Chapter 20: Five Hundred
The first real user of Bridge was a third-year economics student named Kang Jihye, who signed up for the beta because her roommate told her it was “like having a secretary who lives in your laptop,” and who sent the following feedback email eleven minutes after installation:
Dear Bridge team, this thing just grouped my Macroeconomics homework emails with my part-time job schedule and told me I have a conflict on Thursday. I have been aware of this conflict for three weeks and have been ignoring it. I did not need a computer to judge me. That said, 10/10, will keep using. — Kang Jihye, Econ ’08
Dojun read the email three times, then forwarded it to Hana and Minjae with the subject line: “Our first user review. We’re being judged by a computer-judgmental economics student.”
Hana replied in four seconds: Frame it. This is going on the office wall next to the founding handshake photo.
Minjae replied in twelve seconds: She found a conflict I missed in the test data. Our beta users are better testers than I am. I’m simultaneously proud and humiliated.
The campus beta had launched the previous Monday—October 2nd, 2006—after two weeks of frantic preparation that had consumed every waking hour that wasn’t already claimed by classes, research, and Saturday market visits. The launch process had been deceptively simple: Minjae set up a download page on the SNU intranet, Hana designed a poster campaign (“Your laptop is a mess. Bridge can help.”), and Dojun spent three sleepless nights stress-testing the server that would handle five hundred simultaneous installations.
The server was a rented machine in a Guro-gu data center that cost 200,000 won per month—their first real business expense, paid from the Choi Eunji investment. It was modest by any standard: a single-core processor, 2GB of RAM, and a hard drive that made concerning clicking sounds when the load exceeded 70%. But it was theirs, and it worked, and when the first installation request hit the server at 9:03 AM on launch day, all three of them were sitting in the twelve-square-meter office watching the log file scroll.
“One,” Minjae whispered.
“One,” Hana confirmed.
“Don’t jinx it,” Dojun said.
By noon, they had fifty installations. By 6 PM, two hundred. By midnight, the counter read 347, and Minjae was asleep on the office floor using his backpack as a pillow.
“We should go home,” Hana said, but she was still watching the counter, unable to look away. Each new number was a person—a real human being who had chosen to trust their digital life to a product built by three sophomores in a closet-sized office.
“Three hundred and forty-seven people are using Bridge right now,” she said. “Three hundred and forty-seven people we’ve never met decided that our idea was worth trying. That’s… I don’t have a word for what that feels like.”
“Terrifying,” Dojun said.
“Terrifying in a good way.”
“Is there a good way to be terrified?”
“The way you feel before jumping off a diving board. You know the water is there. You know you’ll be fine. But your body doesn’t believe your brain, and the gap between knowing and feeling is where the terror lives.” She tapped the counter. “Three hundred and forty-eight.”
By Friday, they hit five hundred. The beta was full.
The feedback came in waves.
The first wave was enthusiasm—Jihye-level excitement from students who had never experienced automatic task organization and found it revelatory. “Why doesn’t every computer do this?” was the most common question, which Hana took as the highest possible compliment.
The second wave was complaints. Not about the concept—nobody questioned the value of task-centric organization—but about the edges. The task detection misclassified emails with ambiguous subjects. The calendar parser struggled with recurring events that had been modified individually. The file system scanner was too aggressive, indexing temporary downloads and browser cache alongside actual documents.
“Every bug report is a gift,” Hana said during their daily standup—a fifteen-minute meeting she had instituted, standing in the office because sitting implied complacency. “Each one tells us where the gap between our model of users and actual users exists.”
“There are forty-seven bugs in the tracker,” Minjae said, scrolling through his laptop. “Seventeen are UI issues—I can fix those this week. Twenty are task detection errors—that’s Dojun. Ten are data pipeline problems—that’s me.”
“Prioritize by impact,” Dojun said. “Which bugs affect the most users?”
“The recurring calendar issue. Sixty-three beta users have recurring events, and twenty-two of them reported incorrect grouping. That’s a 35% error rate for that specific feature.”
“I’ll fix that today.” Dojun pulled up the code on his laptop. The recurring event parser was a known weak point—he had written it hastily during the pre-Showcase crunch, and the logic for handling modified instances of recurring events was, to put it charitably, optimistic.
“While you’re in there,” Hana said, “can you look at the task naming algorithm? Several users complained that Bridge names their tasks things like ‘Email Cluster 7’ instead of something human-readable. Nobody wants to organize their life around ‘Email Cluster 7.'”
“The naming algorithm uses the most frequent keyword in the cluster. If the keyword is generic—like ‘meeting’ or ‘project’—the name is generic.”
“Then make it smarter. Use the sender name, or the calendar event title, or the most specific noun in the subject line. ‘Professor Kim’s Research Meeting’ is better than ‘Email Cluster 7.'”
“That’s a natural language processing problem.”
“It’s a making-things-not-ugly problem. Which is my department. But it needs your code to implement. Can you do it by Wednesday?”
“I can try.”
“Don’t try. Do. Dr. Yoon taught us both that lesson.” She capped her marker. “Standup over. Back to work.”
The rhythm of the beta became the rhythm of their lives. Code in the morning, fix bugs in the afternoon, analyze feedback in the evening. Minjae built a dashboard that tracked user engagement metrics—daily active users, task creation rates, error frequencies—and projected it on the office wall using a secondhand projector he had bought from a graduating senior for thirty thousand won.
“Our DAU is 312 out of 500 installations,” he reported after the first full week. “That’s a 62% daily retention rate. For a beta with no marketing and no support team, that’s very strong.”
“What’s the industry standard?” Hana asked.
“For consumer software? 20-30% is normal. 40% is good. 62% is—” He paused, checking his numbers. “Unusual.”
“Unusual good or unusual suspicious?”
“Unusual ‘our users actually like the product’ good. The feedback confirms it—the satisfaction scores average 4.2 out of 5. The main complaints are bugs, not concept. Nobody is saying ‘I don’t need this.’ They’re saying ‘I need this, but fix the calendar thing.'”
Dojun listened to the numbers and felt the particular warmth of validation—not personal validation, but product validation. Bridge was working. Not perfectly, not completely, but working in the way that mattered: people were using it, finding value in it, and asking for more of it.
In his previous life, Prometheus Labs had launched with much larger numbers—millions of users from day one, thanks to enterprise partnerships and a massive marketing budget. But those numbers had always felt abstract. Five hundred students on a campus, sending emails about bugs and conflicts and task names—that felt real. Each user had a face, a schedule, a life that Bridge was (imperfectly, incrementally) making easier.
“The next priority is the learning module,” he said during the Friday standup. “We cut it for the Showcase, but the users are generating exactly the kind of correction data we need for training. Every time a user moves a task from one group to another, or renames a cluster, or marks a detection as incorrect—that’s a labeled training example.”
“How many labeled examples do we have?” Hana asked.
“After one week? About 2,300. We need at least 10,000 for a meaningful model. At the current rate, we’ll have that in a month.”
“Then the learning module launches in November?”
“If I can get the compressed state representation working. The memory issue from before hasn’t gone away—I just delayed it.” He rubbed his eyes. “But I have an idea. Instead of storing full correction contexts, I can use a vector embedding—a compressed numerical representation of each correction. Smaller footprint, faster lookup, same information content.”
“Vector embedding,” Minjae repeated. “Is that the thing from the machine learning textbook that made me want to change majors?”
“It’s simpler than the textbook makes it sound. Think of it as translating each correction from a paragraph into a zip file.”
“That I understand. Zip files are my friends.” He made a note. “Okay. Learning module, November. Compressed vectors. I’ll expand the data pipeline to collect and store the training examples.”
“And I’ll redesign the settings page to let users opt in or out of the learning feature,” Hana said. “Privacy matters. We don’t train on anyone’s data without their explicit consent.”
“Agreed,” Dojun said. And meant it more deeply than they could know. In his previous life, Prometheus Labs had been accused of opaque data practices—training AI models on user data without clear consent mechanisms. The scandal had been small but damaging, a crack in the company’s reputation that had widened over years. He would not repeat that mistake.
October settled into a rhythm that felt, for the first time in Dojun’s second life, like normalcy. Not the frantic normalcy of constant crisis management, but the steady, productive normalcy of people doing meaningful work together.
Mornings: classes. Dojun was now taking graduate-level courses—Advanced Computer Architecture with a visiting professor from MIT, Kim Taesik’s Embedded Systems seminar, and Dr. Yoon’s Advanced Graph Theory. The material was familiar but the teaching was new, and he found himself genuinely learning things about pedagogy—how different professors structured explanations, how questions could illuminate or obscure, how the same concept could be revolutionary or obvious depending on the context in which it was presented.
He was also, quietly, becoming a better teacher himself. Yuri’s math grades had gone from failing to B+, and Uncle Sangchul had increased his tutoring fee to forty thousand won per session—”Because you’re worth it, and because my wife’s salon had a good month.” Dojun used the money to buy his mother a new cutting board and a set of premium containers for the banchan stall.
Afternoons: Bridge. The bug count was dropping steadily—from forty-seven to thirty-one to eighteen to a manageable seven. User satisfaction crept up from 4.2 to 4.4 to 4.5. The recurring calendar fix alone generated a wave of grateful feedback that Minjae printed and taped to the office wall under the heading “PROOF THAT MINJAE’S SUFFERING WAS NOT IN VAIN.”
Evenings: life. This was the new part. In his previous life, evenings had been absorbed by work—more code, more meetings, more optimization of an already-optimized machine. Here, he was deliberately reclaiming them. Tuesday evenings were reserved for phone calls with Seokho, who was working on a research paper about distributed consensus algorithms and used Dojun as a sounding board. (“Your criticism is always annoyingly precise,” Seokho told him. “It’s the only criticism I don’t ignore.”)
Thursday evenings were jjigae dinners with Hana. These had evolved from project meetings into something that neither of them labeled but both recognized—a space where work blurred into personal, where algorithms gave way to stories, where two people who were building a company together were also, carefully, building something else.
“Tell me something about yourself that isn’t on your resume,” Hana said one Thursday, over stone pots of sundubu jjigae—a variation from their usual kimchi, because the ajumma had declared it “sundubu season” with the authoritative finality of someone who controlled the menu and did not accept appeals.
“I’m terrible at cooking,” Dojun said.
“Everyone knows that. Something real.”
“I have a fear of heights.”
“Heights? You don’t seem like someone who’s afraid of anything.”
“Everyone’s afraid of something. Heights, for me. And—” He paused. “Losing people. I’m afraid of losing people I care about. Not physically—emotionally. Losing their trust, their respect, their willingness to stay.”
Hana set down her spoon. “Has that happened to you before? Losing someone?”
“Yes.”
“Who?”
“Someone I should have held on to. Someone I let go because I was too focused on building things to notice that the most important thing I’d built was the relationship itself.” He looked at his sundubu. “I don’t talk about it.”
“You just did.”
“I did.” He met her eyes. “I’m trying to be less edited. Like you asked.”
“You’re getting better at it.” Her voice was soft. “Slowly. Like a software update that improves performance by 2% each iteration. Not dramatic, but measurable.”
“You’re comparing my emotional growth to a software update.”
“I’m a designer working with an engineer. Metaphors adapt to the audience.” She reached across the table and covered his hand with hers. “I’m not going anywhere, Dojun. Whatever you’re afraid of losing—I’m not going anywhere.”
He turned his hand over and held hers. The ajumma, passing with a tray of banchan, glanced at them, made a small “tsk” sound that might have been disapproval or might have been approval, and continued without comment.
“Saturdays at the market,” Hana said. “Tuesdays with Seokho. Thursdays with me. You’re building a schedule that has people in it, not just tasks. Bridge should take notes.”
“Bridge doesn’t organize relationships. Only work.”
“Maybe it should. Maybe version three should understand that the most important tasks in someone’s life aren’t emails or meetings. They’re the people they keep choosing to show up for.”
She said it casually, as a product idea, but the truth in it hit Dojun with the force of something he had spent sixty-three years learning and was only now, at twenty, beginning to understand.
The most important tasks weren’t tasks at all. They were people.
“Write that down,” he said. “Bridge v3: relationship-aware task organization.”
“I already did.” She held up her notebook, where she had sketched a wireframe of a new Bridge feature—a “People” layer that tracked not tasks but connections, showing which relationships needed attention, which conversations were overdue, which people you hadn’t reached out to in too long.
“That’s genius,” he said.
“That’s empathy translated into interface design.” She closed the notebook. “Eat your sundubu. It’s getting cold and the ajumma will blame me.”
They ate. The sundubu was perfect—silky, spicy, with a depth of flavor that came from the ajumma’s twenty years of practice and her refusal to reveal her recipe to anyone, including her own daughter.
Outside, October was turning toward November. The ginkgo trees had gone fully gold, lining the streets with a canopy of light that made Seoul look, for a few weeks each year, like a city built inside a painting.
Five hundred users. A funded company. A paper under review. A rival who was becoming a friend. A partner who was becoming something more. A mother who was proud.
And a twenty-year-old man with the memories of sixty-three years, sitting in a basement restaurant, holding hands over sundubu jjigae, learning that the most invisible technology of all was the kind that helped you remember to show up for the people you loved.
He was, against all odds, happy.
Not the desperate, fragile happiness of someone clinging to a second chance. But the quiet, sturdy happiness of someone who was, for the first time, building a life that deserved the word.