Singapore Leads the World on AI Adoption. Most Teams Still Don't.
What I learned at Notion Sessions Singapore about the gap between buying AI and actually using it.
The teams that scale write down what lives in the founder's head.
Founder-led design: everything Notion’s Head of Design, Randy Hunt, covered at the Notion AMA
If you’re in a startup or SMB that’s new to Notion, try out Notion Business with AI for 3 months here.📺 If you prefer watching instead of reading, check this out: https://youtu.be/UiIUsPNN3H8
Last Wednesday (11 June), I sat in on Founder-led Design: An AMA for Startups with Randy Hunt, Head of Design at Notion, and it was the clearest articulation I've heard of why "taste" survives the AI era instead of being flattened by it.
The thread running through every answer: design shapes perception to change behaviour, and trust is the lever. Most of his answers were really one argument applied to a different surface, so that's how I've laid this out: each section is a place where Randy takes something teams usually treat as a vague feeling and turns it into a concrete decision you can actually act on.
The first one is the one nobody expects to be a design problem at all.
Speed is a design decision, not an engineering one
- Randy's most impactful intervention as a designer: choosing to make the front-end faster. Speed correlates directly with how much users trust the system.
- Concrete Notion work: speeding up database loading and agent response times.
- The counter-intuitive bit: for AI interactions, dripping information out over a couple of seconds (showing the thinking) makes a longer wait feel shorter than a loading spinner followed by a finished result. That perceived speed increases trust and willingness to engage.
- This is animated through Notion's avatar, internally called "Nosy". It reads software state changes and transitions fluidly between writing, searching, and resting states. Cute and endearing, but functionally it's moderating how you perceive each moment.
Speed makes the product feel a certain way to a user. The next move is the same trick aimed inward: making the team's progress visible to itself.
Design Lightning, a ritual for organisational transparency
- A ritual Randy has carried across multiple companies. Purpose: less about craft, more about moving design through a complex org and making progress legible.
- Mechanics: 20–25 people each share what they're working on for ~1 minute. One "driver" clicks through each person's shared screen while everyone shouts "next, next, next." The absurd pace is the point; it's playful and silly on purpose.
- Output: a 30-minute video every two weeks. Watchable at 1.5x or 2x. Randy has run this for 2+ years, with a Notion database tracking every session, a rolling archive of how the team pushed the product forward.
- The twist: the biggest consumers are outside design. Many call it the most valuable 15 minutes of their fortnight, a status update on all new value being created.
That covers how Randy makes the work visible. The next stretch is about who he makes it with: the founder. Founder-led design just means the founder holds the taste and product vision directly rather than handing it off to a design function, and a few of his answers were about what that actually takes.
Founder-led design, two cohorts of founders
- Founders who are designers themselves, and founders who aren't. Both can work, but it's smoothest with designer-founders.
- Why: deep empathy for the craft eliminates the work of explaining or advocating for design. No educating people on why it's good, why it's hard, how long it takes. Lower friction, more focus on the actual work.
The same closeness that makes designer-founders effective is also where it can go wrong.
The dark side, holding on too long
- The trap: founders execute through the discipline they came from to get the company off the ground, then hold onto it too long. The thing they cared about most atrophies because they never let it scale or evolve organisationally.
- The best founders scale themselves alongside or ahead of the org. Randy avoids it by zooming up a level and making broader choices. At Notion, he says it's a non-issue; part of why he joined was the intuition that working closely with Ivan would enable the work he wanted to do.
If the founder can't personally review everything as the org grows, the question becomes how their standard survives without them in the room. Randy's answer starts before anyone does any work.
Protecting founder vision happens at hiring, not in design reviews
- You don't protect the founder's spirit through feedback on individual designs. You do it at the human level: values, intent, and who you hire.
- Notion hires people intrinsically motivated to reinforce the same values, so the team self-checks around care level, personality, and experience.
- The anecdote: at a Notion event in Sydney, Randy watched Sophie agonising over the placement of rugs, trees, and plants, the same standard of care at every level of the org. It made him realise he had to operate to that bar too.
Hiring solves how a human team holds the standard. The bigger question in the room was whether AI erases the need for that standard at all.
AI, taste, and codification
- The question: does AI make features free and flatten everything into sameness, or does it make founder taste more valuable? Randy's answer: taste stays critical. AI doesn't flatten if you codify what you want.
- Codification isn't new. It's what designers have always done: write down what's good, show examples, describe components in software or Figma. Values and principles go into Notion docs as guidelines and standards; branding has done this for decades with manuals. All that prose becomes context for the models.
- If you blindly prompt with no direction, you get a magical-feeling result fast, but its alignment with your org is poor.
- For non-designer or technical founders: make artifacts that reflect how you want things to be. Prose on how you make decisions, your design principles, usability considerations, plus the assets the models need to stay on-brand (interface elements, behaviours, speed, interaction patterns).
- The shortcut if you already like your product: point the models at your codebase and say more like this. If designers and engineers have shaped it and you like where it's at, that's the best context you have: "do the next thing like this thing."
- The fallback if you don't have the taste or the experts in-house: bring someone on, or work with a product agency, to arrive at a point of view you can then codify. You can't codify what you haven't formed a view on yet.
- Randy's sharpest line on this: anyone doing good work today is creating output for humans and agents at the same time. The audience for your guidelines and artifacts is as much the models as it is people. You have to author for both.
Codifying your taste lets a model act without you. The same logic applies to your team: the goal is people who can ship without waiting on you.
Empowering a team without becoming the bottleneck
- One-way doors vs two-way doors. If a decision is easily reversible, the team ships it directly, no approval needed. The org has to be culturally comfortable making something then fixing it.
- When someone ships unreviewed, it isn't a gamble, because the confidence was built upstream. Hiring and shared values mean Randy already trusts how that person thinks, so he doesn't need to inspect the output to know it's roughly right. If the decision does get questioned later, the person doesn't defend the pixels; they explain the reasoning behind the choice and the plan for addressing the concern. That reframes the moment: instead of "you shipped something wrong," it becomes "here's how I was thinking about it, here's what I'll adjust." Friction turns into collaboration because the conversation is about reasoning, not blame. And because most design decisions are reversible (the two-way doors above), the cost of being wrong is a fix, not a disaster.
That whole model rests on hiring the right people, so it's worth knowing what Randy actually screens for.
Hiring for a continuous-improvement mindset
- The interview technique: ask candidates to show something not good, something they're unhappy with or that didn't go as planned. The logic is that a polished case study only proves someone can present a finished result; it tells you nothing about how they think when the work is still rough. Asking for the unfinished or failed thing surfaces the actual skill Randy is hiring for: can they see what's wrong with their own work, and do they have a view on how it should be better? Candidates who only show three glossy case studies miss the point, because they're optimising to look impressive rather than to demonstrate judgement.
- Strong signal: someone says "let me find the thing," opens a GitHub project they've kept for years, and pulls up work in progress instantly. Randy reads three things from that single move. One, they're a constant maker (the project exists because they build for its own sake, not for a portfolio). Two, they're organised enough to retrieve it on demand, which is its own form of care. Three, they're comfortable showing unfinished work, which means they're not precious about it. Taken together it says this is someone in love with making software anyway, whether or not anyone is watching, and that intrinsic pull is exactly the continuous-improvement mindset the rest of the talk keeps circling back to.
Vision and the "pivot to AI"
- Someone framed Notion's recent AI focus as a dramatic top-down pivot. Randy pushed back: it didn't happen that way. Ivan has an insatiable drive to push things forward and a clear view of what ought to happen next (if not always how).
- The design team's job: listen carefully, add precision to Ivan's vision when he knows something but hasn't articulated it yet. Collaborative everywhere, but the founder holds and pushes the vision rather than outsourcing it.
Culture as religion
- Randy's underlying point: culture isn't a set of stated values, it's built from repeated physical and ritual choices, the same way a religion is. Referencing an Ivan Zhao interview on building a culture people commit to like a religion: the building blocks are rituals, symbolism, myth, and storytelling, the same across religions, cults, and brands. You do each piece and repeat, repeat, repeat.
- At Notion it's felt physically: quality, detail, timelessness. Lighting is always warm, no down-lighting, only uplighting at ceilings and floor-lamp level. Randy's framing: the feeling is engineered, not declared, more like standing in a cathedral than reading the Bible. You sense it rather than getting told it.
Craft = skill + practice + judgement
- Randy breaks craft into three parts, each with its own gradient from none to a lot.
- Skill is the ability to make the thing.
- Practice is the repetitive ritual of doing it over and over; he calls this maybe the most important part.
- Judgement (taste) is recognising what's good, seeing the gap between what you can make and what you want to make, then applying the skill through the practice to close it.
- The pottery metaphor: skill is knowing the wheel, the pedal, hands on clay. But you only get good by throwing pot after pot. Judgement is knowing when the pot is right.
- It's all teachable; people get much better. Some just start closer to the finish line ("DNA or the power of God," in his words).
The fourth ingredient, intrinsic motivation
- Beyond the three parts is the one that actually powers high-craft work: an intrinsic drive to always make the thing better. The best people operate in a state of constant dissatisfaction, and are completely fine with it. "It can always be better. Always. That's the joy of it."
- Three sources of that drive: a belief that people deserve nice things (a kind of humanism; making something good is a gift to others), the pleasure of experiencing beauty in the world, and the fact that the pursuit itself is deeply satisfying, putting in effort and seeing the thing improve.
- Find that hook and high craft becomes the default: you're always chasing the better version, doing it over and over, dissatisfied but having fun doing it.
- It leaks into everything. Randy's own example: unpacking a hotel room on the Singapore trip. Clothes unpacked in a set order and hung immediately, shoes placed just so, toiletries laid out with a spatial relationship to each other. There's a quick operational analysis running underneath: “do I have 30 minutes to iron everything or 5, how long does the iron take to heat, can I do it all in one pass?” The need for order doesn't switch off.
Developing taste when you're not a designer
- For an early founder building a product with no design background: how do you even find your own footprint? Look to external examples you judge to be genuinely good. Pick one or two as reference points and start there.
- The mistake nearly everyone makes: reaching for something familiar and close, a semi-competitor, the startup down the hall that seems to be doing a good job. That just averages you toward your immediate space.
- Instead, look across disciplines for permanent things you find good, using your own judgement rather than asking around: a sneaker you love wearing, a building that makes you feel good, a restaurant doing it right, an apartment that impressed you. Draw the bar from quality experiences anywhere, not just from products that look like yours.
The common theme for me: Randy takes the things most teams can only gesture at and gives each one a concrete form you can measure, build, or repeat.
- Trust stops being a hope and becomes a speed number you can tune.
- Taste stops being "I'll know it when I see it" and becomes a written Notion doc you can hand to a new hire or a model.
- Progress stops being a status meeting and becomes a 30-minute video archive.
- Culture stops being a values poster and becomes a literal lighting spec.
- Craft stops being talent and becomes a ritual you run on a schedule.
Every one of those moves does the same job: it takes something that lived only in one person's head and turns it into an object other people, and now AI models, can use without that person in the room.
That's the whole difference I see in the teams I work with. Most treat trust, taste, and culture as feelings, which is exactly why they can't grow them past the founder or encode them for a model. Randy treats them as things you can write down, and that's what lets them scale.
Want your taste, processes, and standards written down so they scale past you?
That is the work we do at work.flowers. We turn the way you run things into a Notion system your team and your AI tools can actually use, so the standard holds even when you are not in the room.
Reach out and tell us what is living only in your head right now.
Try out Notion Business with AI for 3 months here: affiliate.notion.so/pyn572ygxpb0🤗 Join NotionSG Community: https://t.me/notionsingapore
Related Posts
Why Notion Workers + Dashboards are my New BI Stack
After Notion's developer platform launch, Workers and Dashboards make a credible BI stack for most small teams. Why it's now my default recommendation, and where the warehouse-plus-Tableau setup still has the edge.
Getting Started with Notion Custom Agents
Notion's new Custom Agents let you automate workflows directly inside your workspace. Here's what they do, how to use them effectively, and where we think Notion is heading next.