What I'm working on (2025)
It feels like the right moment to write this.
Atlas marked another anniversary of the moment things became real: processing our first order. It feels like a good time to pause, reflect on what we believed, see how the world has shifted, and how our assumptions have evolved.
Where is Atlas now
Singapore is still home, and where most of our customers operate.
Every day, hundreds of restaurants rely on Atlas. We help them make hundreds of millions of dollars annually, generating millions in revenue for ourselves. Some of the most loved brands—Killiney, Haidilao, SaladStop, Les Amis Group—run on us today. A few personal favourites are in the mix too: PPP Coffee, Huevos—happy places I keep finding reasons to ‘check in on.’ And we continue to onboard new restaurants every single week.
Restaurants now use Atlas to manage everything: online ordering, point-of-sale, QR code ordering, kiosks, aggregating orders from all food delivery platforms, loyalty, and a growing set of AI tools. Internally, we joke that we have more products than engineers. Everything runs 24/7—it kicks off with an 8am coffee at CSHH, one of my favourite spots (and a morning coffee institution), and wraps up past 3am at Backdrop, where Dario’s team is slinging proper cocktails. In between, at spots like Willow—a Michelin-starred restaurant—the staff deliver world-class service, customers place orders around the clock from across the country, and Atlas quietly keeps everything moving in the background.
The team is still lean and mean. We recently welcomed another engineer, we’re still looking for a truly world-class designer (our CPO covers design for now). Sales and support have expanded, and we’ve got a hobby entrepreneur-in-residence helping build a great community of restaurant operators.
We’re profitable, steadily growing, and focused on building something our customers—and we ourselves—genuinely love. Each day is an adventure, and the ambition to create world-class infrastructure that’s reliable, invisible, and built to last is what keeps us going.
What we are learning
Singapore’s restaurant scene is vibrant and brutal. Over 10,000 restaurants make 12 billion dollars each year. In 2024, 3,791 opened and 3,047 closed — the highest count in almost two decades. Running a restaurant is tough.
Diners are returning to restaurants, but with new expectations. They now move fluidly between online ordering, dining in, and delivery. Loyalty and consistent experiences across these channels are increasingly important to them.
Restaurants are adapting to keep up with how diners now behave, even as they manage rising manpower costs and maintain good service. Dine-in places are adding online ordering and delivery, while online-first brands are figuring out whether to stick with legacy POS systems or experiment with something new. In short, restaurants want tools that just work—loyalty, ordering, POS, delivery—all integrated smoothly, without extra headaches.
Restaurant closures are always a risk, but even with them rising, we see resilience in four types. First, long-standing institutions are updating their systems to stay relevant and adapt to changing customer behaviors. Second, independent restaurants are finding breakout niches or dishes and strategically scaling them to millions. Third, larger groups continue acquiring and consolidating various brands under one roof. Fourth, new formats experiment by launching fresh concepts to existing customer pools, then rapidly scale successful ones.
Restaurants are moving quickly. We’re building tools to transform them.
What I’m working on
Restaurant operators aren’t engineers. They don’t care about our architecture or tech-stack choices. They care that orders go through, payments land, and things run smoothly, even on their busiest days. Internally, this perspective has turned into a bit of a running joke: ‘but merchants don’t care.’ (I’ve even made some memes about it)
As Atlas grows, we’re supporting more restaurants, processing more transactions, and facing more ways things could break—with more people counting on us to get it right. Speed and scale clarify things quickly: how fast we move shows how clearly we’re thinking, how well we’re executing, and how ready we truly are.
When I talk about scale, I mean something simple and specific: It has to work. At any hour, in any setup, at any volume.
Over the past few months, I’ve spent more time making sure our foundations are solid enough to reliably handle 3x, 5x, and eventually even 10x our current volume. Most of this meant thinking ahead, simplifying wherever we could, and staying disciplined in how we built, tested, and shipped. It involved placing some wild bets, talking to people who’ve built at scale, borrowing best ideas, carefully migrating systems, and steadily raising our own bar.
We’ve rewritten the entire infrastructure multiple times and improved the way we build, test, release, observe, and recover when things go wrong.
Here are a few engineering highlights (I spend most of my time here):
Infrastructure
- We migrated away from a serverless infrastructure that served us well in the early days to a Kubernetes setup that fits our model of isolated compute far better. It unlocked effectively unlimited scale on compute while keeping costs manageable (by resharing the same resources with different utilisation limits for different types of restaurants). It gave us first-class support for database-backed background jobs—critical for time-sensitive operations—and let us easily spin up new workflows that fit our setup, almost like we are building our own platform as a service tool.
- Scaling databases has been one of our toughest challenges. We offer tenant isolation on a shared database, making it tricky to manage resources and connections. By tuning databases, optimizing queries (repeatedly!) to reduce CPU and memory load, and fine-tuning connection pooling, we’ve kept utilization low and unlocked roughly 10x more connections per database. This gives us plenty of headroom to spin up extra processing capacity, maintain resilient standby servers, and quickly deploy new workloads. We also added support for self-hosted databases to power short-term, trial, and testing accounts.
Platform engineering, operations and building the team
- We opened up more of our APIs. Some now power unmanned robots, new experiences, and workflows beyond what we imagined internally. We plan to open up even more, in a structured way, so agents online and MCP servers can start interacting directly with us.
- Observability, performance audits, and our release processes are getting sharper with each new line of code we ship. We now log, trace (distributed!), and observe millions of requests every day across many many services.
- Engineering onboarding for new tenants—from provisioning compute, storage, networks, accounts, and all scaffolding—now happens in just seconds, down from tens of minutes. This lets us instantly spin up new tenants and saves time across the board.
- Expanded the team with some fresh, sharp engineers—and streamlined our onboarding, getting them shipping code to production by their second day.
AI
- We joined Google’s AI‑First Accelerator, giving us access to mentorship and space to experiment with AI. We’ve since rolled out use cases—covering product copy generation, marketing tools, auto-coursing, and insights. Most importantly, it unlocked the technical capabilities we needed to integrate AI workflows into our systems, and sparked the conviction to explore more of what’s possible.
We ship a lot, deploy multiple times a day, and add new restaurants every week. Everything we build is operationally critical, so consistently getting things right matters even more and takes a ton of effort. Nothing is ever truly finished and there’s a lot more to do which is uncomfortably exciting.
What’s next
Restaurant needs keep changing. More features, more integrations, new workflows—some we didn’t expect to build this early. We stay close to where they are and ready for where they’re going.
Our mission is simple: to help restaurants focus on things that matter. Restaurant business models are fundamentally changing as everything moves online, and seamless experiences across online and offline commerce have become crucial. We believe the best way to support restaurants in this shift is by building an open platform that easily integrates with everything they need—this openness is our competitive advantage.
To execute this, we continue to invest in Atlas’s core: a unified source of truth with an ecosystem of features that work magically across all sales channels. It gets stronger with every restaurant that joins, accelerates adoption and innovation through open APIs, and cuts operating costs as they use more of it—driving overall software costs down for everyone.
This time, we’re building with more AI in the stack. Internally, the goal is to make it easier for everyone at Atlas—not just engineers—to build and contribute. Lower the barrier, unlock more experiments, and move faster. Externally, we’re focused on building more experiences, adding and improving workflows, perfecting everyday interactions, opening up more APIs and bringing AI to restaurants where it can truly help.
YC always reminds us to stay focused on the product and customers—and that’s exactly what we’re doing.
That’s all for now—thanks for reading this monologue.
PS: I spend lots of time on infrastructure and engineering operations—happy to swap notes if you’re tackling similar problems. We’re also deep into AI-powered workflows now and would love to learn from folks who’ve done it well. (Email’s open at hey@twkrr.com)