeds AI seamlessly into user flows will win. It's not about building the flashiest model - it's about removing friction for users. Embedding AI directly into familiar workflows allows users to stay in one tool, get more done, and realize value faster. That’s where Canva focuses: making AI invisible, intuitive, and native to the design process.

Q: You shared a metaphor about shampoo…what can consumer packaged goods teach us about GTM strategy?

Rob: Let me give you a metaphor.

I spent a decade in consumer packaged goods, selling literal commodities. Razor-thin margins. So how do you compete?

1. Brand

2. Distribution expansion

That’s where "2-in-1 shampoo" came from. You take shampoo, add conditioner, and boom - you’ve just solved two problems in one bottle.

That’s what Canva is doing. We already have the platform (shampoo). Now we’re adding AI (conditioner). We’re not forcing people to find another tool. We’re solving the full use case in one experience.

Pony AI is the largest US IPO of autonomous vehicle in 2024. The company’s CEO & Founder, Jun (James) Peng unpacks what it takes to commercialize autonomous vehicles at scale - and how Pony.ai is expanding to global markets.

Q: Pony.ai has seen tremendous success in China. As you explore global expansion, how do you assess "autonomy readiness" in each market? Given the diversity in infrastructure, regulatory maturity, and mobility patterns, how does your go-to-market strategy adapt across geographies?

Jun: We look at a few key things when assessing a market's readiness: regulatory support, infrastructure maturity, and the mobility needs people actually have there. Places like South Korea, the UAE (Dubai), and Luxembourg really stand out because they've made significant progress in several areas: permits for L4 testing; a robust economy and high mobility demand; and smart city visions with future-tech adoption. This makes them great places for pilot programs.

Dubai's a perfect example of a good fit. They have strong infrastructure and a clear plan for building a sustainable public transit that aligns well with what we do. Plus, their "Smart City 2030" goal targeting 25% autonomous transport gives us a concrete path forward: we're planning supervised trials starting in 2025, aiming for smooth integration into their transport system long-term. Luxembourg is similar – their focus on sustainable cities and proactive regulators create excellent conditions for pilots.

Our core tech, the "AI driver," is built to scale globally right from the start. It adapts really well to different environments – like switching to left-side driving in Singapore or handling harsh weather – without needing to be fundamentally rebuilt. This wasn't just theory; it was proven across China's Tier-1 cities with diverse weather conditions. Think Beijing's freezing winters and heavy snowfall, or the intense rainstorms you get in Guangzhou and Shenzhen – our system learned to navigate those successfully. This flexibility is key to our efficient rollout.

So, our go-to-market strategy is all about starting small and proving things step-by-step. We begin with supervised trials in defined areas, like we're planning with Dubai's RTA in 2025. As we demonstrate success, we gradually expand the service area and eventually move to fully driverless operations where people pay fares. This phased approach worked well in China's Tier-1 cities: we proved safety in controlled zones first, then scaled out, and finally removed the safety operators. Crucially, we work very closely with local authorities and partners in each region. Sharing our experience from China – like the 500,000+ hours of fully driverless operations we've done across Beijing, Shanghai, Guangzhou, and Shenzhen – builds the trust needed to move things forward smoothly.

Q: As Pony.ai enters new markets globally, it will likely encounter a mix of global players, regional operators, and local champions. How do you think about differentiation - not just from a technology standpoint, but also from a commercial and regulatory standpoint?

Jun: Our deep experience in China gives us a major edge globally. We operate about 300 robotaxis across Beijing, Shanghai, Guangzhou, and Shenzhen, covering over 2,000 square kilometers – that demonstrates serious scale and capability. We're actually the only company approved for paid robotaxi services in all four of those major cities. And we just launched true 24/7 robotaxi services in Guangzhou and Shenzhen. This didn't happen overnight; it's built on years of working closely with regulators and building a safety record that both authorities and riders trust.

On the tech side, our differentiation comes down to proven scale, safety, real-world validated tech stack. Those 500,000+ driverless hours in China show our safety performance is estimated at ten times better than a human driver. At the heart of our autonomoud driving system is our 7th-gen L4 domain controller – it's production-ready, built to last 10 years or 600,000 km, and the gen-7 AD system got full-vehicle redundancy. We validated the gen-7 domain controller over 2 million km on real roads and managed to cut its cost by 80% compared to the previous version. The latest AD system is modular too, so it adapts to different vehicles and platforms. Our in-house developed foundation model, PonyWorld, keeps making the system smarter and better at handling real world driving cases, using automotive-grade chips.

Commercially, we seek ecosystem synergies by building partnerships along the mobility chain. We team up with global mobility giants like Uber, and key regional players like ComfortDelGro in Singapore or Emile Weber in Luxembourg. This lets us tap into their user base and local know-how, freeing us to focus our resources where it matters most.

Q: The autonomous vehicle value chain spans everything from vehicle manufacturing and ownership, to autonomy software, to downstream business models and end-user engagement. How does Pony.ai think about where to focus - and what to build versus partner on - across this stack?

Jun: Our core focus is laser-sharp: we research, develop, and commercialize autonomous driving technology. To get that tech deployed widely, we strategically partner across the ecosystem – especially with automakers (OEMs), transportation companies (TNCs), and key suppliers.

Here's how that plays out: Vehicle making and ownership? That's handled through our deep partnerships with OEMs like Toyota, GAC, and BAIC. We work together to develop and build robotaxis that come with our self-driving system pre-installed. When it comes to operating the fleet and getting users, we collaborate with fleet operators and tap into platforms like Uber, Amap, Ruqi Mobility, or Tencent Cloud.

This "asset light+AI empowered" model is really efficient. It optimizes our capital, shares the operational benefits, and helps us scale faster. Take our partnership with Shenzhen Xihu Group: we'll jointly deploy over 1,000 of our Gen 7 robotaxis. Xihu brings incredible expertise in managing fleets, handling assets, safety protocols, and service guarantees – and they benefit economically from those operations. Meanwhile, we concentrate on pushing the AI and scaling our robotaxi dispatch and service platforms. This mode of collaboration helps both sides and dramatically speeds up large-scale deployment.

By leveraging our tech and partners' strength for market scale and execution, we optimize resources throughout the whole autonomy stack. Partners like Dubai’s RTA or Emile Weber also offer crucial local market knowledge and regulatory insights, which is invaluable for getting approvals and tailoring vehicles or services quickly to fit specific needs.

Q: On a more personal note - as an engineer-turned-CEO, how has your perspective evolved when it comes to balancing product innovation with commercial expansion? What has been the most challenging lesson in leading an autonomous driving company through both phases?

Jun: Moving from engineer to CEO shifted my perspective pretty fundamentally. Early on, innovation was almost the end goal itself – chasing the perfect sensor fusion or solving the toughest edge cases for our "AI driver".

Taking Pony.ai through commercialization changed that: now, I see innovation as a crucial path to creating real, scalable value. Take our 7th-gen L4 AD system for example. It wasn't just a tech breakthrough; the 70% BOM cost reduction and automotive-grade durability were critical design choices to enable mass deployment.

The toughest lesson? Dealing with the time mismatch. Building safe autonomous driving is a marathon – it takes immense R&D effort, like validating your system over 500,000 driverless hours. But running a company demands you move at a sprint commercially to keep things going. Balancing those timelines meant making hard trade-offs. We learned the hard way that favorable policies aren't automatic; you have to actively cultivate them. Engaging regulators early and actively was key. By aligning our tech innovation with national or regional development goals, we turned policies into accelerants for both innovation and business growth.

In short, I think the essence is shifting from the engineer's drive for zero-fail systems to the CEO's job of juggling parallel paths – where tech breakthroughs fuel the business model, and the commercial success funds the next wave of innovation.

Chief Executive of GovTech Singapore, Wei Boon Goh, brings a rare blend of technical depth and strategic clarity to one of the world’s most admired public digital agencies. A former cybersecurity leader at the Ministry of Defence, Wei Boon now oversees an agency building platforms that power everyday life in Singapore.

In this Evolving Edge chapter, Wei Boon shares how GovTech decides what to build, how AI is being deployed in the public sector, and why capability-building is key to digital government.

Q: You have a technical background, which is uncommon among public sector leaders. What brought you to GovTech?

Wei Boon: I spent 26 years in the Ministry of Defence, working on a broad range of technical challenges - primarily in cybersecurity. When GovTech was undergoing a leadership renewal, there was a sense that the organisation could benefit from someone with deep hands-on engineering and systems experience to complement its strong strategic foundation.

They approached me because there aren’t many public sector leaders with that profile. Most technologists in government are in defence-related agencies. I agreed to take on the role because I saw the opportunity to shape how Singapore delivers technology at scale - not just for defence, but across society.

Q: GovTech has built several national platforms in-house, but also partners with external parties when needed. What’s your decision framework for when to build internally vs. when to co-develop or buy from external tech players?

Wei Boon: The most important consideration is the criticality of the product to Singapore. Singpass, for instance, is the foundation to how citizens authenticate themselves across hundreds of government and commercial services. It carries significant security and trust implications, so we build it entirely in-house and retain full control. The same goes for our anti-scam systems, where methodologies and algorithms must remain confidential.

The second lens is scalability. If a platform is going to be used across multiple agencies, it’s far more efficient for GovTech to build it once and make it available as a shared service. A good example is our Q&A AI bot that enables officers to interact with lengthy policy documents across departments.

For less critical, agency-specific tools, we often co-develop or fully outsource to vendors. But even then, we’re deliberate in our choices. We work with multinational tech companies, but also with smaller, agile local firms. Ultimately, the key criteria are having the right capabilities and capacity to deliver, not company size.

Q: Governments are often viewed as slow and rigid. Yet Singapore has a reputation for being digitally agile. How does GovTech strike the right balance between innovation and stability?

Wei Boon: It’s definitely a challenge, but there are three things that help us. First, strong political will. The Smart Nation initiative was championed by our previous Prime Minister and gave GovTech a powerful mandate.

Second, we invested in building internal capabilities. If you want to know what “good” looks like, you have to build it yourself. That’s why we brought in software engineers, product managers, UX designers - so we wouldn’t just be a project manager outsourcing everything.

Third, we help agencies become more self-sufficient. Every ministry and statutory board has unique needs, and we want them to take ownership of their digital products. So we build the platforms, tools, and design systems that let them move fast without compromising quality or stability. We’re making digital capability a whole-of-government skill set, not a central bottleneck.

Q: AI is evolving quickly. Where do you see the most immediate impact in government?

Wei Boon: Our first priority was safety. We didn’t want officers pasting sensitive information into consumer tools like ChatGPT. So we partnered with AI companies to create a secure, sandboxed environment where prompts wouldn’t be stored or used for training. That helped unlock usage across the public service without compromising trust.

Then we started building use cases. One example is generating student testimonials. Teachers normally spend 20–30 minutes writing each one. Now with just a few bullet points, AI can draft a solid first version, potentially saving hours of effort during graduation season.

Another is document Q&A bots. Public policy documents are dense and long. Officers can now upload them and use chat interfaces to extract key info. These are practical, high-ROI wins. They don’t require major system changes, but deliver real productivity gains.

We also launched hackathons to bridge business and engineering teams. These events bring together public officers who have ideas or problems to solve with engineers who can quickly build prototypes. Once a prototype demonstrates value, we assess whether it’s suitable for scale. This model has not only accelerated innovation, but also created ownership and confidence across the ecosystem.

Q: What about longer-term AI applications?

Wei Boon: The possibilities are limitless, limited only by our imagination.

For example, we’re seeing very promising pilots in education. One team built a prototype that uses AI to grade chemistry papers. It was trained on past test answers and showed high accuracy. It’s still early, but it shows that AI can augment—and perhaps eventually automate—some of the most time-consuming administrative work in teaching.

Another exciting prototype is “Language Buddy.” It’s designed to help students practise conversational Mandarin by chatting with a friendly AI. As a parent, I know how hard it is to get kids to speak Mandarin at home. This could be a fun, effective way to reinforce language learning.

And then there are creative, unconventional ideas. One team prototyped a virtual memorial system - where you can pay respects to deceased loved ones online and even “chat” with an AI that has been trained on their writings. It’s not for everyone, but it opens up new forms of digital remembrance.

Q: Beyond AI, what other technologies excite you?

Wei Boon: I’m particularly excited about smart city infrastructure and digital twins. At GovTech, we’ve developed the Open Digital Platform, a foundational layer that enables interoperability across cyber-physical systems. We’ve already seen this in action within the built environment, where we’ve connected systems like sensors, elevators, HVAC, and robots into a unified, intelligent network. For example, if a robot needs to move between floors, it can autonomously call the elevator. If a room gets too hot, the system can automatically adjust the air-conditioning with no human intervention needed. It’s about making public buildings more responsive, sustainable, and autonomous.

We’re also working with agencies like MPA and JTC on digital twins - virtual replicas of ports and industrial estates that support real-time simulation, scenario planning, and environmental control. These tools will be game changers for operational command, resource optimisation, and long-term urban planning.

Together, these efforts show how smart infrastructure isn’t just about technology - it’s about using tech as a force for public good, creating spaces that are more efficient, livable, and future-ready.

Q: Any closing thoughts?

Wei Boon: Singapore’s digital government is only possible because we’ve invested in deep technical capabilities, earned public trust, and had the courage to lead decisively. From generative AI to robotics and digital twins, we’re not just watching from the sidelines - we’re shaping what comes next.

But it’s not innovation for its own sake. We build technology to serve people - to improve lives, strengthen communities, and make public service more responsive and inclusive. That’s what tech for public good means to us.

And to do that well, we must always be building. That mindset of continuous experimentation, delivery and improvement is what will keep us moving forward.