Chapter 61: Fourteen Days
Junior spent the first three days of his fourteen-day deadline doing something the board didn’t expect: he wrote code.
Not strategy documents. Not investor presentations. Not the PowerPoint decks that corporate culture demanded as proof that thinking was happening. He sat in his office—Dojun’s old office, which he still couldn’t bring himself to redecorate—and wrote code for eighteen hours a day, fueled by energy drinks and the desperate clarity that comes from having everything on the line.
The code was for Project Lighthouse.
Dojun had started Lighthouse two years before his death—a skunkworks project, off the books, known only to Junior and a team of seven senior engineers. The concept was simple and terrifying: an AI system that could audit its own code for bias, errors, and ethical violations in real time. Not a safety filter bolted onto an existing AI—a fundamental redesign of how artificial intelligence monitored itself.
The industry called it “AI alignment.” Dojun had called it “teaching a god to check its own homework.”
The project was 70% complete when Dojun died. The remaining 30% was the hardest part—the recursive self-evaluation loop that allowed the AI to assess not just its outputs but its reasoning process. It was the kind of problem that had stumped every major AI lab in the world, because it required the system to be simultaneously the judge, the defendant, and the jury.
Dojun had cracked the theoretical framework. The implementation was up to Junior.
On day four, Ha Yeonhee—lead researcher, genius-level systems architect, the woman who three competitors had been trying to poach—walked into Junior’s office at 2 AM and found him asleep on his keyboard, the screen filled with a recursive function that was, despite being written by an unconscious person, remarkably close to correct.
“You’re killing yourself,” she said, setting a cup of coffee on his desk.
Junior lifted his head. The keyboard had left an imprint on his cheek. “I’m close.”
“Close to what? A breakthrough or a cardiac event?”
“Both, probably.” He showed her the screen. “The recursive loop. I think I’ve solved the halting problem.”
Yeonhee’s eyebrows rose. In computer science, the halting problem was famously unsolvable—proven impossible by Alan Turing in 1936. What Junior meant was that he’d found a practical workaround: a probabilistic approximation that could predict, with 99.7% accuracy, whether a self-evaluating AI would reach a stable conclusion or spiral into infinite recursion.
“Show me,” she said.
He showed her. She read the code line by line, her expression shifting from skepticism to interest to the wide-eyed, slightly terrified look of a scientist encountering something genuinely new.
“This works,” she said.
“I know.”
“No, Junior. This works. Not theoretically. Actually. If we implement this, Lighthouse becomes the first commercially viable self-auditing AI system in the world.”
“I know.”
“Every tech company on the planet will need this. Every government regulator will require it. This isn’t a product. It’s the next decade of AI safety.”
“I know.” He rubbed his face. The keyboard imprint was still there. “And Nexion knows too. That’s why they’re offering 40 trillion. Not because Prometheus is worth 40 trillion today. Because with Lighthouse, it’s worth ten times that.”
“They know about Lighthouse?”
“Chen Wei didn’t come to Korea to buy a company. She came to buy this.” He pointed at the screen. “Someone leaked it. Not the code—the concept. Enough for Nexion to understand what we’re building and how far ahead we are.”
Yeonhee was quiet. Then: “Who?”
“I don’t know yet. But I’m going to find out.”
She sat down across from him—in the chair that had been Dojun’s, because Yeonhee was not the kind of person who avoided ghosts. “What do you need?”
“Ten more days. Your team. And absolute secrecy.”
“You have all three.” She stood. At the door, she paused. “Junior.”
“Yeah?”
“Dojun would be proud.”
“Dojun would tell me my variable names are too long.”
“Also that.”
She left. Junior turned back to the screen and kept coding. Nine days left. One impossible problem. And somewhere in the building, someone who had told Nexion about the most valuable technology Prometheus had ever created.