Written by Peh Cheng Ye.
Singapore Is Hungry for the AI Frontier
16 min read · Part of Sponsoring students at AIE became a test of Singapore's AI ambition
Written by Peh Cheng Ye.
16 min read · Part of Sponsoring students at AIE became a test of Singapore's AI ambition
I’m a graduating student from SUTD (Computer Science and Design pillar), looking to start my business helping SMEs in the premium food sector tell better stories through the use of NFC technology on their packaging.
Before matriculating into SUTD, at the start of 2022, I was self learning graphic design using Adobe Illustrator because I believed that developing my design skills would provide me an unique advantage as a CS graduate 4+ years later.
“You can only connect the dots looking back”
At the same time in 2022, I started going to Crypto events as I’ve been a long time lurker in this space, wanting to meet more crypto natives just like myself. I quickly realised that this is the community I want to be part of. I’ve now been to many crypto events and conferences (TOKEN2049 Singapore in 2024 & 2025, Devcon 7 Bangkok in 2024, Catstanbul in Istanbul 2025, etc) and got to experience what it feels like to be part of an emerging industry.
In my Year 2 at SUTD, I joined SEVEN (SUTD Entrepreneurship Venture and Enterprise Network) as the Vice President for 2 years and ran many events (including Blockchain focused ones) and got to build the entrepreneurship ecosystem in SUTD.
The community is everything. It is worth investing in people.
Now, Singapore has the privilege of claiming titles such as “The Hub of Asia”, being THE place for many things: Financial Hub, Innovation Hub, hosting F1 & the many iconic conferences like TOKEN2049.
Yet, especially if you’re in the entrepreneurship space, you’ll realise that the most brilliant people aren’t here: they're in Silicon Valley. San Francisco is undoubtedly where you’ll like to be if you want to be at the epicenter of innovation. Right at the heart of it.
However, that being said, from my own experience as a crypto native I can assure you that brilliance has no boundaries. The community is what matters. People matter. You don’t have to be in SF to meet builders who are just as passionate as you are. You can start right here, right now!
This is my framing going into this new AI frontier. It feels like crypto back in 2021. It’s kinda crazy how I could feel the same excitement now as if I was discovering the crypto space for the first time in 2021.
I’m lucky to be one of the 20 students who were sponsored (out of their own pockets) by 65Labs, and the many other individual contributors, to attend the AIE conference.
In the months leading up to the conference, I’ve been participating in some AI events myself. I can still remember the first AI event I participated in (which was also partly hosted by SEVEN, shoutout to Aaron for making this happen!), featuring Gabriel from OpenAI as one of the speakers sharing about their workflows.
Slowly, I started seeing more and more AI events popping out on Luma, and then it feels like there’s at least one event happening every other day. Time and time again, each event was oversubscribed, and people filled the venue to max capacity. Something special is happening in real time.
Singapore might feel mundane at times. People working their ass off in the day and rushing home in the evening to wind down & repeat. Yet, it feels like there’s not enough events to keep up with Singapore’s demand for participating in the AI frontier!
Singapore is hungry. VERY hungry.
Here’s 5 things that really stood out to me across these 3 days:
- Mike Cann’s (Developer Experience Engineer at Convex) Workshop
- Vivian Balakrishnan’s (Minister for Foreign Affairs) sharing
- Aosheng Ran’s (Designer at Figma) sharing
- Conversation with Dr Feng Yuzhang (Head of AI Practice & Director of AI Programme at GovTech)
- Conversation with Ryo Lu (Head of Design at Cursor)
I’ve heard of Convex before but never got a chance to understand how people are using it. Mike’s workshop went into the details on how to get started, what you could do with it and what are things to take note (e.g. schema migration).
To me, Convex is essentially compressing the backend into the database itself, allowing developers to build applications in a faster way whereby web technologies are treated as first class citizens (Convex itself is in Typescript), with the magic of creating React-like liveliness.
This is major, as I remember vibecoding my first application which was a real time booking application for an investor event, whereby investors can schedule slots with the various startups simply by indicating who they want to talk to, and the application will put everyone’s request into a timeslot in real time. The major challenge was implementing this with free tools, and the free way was just not possible. The bottleneck is the database: getting stale data and running into race conditions.
Looking back, there was indeed a free manner of powering the application (running on Railway with the free CPU allowance & single process backend to interface with the database) but the inexperienced me could not figure it out. I was only familiar with Vercel deployments, which is also the most popular platform builders of today’s AI accelerated environment are choosing.
It is so important to build tools around how humans intuitively utilise the technology, not to ask developers to change their behaviour to suit the tool -> which is ironic, because in a world that’s rapidly filling with AI generated stuff, it becomes even more important to design tools for the ultimate agent: humans. I think Convex is uniquely positioned to elevate the builder experience.
This neatly brings me to the next highlight, and there’s a clear distinction I need to make here: every developer is a builder, but not every builder is a developer. The builder profile is uniquely able to maximise the full benefits of this AI era.
This is quite pivotal. And undoubtedly, Vivian Balakrishnan is a builder. Here are some takeaways from his morning sharing:
I absolutely did not know he used to be an eye surgeon. As a Singaporean, Vivian is just a minister, but his talk (or more precisely, his famous article on LinkedIn/Facebook) was a watershed moment.
Hearing the same quote “You can outsource memory and computation, but you cannot outsource understanding” which I first heard in YC podcasts spoken by the minister himself was really mind blowing to me.
For some time, the government has been paddling AI narratives all over, in every parliament debate, in every public address, AI and the need to be an AI-first nation has been plastered all over. And for the longest time I could not tell whether or not the government itself really understood what they’re talking about.
Yet, here I was learning about the minister’s Nanoclaw workflow with some technical terms (e.g. edge deployment, semantic search) sprinkled around. This is the moment that changed my view on the government’s AI narrative. It is real. It is happening. And the government & its ministers are serious about it.
As an economy mostly driven by our services sector (because human resource is Singapore’s ONLY resource), we have to be AI first. And I’m glad that our government is putting serious work behind their AI narrative.
“You cannot govern a technology you have only been briefed on” <- this is the exact philosophy which will be the key to ensuring our government’s edge in the AI era.
To me, the magic of Figma (and Canva, alongside other players like Miro) is real time collaboration. Seeing someone’s cursor in action is significant both on a technical and productivity level. Aosheng demo-ed his canvas at [https://aoshe.ng/aie]\(https://aoshe.ng/aie) and it was wild to me how he’s pushing co-creation in the uniquely Figma way: featuring visual branching to represent version control, and real time prompting which also allows others to edit the prompts on-the-fly.
Right now, the technical challenges might make co-creation workflows like these very finicky, but that should not be the focus. The only focus is to accelerate towards tools which complement human intuition, not to be hung up on the small technical details which will eventually iron out.
I am super optimistic about this future.
## In-person conversations at the booths
The 2 exhibition areas (one at Pullman and one at the hotel itself) is where the alpha is really at. I had many meaningful conversations with booths like The Robot Company, OpenAI, CloudFlare, and Exa. But 2 conversations I had stood out which I would like to share:
Conversation with Dr Feng Yuzhang, Head of AI Practice & Director of AI Programme at GovTech:
I couldn’t help myself but to hear from the man himself first-hand, how AI is changing the government from the inside. From our conversation, I learnt that our various government branches are somewhat autonomous, whereby GovTech is only empowered to push best practices and provide advisory to the other branches.
The reality back then was each branch is really only focused on their core functions: MOE on education, etc, whereby technical expertise relied more on contractors to fulfil. However, the trend is moving towards building those expertise in-house, to support greater technological integration.
GovTech does get involved, and I can’t help but to think they’re probably the one of the earliest organisations to be practising the “Forward Deployed Engineer” strategy -> helping different branches of the government set up tech integrations/stack to ensure that Singapore always maintain an edge, largely driven my frontier tech adoptions.
There is always much to be desired, but relative to what other governments are doing, I am confident to say that we are definitely on the ball here.
Last, and definitely not least, is my conversation with Ryo Lu, Head of Design at Cursor:
I remember the first time I got to know about Ryo was a Y Combinator YouTube video where he and Aaron Epstein sat down together to roast various startup websites. I usually only listen to videos like a podcast while doing household chores, but Ryo’s critics were really on point and resonated strongly with me, and I had to watch the rest of the video with my full attention.
Honestly, videos and podcasts undersell him. In person, he's visibly passionate and you’ll feel it: high-conviction opinions delivered with real energy. There were moments that genuinely gave off Alex Karp vibes. I stayed around him for about an hour, listening to other people's questions and asking my own.
Most people there knew him from his main stage talk. A lot were product designers (some lowkey feels like they have friction with their engineers), one person asked how to "influence engineers when you change something directly in Figma?", which felt oddly narrow. Another guy pushed back on Ryo’s main stage’s phrasing of "no black box" ideology, arguing it's misleading since Cursor doesn't show you exactly what's happening under the hood.
Ryo held his ground. Here’s my take on his explanation:
Abstracting complexity from users isn't deceptive; it's good design. Non-black-box doesn't mean open-sourcing everything or giving users 500 toggles/configs. The magic of Cursor is that the defaults work really well for most people, while power users still have enough room to customise. It is never suffocating.
Ryo doesn't think much of Figma, to him it's just a sketching tool, not the actual product. Here’s his bigger idea: the codebase as the single source of truth, now made possible by AI:
- Need a mockup? Work it out in the codebase.
- Documentation? Lives in and refers directly to the codebase.
- Need a report or presentation? Point an agent at your codebase and draft it from there.
Here are some answers that actually shifted something for me:
I asked: "How do you grow Cursor's utility without scope creep, adding more things you can do without losing focus?"
Ryo: Whatever you want to do, your current features can probably already do it. Don't reach for a new feature; look at what exists and figure out how to tweak or improve it so it closes the gap. You’ll have to ask yourself: how can I change this feature(s) in a way that supports what the user wants?
I asked: "Do you study user intent to decide what to improve?"
Ryo: User intent is only half the equation. If you build a feature for every user intent, you end up with a bloated product: one feature per use case. The better question is: how do these intents map back to existing features and workflows? Everything Cursor has shipped that looks "new" is really the same core features repackaged: cloud agents, CLI, IDE, chat, etc.
Same building blocks, different interfaces.
This mirrors what he practiced at Notion: a to-do list is just a block of ordered information. A calendar is the same block arranged by timestamp. A kanban is the same block with assignments.
Same core unit, extended through composition rather than invention. Notion's entire product surface is basically that one principle repeated.
A question I threw out that got him thinking:
“Design has WYSIWYG (what you see is what you get). A poster looks the same for everyone. Drag something, export it, it's identical.
Could programming have its own version of WYSIWYG: where code always behaves exactly as designed, matching the engineer's diagrams and decision trees? Could developers design behavior visually instead of writing code, running test suites, and manually updating documentation every time something changes?”
Ryo was intrigued by this for sure, and expressed that this is something interesting to explore.
I asked which parts of Cursor are opinionated -> whether Cursor nudges users toward specific tools or behaviors.
Ryo: Not really. Cursor gives you the tools and stays out of your way. The one opinionated layer is the defaults. Defaults are tuned to work well for 90% of users. Beyond that, it's yours to configure.
On shipping cadence: they give engineers and designers high autonomy, but not everything ships immediately. Some features sit in a sandbox to mature. They're experimenting with limited rollouts to observe behavior before going wide.
Also at the Cursor booth, I learned they'd shipped a canvas feature where the agent gives you a bird's-eye view rendered as HTML. That made me think a little further.
If the codebase is now the single source of truth for documentation, presentations, demos, and project management… then what is the single source of truth for the codebase itself?
Maybe it's HTML.
Not as a file format, but as the live interface layer that sits between you and your agents. You operate at the level of design and intended behavior. The agent interprets your prompts against a persistent, structured representation of the system. The HTML canvas isn't just a visualization: it's the context window your collaboration with the agent actually runs on.
Cursor's canvas feature might be an early signal of exactly this. This is a development worth watching.
What do you think Singapore’s ecosystem needs to be at AI’s frontier?
Here’s what we would like but fundamentally cannot have in Singapore:
- Mega data centers with massive compute training frontier AI models
- Frontier AI researchers (SF is more attractive for these top 0.00000001% talent)
That’s it! It may look like a handicap here but the vast impact of AI innovation does not happen at the very cutting edge, like how F1 pioneered the usage of disc brakes, which eventually saw most of the benefits in everyday cars, used by everyday people.
My point is, most of the value creation will happen at the application layer. Think back to the most useful tools (which were designed around human intuition). The tools can be engineered by mega brains but it’s only useful if a lot of us are able to use it.
So, Singapore’s ecosystem needs to build an environment which encourages innovation at the application layer, answering the hard question of: ok this is cool, but how does it help me with [insert pain points here]?
Our environment needs to empower people to solve problems.
AI has the potential to create completely new markets by automating the tasks which do not require the ultimate agent that is the human.
When AI agents could start doing the current job of a Lawyer, the Lawyer is not out of a job, but the Lawyer is now empowered to address other issues which an AI agent could not do.
There’ll always be more things to do, because human creativity and human wants have no end. It resizes infinitely. Value creation can be infinitely created. It’s like creating entirely new pies from just a small slice of the previous pies.
That gives a clue…
The people who do this “infinite value creation”, creating new markets, or what Jensen Huang likes to call the “Zero-Billion-Dollar market” are entrepreneurs!
So my answer is: Singapore needs to empower entrepreneurs to change the world. We have to let people try. It’s all about trying. And the best part about the AI era is we can try a whole lot with just a little.
The problematic thing about our culture is that we love to optimise processes down to a science; there’s always the best way of doing things. However, great entrepreneurs are born in all kinds of circumstances, with all types of backgrounds. You cannot just put your bets in a narrow band -> on just “the best” people.
You have to let young people try.
You have to just let people try.
To get more people to try, empower them to try.
Small efforts go a long, long way. Just like how the folks at 65Labs & other individual contributors did with sponsoring 20 students. We were given a chance. Give people a chance.
Let people try.
It’s a culture. Letting people try can be holding back your “NO” to your employee who suggests a different way of doing things. Letting people try can be giving AI credits to builders. Letting people try can be restructuring entrepreneurship programs in academia (from Primary Schools to Universities) to be more accessible, and asking for less concrete outcomes. You have to let people try.
What we must do more of is to eliminate (or at least lower) expectations of concrete deliverables for every attempt.
The best part about AI is that someone can spend a few hours and a few AI credits and create something new. If it sucks? So what. You only wasted a couple hours and dollars. The upside to this is ridiculously higher than any downside. Lean into it.
Just let people try.