이 포스팅은 쿠팡 파트너스 활동의 일환으로, 이에 따른 일정액의 수수료를 제공받습니다.
Chapter 40: Version Three
Hana came back on the fourth day, not the third, and she came back different.
Not in the ways Dojun had feared—not diminished, not defeated, not hollowed out by the particular exhaustion that had consumed her predecessor in another timeline. She came back rested, clear-eyed, and carrying a single sheet of paper that she taped to the conference room wall before anyone else arrived.
The sheet read:
ARIA 3.0 DESIGN PRINCIPLES
1. Ship at 4.7. Perfect at 5.0 in version 3.1.
2. The user doesn’t care about TaskFlow. The user cares about their Tuesday.
3. If Hana hasn’t slept, Hana doesn’t design.
“Rule three is new,” Dojun said when he saw it.
“Rule three is the most important one.” She uncapped a marker. “I spent four days sleeping, cooking, and drawing things that had nothing to do with Aria. On day two, I sketched my grandmother’s rice cake stall from memory. On day three, I redesigned my apartment’s bookshelf layout. On day four, I woke up at 6 AM and knew—with absolute clarity—exactly what was wrong with the 3.0 interface.”
“What was wrong with it?”
“It was trying too hard. The conversation metaphor—the suggestions, the anticipatory messages—they were too eager. Like a waiter who refills your water every thirty seconds. Technically attentive. Emotionally exhausting.” She drew a simple diagram on the whiteboard. “The fix is restraint. Aria doesn’t suggest unless asked. It doesn’t anticipate unless the pattern is strong. It waits. Like a good assistant—present but not pushy. Available but not aggressive.”
“That’s a smaller change than I expected.”
“The best design changes are small. Big changes mean you got the fundamentals wrong. Small changes mean you got the tone wrong. Tone is fixable in a day.” She smiled—the first real smile in weeks. “Taeyoung, how long to adjust the suggestion trigger threshold?”
“If I’m just changing when suggestions appear? Two hours.”
“Do it. Increase the confidence threshold from 70% to 85%. Aria only speaks when it’s sure. When it’s unsure, it stays quiet.”
“That’ll reduce the number of suggestions by about 40%.”
“Good. Fewer suggestions, each one right. That’s the design.”
The change was implemented by noon, tested by 3 PM, and deployed to the beta group by 5 PM. The user feedback was immediate and unanimous: “It’s like Aria grew up. Less chatty, more useful. Finally feels like an assistant, not an attention-seeking intern.”
Hana read the feedback and pinned it next to her design principles sheet. “Attention-seeking intern. That’s what 4.7 felt like. ‘Grew up’ is what 4.9 feels like.” She turned to Dojun. “Ship it.”
“You’re sure?”
“I’m sure because I slept. Rule three works.”
Aria 3.0 launched on June 15th, 2010, across all platforms simultaneously—iOS, desktop, web, and the Google Workspace integration. The launch was quiet by design. No press event. No TechCrunch exclusive. No countdown timer on the website. Just an update notification that appeared for existing users: Aria 3.0 is here. Same Aria, smarter conversations. Update now.
“That’s the most understated product launch in the history of technology,” Jiyoung said.
“We’re not launching to the press,” Hana said. “We’re launching to our users. They’re the ones who matter. If the product is good, they’ll tell everyone else. If it’s not, no amount of marketing will save it.”
The product was good.
Within 48 hours, Aria 3.0’s user satisfaction scores hit 4.9 out of 5—the highest in the company’s history. The LLM-powered task detection, combined with Hana’s restrained suggestion model, produced an experience that users described with words like “magical,” “thoughtful,” and “it just knows.” Daily active user rates jumped from 71% to 78%. Enterprise clients who had been evaluating TaskFlow’s beta paused their evaluations—not because TaskFlow was bad, but because Aria 3.0 was better in the ways that mattered.
“TaskFlow is smarter,” one enterprise IT director told Jiyoung during a retention call. “But Aria is wiser. There’s a difference.”
“What’s the difference?”
“Smart means knowing the answer. Wise means knowing when not to answer. Your system doesn’t flood me with suggestions I don’t need. It waits until I need help and then it’s there. That’s not AI—that’s design.”
Jiyoung forwarded the quote to the team Slack. Hana framed it.
The market responded. Not with the dramatic fireworks of a viral moment, but with the steady, compounding growth of a product that had found its voice.
Consumer users: 134,000 before 3.0. 180,000 by August. 220,000 by October.
Enterprise accounts: 2,800 before 3.0. 3,600 by August. 4,500 by October.
Monthly recurring revenue: 89 million won before 3.0. 142 million by August. 198 million by October.
“We’ll cross 200 million in monthly revenue by November,” Minjae reported at the October standup, his voice carrying the particular disbelief of someone who had once tracked company finances on a spreadsheet held together by prayers. “Annual run rate: 2.4 billion won. Profitable. Growing. And—for the first time—we’re growing faster than TaskFlow in the Asian market.”
“How’s TaskFlow doing globally?” Dojun asked.
“Growing fast in the US and Europe. They’re dominant in English-language markets—their LLM is optimized for English, and their Sequoia connections give them distribution in Silicon Valley. But in Asia—Korea, Japan, Southeast Asia—we’re winning. The localization, the design sensitivity, the enterprise relationships. They can’t replicate that from San Francisco.”
“Two hemispheres, two leaders,” Seokho observed during their Tuesday call. He had been following the competitive dynamics with the analytical intensity of someone who had a financial stake in the outcome—Nova Systems hosted both Aria’s and several of TaskFlow’s Asian clients’ infrastructure. “TaskFlow owns the West. Aria owns the East. The question is whether it stays that way or whether one side crosses into the other’s territory.”
“We’re not crossing,” Dojun said. “Not yet. Asia-Pacific has two billion potential users. We haven’t even scratched India, Indonesia, or Southeast Asia. There’s a decade of growth in our backyard before we need to think about San Francisco.”
“Wise. In the military sense—don’t open a second front when the first front is still unsecured.” A pause. “The competitor isn’t your biggest risk, Park. Growth is. You’re adding employees, offices, and clients faster than your culture can absorb them. Forty-five people in June. Sixty by year-end projection. At that rate, you’ll be a hundred by mid-2011.”
“I know.”
“Do you? Because the company that has a hundred people is not the company that had twelve. The relationships change. The communication channels multiply. The founder’s vision gets filtered through three layers of management. And the magic that made Bridge—that made Aria—start to dilute.”
“We’re building for that. The two-track structure, the design principles, the co-equal governance—they’re all scaffolding for scale.”
“Scaffolding helps. But scaffolding isn’t culture. Culture is the stories people tell about what matters. Make sure the right stories survive the growth.” He paused. “The japchae story. The banchan vendor. The 3 AM phone call. Those are your cultural stories. When a new engineer joins and someone tells them about the ajumma in Namdaemun Market who uses Aria to track her kimchi—that story does more for your culture than any HR onboarding deck.”
“Since when do you give culture advice?”
“Since I hired my twentieth employee at Nova and realized that I couldn’t personally teach each one how to eat naengmyeon at the correct speed. Culture scales or it doesn’t. I’m learning.”
“From whom?”
“From you, mostly. Which is embarrassing. But I’ve made peace with it.”
November. A year since the Google partnership, four and a half years since the lecture hall. Dojun sat in his office—he had a private office now, a concession to the reality that a CEO of sixty people occasionally needed to close a door—and opened his journal.
The gap is closing.
Not the gap between Aria and TaskFlow. That gap is stable—we lead in Asia, they lead in the West. Both companies are growing. The market is big enough for two.
The gap I mean is the one between who I am and who people see. The secret gap. The impossible-knowledge gap.
Four and a half years. In that time, I’ve built a company, published papers, won competitions, and made decisions that consistently outperform what a twenty-four-year-old should be capable of. The explanations—prodigy, hard work, good mentorship—have held so far. But they’re straining.
Hana asks less often now. Not because she’s stopped wondering, but because she’s decided to trust the person instead of solving the mystery. Kim Taesik stopped asking a year ago. Seokho never directly asks—he just watches, records, and files the anomalies in his mental database.
But the world is getting smaller. As Aria grows, I become more visible. Interviews, conferences, industry panels. Each one is an opportunity for someone to notice that my knowledge doesn’t match my years. Each one is a chance for the wrong question, the too-specific answer, the slip that unravels everything.
Someday.
Hana said someday better come soon. Kim Taesik said secrets have a half-life. Seokho said data accumulates. They’re all right. The weight is getting heavier, not lighter, and the people carrying it with me deserve to know what they’re carrying.
But not today. Today, Aria 3.0 is the best product we’ve ever built. Today, the team is healthy. Today, Hana sleeps before she designs. Today, my mother’s stall accepts card payments and Mrs. Kang’s grandson is our newest intern.
Today is good. Tomorrow might be the day I finally tell the truth.
But probably not tomorrow either.
He closed the journal. Outside, Seoul’s autumn was deepening—the ginkgo trees along the Gangnam boulevard dropping golden leaves that collected in drifts along the curbs. A street sweeper moved methodically through the leaf piles, restoring order to the chaos of seasonal change.
His phone buzzed. His mother:
Dojun-ah. Mrs. Kang’s grandson showed me how to take a photo with the laptop camera. I took a photo of today’s japchae display. Hana says it’s “peak visual merchandising.” I don’t know what that means but she used three exclamation marks so I assume it’s good. Come Saturday. I’ll make tteokguk early this year because the weather is cold and you’re too thin. Love, Mom.
He smiled. Typed back: I’ll be there. Always.
Always.
That was the promise. That was the code he’d written into the foundation of his second life, deeper than any algorithm, stronger than any architecture.
Always show up. Always tell the truth—eventually. Always put the japchae in the front.
The ginkgo leaves fell. The company grew. The secret waited.
And somewhere between the gold of autumn and the cold of approaching winter, Park Dojun—twenty-four years old, sixty-seven years of memory, four and a half years of second chances—was learning that the hardest line of code wasn’t the one that made things work.
It was the one that kept things from breaking.