이 포스팅은 쿠팡 파트너스 활동의 일환으로, 이에 따른 일정액의 수수료를 제공받습니다.
Chapter 39: The Clone
The competitor appeared without warning, the way competitors always did—not in a dramatic announcement but in a TechCrunch article that Minjae found at 6 AM while checking the morning tech news over his first coffee.
“TaskFlow Raises $50M Series A to Build the ‘Aria of the West’ — Former Google Engineers Promise AI-Powered Task Management”
He forwarded the article to the team Slack channel with a single word: Problem.
By 7 AM, everyone had read it. By 8 AM, the office was full—an hour before the normal start time—and the atmosphere had the particular electric tension of a team that had just discovered it was no longer alone in its market.
TaskFlow was based in San Francisco. Founded by three ex-Google engineers who had, according to the article, “studied Aria’s architecture through its published patents and academic papers, and built a next-generation version using state-of-the-art machine learning.” They had $50 million in funding—fifty times Aria’s Series B—from Sequoia Capital, the most prestigious VC firm in Silicon Valley.
“They literally call themselves ‘the Aria of the West,'” Taeyoung said, reading the article aloud. “They’re not even pretending to be original. They’re a clone. A very well-funded clone.”
“They’re not a clone,” Dojun said. He had been reading the article with the particular stillness of a man who had seen this exact scenario before. “They’re a competitor. There’s a difference. A clone copies the product. A competitor takes the concept and builds something new on top of it.”
“Their product screenshots look identical to ours,” Soojin said, pulling up TaskFlow’s website on the conference room screen. “Task cards. Context toggles. Learning module. Even the color palette is similar.”
“The concept is similar. The execution will be different. They’re building on machine learning—specifically, large language models for task detection. That’s a fundamentally different approach than our keyword-and-temporal-correlation engine.” He paused. “And it’s potentially better.”
The room went very quiet.
“Better?” Hana said.
“LLM-based task detection can understand context in ways our heuristic engine can’t. It can read an email and understand that ‘Can we push the deadline to next week?’ means a task needs to be rescheduled—not because of keywords, but because of semantic understanding. Our engine would miss that. Theirs won’t.”
“So we’re outgunned.”
“Technologically, in task detection specifically, yes. They have fifty million dollars, a team of Google ML engineers, and access to compute resources we can’t match.” He looked at the team—forty-five faces, some scared, some determined, all looking to him for direction. “But technology isn’t the only dimension of competition. And it’s usually not the most important one.”
“What’s the most important one?” Minjae asked.
“Design. Distribution. Trust. We have a hundred thirty-four thousand users who chose Aria because it works the way they think. TaskFlow has zero users and a demo. We have Sony, Google Workspace integration, and twenty-eight hundred enterprise clients. They have a pitch deck and a TechCrunch article.” He turned to the whiteboard. “The question isn’t whether TaskFlow is a threat. It is. The question is what we do about it.”
The strategy session lasted four hours—the longest continuous meeting in Aria’s history. By the end, the whiteboard was covered in three columns: DEFEND, DIFFERENTIATE, ACCELERATE.
DEFEND: Lock in existing enterprise clients with long-term contracts and deeper integrations. Sony, Google, and the Korean market were defensible because Aria had first-mover advantage and localized relationships that TaskFlow couldn’t replicate from San Francisco.
DIFFERENTIATE: Double down on what made Aria different—the design philosophy, the human-centric approach, the invisible technology ethos. TaskFlow would compete on AI intelligence. Aria would compete on AI empathy. “A smarter assistant isn’t always a better assistant,” Hana said. “A better assistant understands you. Our learning module doesn’t just detect tasks—it adapts to how you work. TaskFlow’s LLM will be powerful, but it’ll be generic. We’re personal.”
ACCELERATE: Invest in AI capabilities. Not to match TaskFlow—they had fifty million dollars and Google-level talent, and trying to match them dollar-for-dollar would be suicidal. Instead, build a hybrid approach: use Aria’s existing learning module as the personalization layer, and integrate third-party AI services for the heavy computational lifting.
“Third-party AI?” Taeyoung asked. “Like what?”
“OpenAI just launched their API,” Dojun said. “It’s expensive, but the language understanding capabilities are beyond anything we could build in-house. If we integrate OpenAI’s models as an optional layer beneath our learning engine, we get LLM-level task understanding without building the LLM ourselves.”
“You want to use our competitor’s underlying technology—AI models—to differentiate against our competitor’s product?”
“The models aren’t TaskFlow’s. They’re OpenAI’s, available to anyone with an API key. TaskFlow built their own models because they have fifty million dollars. We’re smarter about it—we use the best available models and focus our engineering on the personalization and design layers that TaskFlow can’t copy.” He looked at Hana. “The integration needs to be invisible. The user shouldn’t know or care whether task detection uses our heuristic engine, the learning module, or an LLM. They should just see tasks that make sense.”
“Invisible AI,” Hana said. “I can design that.”
“I know you can.”
The weeks that followed were the most intense in Aria’s history. Not the frantic, sleep-deprived intensity of the Showcase preparation or the App Store launch—those had been sprints. This was a sustained marathon of strategic execution, every team member working at the edge of their capacity.
Taeyoung led the AI integration—connecting OpenAI’s API to Aria’s task detection pipeline, building a routing layer that sent simple tasks to the heuristic engine and complex tasks to the LLM, and optimizing the latency so that users couldn’t tell the difference. The first prototype was ready in three weeks. The accuracy jumped from 93% to 97%.
“Four percent sounds small,” he reported at the standup. “But it’s the hard four percent—the ambiguous emails, the multi-step tasks, the context-dependent requests that our heuristic engine couldn’t handle. The LLM gets them right almost every time.”
Jiyoung led the client retention campaign—calling every enterprise customer personally, offering contract extensions with locked-in pricing, and positioning Aria’s AI upgrade as a free benefit of existing subscriptions. “We’re not panicking,” she told each client. “We’re investing. The upgrade you’re getting for free would cost a new competitor’s customer fifty dollars a month.”
Hana led the design response—a comprehensive UI refresh that she called “Aria 3.0: The Conversation.” The new interface treated every interaction as a dialogue between the user and the system. Instead of static task cards, Aria now presented dynamic suggestions: “It looks like you’re preparing for the budget meeting. Here are the relevant emails, the latest spreadsheet, and a reminder that the meeting was moved to 3 PM.” The tone was warm, the timing was precise, and the suggestions were—thanks to the LLM integration—eerily accurate.
“It’s like having a colleague who read all your email and actually remembers what’s important,” one beta tester wrote. “Except it doesn’t steal your lunch from the office fridge.”
The 3.0 launch was scheduled for June 2010—three months after TaskFlow’s public beta. The timing was deliberate: let TaskFlow establish the category, then enter with a product that was demonstrably better in the dimensions that mattered.
But the intensity took its toll.
Dojun noticed the signs first, because he had seen them before.
Hana was working later than anyone. Not unusual—she had always been the last to leave, the one who refined a pixel at midnight because it wasn’t right. But the nature of the work had changed. She wasn’t refining anymore. She was redoing. Designs that she had approved on Monday were scrapped on Tuesday. Interface decisions that had been finalized were reopened. The 3.0 mockups went through twenty-seven revisions—each one, she insisted, “not quite there yet.”
“The conversation metaphor isn’t landing,” she said during a Thursday review, her eyes red-rimmed from a screen glare she wouldn’t acknowledge was actually exhaustion. “The suggestions feel pushy. Like the system is telling you what to do instead of helping you do what you want. The tone is wrong.”
“The user testing scores are 4.7 out of 5,” Taeyoung said gently. “The highest we’ve ever had.”
“4.7 isn’t 5. There’s a gap. I can feel the gap.” She pulled up the mockup for the third time. “See this transition? When the suggestion appears, there’s a 200-millisecond delay that feels like hesitation. It should feel like anticipation. The difference is the easing curve—”
“Hana.” Dojun’s voice was quiet but firm. “Can we talk? Privately?”
She looked at him. For a moment, the exhaustion broke through the surface—a flash of vulnerability that she immediately suppressed. “After the review.”
“Now.”
They went to The Silence. Dojun closed the door.
“How many hours did you sleep last night?” he asked.
“Enough.”
“How many?”
“Three. Maybe four.”
“And the night before?”
“I don’t remember.” She sat down in the room’s single chair. “I know what you’re going to say.”
“What am I going to say?”
“That I’m pushing too hard. That the design is good enough. That I need to sleep. That TaskFlow has fifty million dollars and we can’t match them by working ourselves to death.” She rubbed her eyes. “I know all of that. Intellectually, I know.”
“But?”
“But this is the first time someone has tried to copy what we built. And not just copy it—improve on it. With better technology, more money, more engineers. TaskFlow’s demo is… good, Dojun. I saw it. The LLM-based detection is smooth. The interface is clean. It’s not as good as Aria, but it’s close. And ‘close’ from a $50 million company with Sequoia behind them is terrifying.”
“You’re afraid.”
“I’m terrified. I built Aria’s design language from a sketchbook in a basement restaurant. Every interaction, every animation, every pixel—it came from here.” She tapped her chest. “And now someone with a hundred times our budget is saying ‘we can do that too.’ And maybe they can. Maybe they’ll do it better. And then what? What was it all for?”
Dojun sat down on the floor—the room only had one chair—and looked up at her.
“I need to tell you something,” he said. “About the person I mentioned. The one I lost.”
“The mysterious person from the other context.”
“She was a designer. Like you. Brilliant, driven, uncompromising. She built something beautiful—a company’s entire design identity, from the ground up. And when a competitor appeared—a better-funded, more technically sophisticated competitor—she did exactly what you’re doing now. Worked harder. Slept less. Redid everything. Pushed herself past the point of function because she believed that perfection was the only defense against being replaced.”
“What happened?”
“She burned out. Not in a dramatic collapse—in a slow erosion. The joy went first. Then the creativity. Then the trust. She stopped believing that ‘good enough’ was ever enough, and she stopped believing that anyone—including her partner—understood what she was going through. By the time he realized what was happening, she was already gone.”
Hana was very still. “She left?”
“She left. Not because the partner didn’t love her. Because the partner didn’t stop her when she was destroying herself. He let her burn because the company needed her fire. And when the fire went out, there was nothing left.”
The room was silent. The ventilation hummed. Somewhere in the office, a phone rang and went to voicemail.
“I’m not going to let that happen to you,” Dojun said. “Not because Aria needs you—though it does. Because I need you. And I need you whole, not burned to ash in the name of a product that will exist long after both of us.”
“You’re telling me to stop.”
“I’m telling you to rest. The 3.0 design is at 4.7 out of 5. That’s extraordinary. Ship it. Let the users tell us what needs to change. Let the market do the work that you’re killing yourself trying to do alone.”
“TaskFlow—”
“TaskFlow has money. We have you. But only if you’re here. Only if you’re healthy. Only if you remember that the woman who designed Aria’s interface in a basement restaurant with a pen and a sketchbook didn’t need $50 million or an LLM. She needed sleep, and jjigae, and a partner who told her the truth.”
Hana’s eyes filled. Not dramatically—the way water fills a cup, slowly, inevitably, until it spills over the edge.
“I’m so tired,” she whispered.
“I know.”
“I’ve been tired for weeks and I didn’t want to say it because everyone is working so hard and I’m the CDO and I’m supposed to be the one who sets the standard—”
“The standard includes being human. You set that standard three years ago, when you made me sign a schedule that said ‘Saturdays are non-negotiable.’ Remember?”
“I remember.”
“Then let me hold you to the same standard. Go home. Sleep. Skip tomorrow. The 3.0 launch can wait three days.”
“Three days?”
“Three days. The world will not end. TaskFlow will not conquer Asia in seventy-two hours. And when you come back, you’ll see the design with fresh eyes, and you’ll know—from certainty, not from exhaustion—whether it’s ready.”
She wiped her eyes. Took a breath. Nodded.
“Three days,” she said. “But I’m taking the mockups home. Just to look at. Not to work on.”
“You’re going to work on them.”
“Probably. But I’ll do it in pajamas. On the couch. With tea. That counts as resting.”
“For a designer, that counts as a spa day.”
She laughed—watery, tired, real. “Walk me to the subway?”
“I’ll walk you home.”
“That’s forty minutes.”
“Then we’ll have forty minutes to talk about something that isn’t TaskFlow, Aria, or easing curves.”
“What’s left?”
“Your grandmother’s rice cake recipes. My mother’s japchae secrets. Whether the jjigae place ajumma has a first name or was born with the title.” He stood and offered his hand. “Come on. The office can survive without both of us for one evening.”
She took his hand. They left The Silence, walked past the team without explanation (Minjae gave a knowing nod; Taeyoung pretended not to notice; Hyunwoo waved), and stepped out into the Gangnam evening.
The walk home was forty-three minutes. They talked about rice cakes, and japchae, and whether the ajumma’s first name was “Auntie” (Hana’s theory) or “Jjigae” (Dojun’s, offered for comedy). They did not talk about TaskFlow, or 3.0, or the particular terror of watching someone try to copy the thing you loved most in the world.
And when they reached Hana’s apartment—a studio in Seocho-dong that she had decorated with her own artwork and a truly excessive number of plants—Dojun kissed her forehead and said: “Rest. I’ll handle tomorrow.”
“You can’t handle design decisions.”
“I can handle postponing them. That’s a different skill.”
“A crucial one.” She smiled. “Good night, partner.”
“Good night, partner.”
He walked back to Gangnam alone, through streets that smelled of grilled meat and exhaust and the particular nighttime sweetness of a city that never fully slept. His phone buzzed—the team Slack, still active at 10 PM, still building, still fighting the invisible war against a competitor that had money and talent and everything except the one thing that made Aria what it was.
A designer who cared enough to burn. And a partner who cared enough to stop her.
The 3.0 launch could wait. The people couldn’t.