Written by Jesmine Goh.
We Don’t Need to Invent the Future. We Need to Ship It.
5 min read · Part of Sponsoring students at AIE became a test of Singapore's AI ambition
Written by Jesmine Goh.
5 min read · Part of Sponsoring students at AIE became a test of Singapore's AI ambition
On Day 2 of AI Engineer Week, Dr. Vivian Balakrishnan said something that cut through three days of highly technical content like a knife: “You cannot govern a technology you’ve only been briefed on.” Then he described building a personal AI assistant — on a Raspberry Pi.
It was a small moment, but I haven’t stopped thinking about it. Here was a senior public servant not performing familiarity with AI, but demonstrating it — with real code, real hardware, and a willingness to get his hands dirty. That, more than any product demo or benchmark announcement, felt like the most important thing said at the conference.
Because that gap — between being briefed on a technology and actually building with it — is exactly the gap Singapore’s ecosystem needs to close. Urgently.
AI Engineer Week is not a headline-grabbing event. There are no keynote celebrities, no splashy funding announcements, no viral moments designed for social media. What it is, increasingly, is where the people who are actually shipping AI products go to compare notes.
The conversations across three days covered agent primitives, evaluation pipelines, inference infrastructure, sandbox security, and the aesthetics of AI interface design. Speakers came from OpenAI Codex, Cursor, Figma, Google DeepMind, Vercel, and a constellation of fast-moving startups. The throughline wasn’t hype. It was craft.
And for someone watching Singapore’s AI ambitions from the inside, the conference surfaced a quiet tension worth talking about more openly.
The scaling era is over. Did we miss it?
Sara Hooker from Autoscientist gave the most intellectually significant talk of the week. Her thesis: the era of scaling — the belief that bigger models trained on more data will keep delivering proportional gains — is ending. The future of AI is not about who can build the largest model. It’s about who can make intelligence adaptable most efficiently.
“The cost of static intelligence is too high. The next era is defined by the cost of adaptability.”
— Sara Hooker, Autoscientist
This reframing matters enormously for Singapore. We were never going to win a compute arms race with the US or China — the capital requirements alone make that a non-starter. But the post-scaling world plays to different strengths: algorithmic efficiency, domain-specific fine-tuning, high-quality data curation, and the ability to deploy intelligence into real systems with real constraints.
These are strengths Singapore can build. The question is whether we’re building them fast enough — and whether our institutions are moving at the speed the moment requires.
Assembly is the skill. Invention is a distraction.
One of the subtle but important themes across the conference was a shift away from the idea that building AI means building models. JJ from Google DeepMind put it plainly: stop using the LLM as one giant problem-solver. Use it in smaller, focused chunks, surrounded by determinism. The intelligence isn’t in the model alone — it’s in the architecture around it.
Sam Bhagwat from Mastra AI ran a full workshop on agent primitives: memory management, multi-agent orchestration, observability. Exa AI’s Conrad Soon showed how structured communication between agents — passing typed objects rather than free-form text — makes systems dramatically more reliable. Gavriel Cohen from Nanoclaw argued that security isn’t a feature you bolt on later; it’s an architectural choice you make on day one.
The builders who will define this era are assemblers, not inventors. They compose existing capabilities into systems that are reliable, observable, and maintainable. This is not a consolation prize. It is, right now, the hardest and most valuable skill in the industry.
“Every prototype that survives becomes something a person has to maintain.”
— Jimmy Lai, Vercel
For Singapore’s engineering talent — trained in systems thinking, comfortable with constraints, embedded in one of the world’s most demanding enterprise markets — this is good news. We don’t need to have invented the transformer architecture. We need to ship production systems built on top of it, in healthcare, finance, logistics, and government, where the failure modes are real and the stakes are high.
The sovereign stack question
Dr. Feng from GovTech outlined a compelling vision: government must evolve from AI-enabled to AI-native. He described a sovereign agentic harness — an MCP gateway, sandboxed runtime, agent identity management, memory, and a skills platform — as the architecture for deploying AI responsibly at scale.
By 2028, the estimate is 1.3 billion agents operating globally. The question for Singapore isn’t whether our government will be touched by that. It’s whether we’ll be running our own agents, on our own infrastructure, with our own oversight — or dependent on platforms built and governed elsewhere. This is where Singapore’s instinct for institutional seriousness, sometimes mocked as over-caution, becomes an asset.
What the next generation of builders needs to hear
The people who are least at risk in this new landscape are the ones who got their hands dirty. Not the ones with the most credentials. Not the ones who attended the most briefings. The ones who built something — broke something — figured out why — and built again.
Jan Liphart from Openmind said something that stuck with me: the best evaluation criteria for AI systems aren’t benchmarks. They’re smiles, tears, trust, and memories. Advancing humanity starts with understanding it.
That is what Singapore’s ecosystem needs most. Not just engineers who can tune models. Builders who understand the people those models serve — who can hold the technical and the human in the same hand. Who are willing to pick up a Raspberry Pi, make something work, and learn from the failure.
The frontier isn’t waiting for us to be ready. It’s moving whether we show up or not. The question is whether we show up prepared to build — or just to watch.