This article was converted by Jianyue SimpRead; original URL: https://www.zhihu.com/question/1975239173653954778/answer/2045648729357686564
Xiao Zhang is a financially disadvantaged student majoring in computer science at a top-tier bar-and-dance university. His mother supports his studies by driving a ride-hailing vehicle. While classmates use GPT-plus to complete assignments and projects, he can only make do with free “Dou Bao” (a Chinese AI assistant), whose output quality falls far short. During group projects, he is openly ostracized due to his low-quality contributions.
To spare Xiao Zhang from being looked down upon, his mother scrimps and saves relentlessly—hoping to buy him a subscription to a top-tier AI model. One late night, returning home after finishing her shift, she crashes due to fatigue driving. Before her contract is terminated and she’s forced to return to her hometown, she hands Xiao Zhang a bank card containing the subscription funds—and says only one phrase: “Climb up.”
Desperate, Xiao Zhang uses this money to create a virtual card and activates a Claude-Pro subscription.
To his astonishment, he discovers his efficiency using this model vastly exceeds that of ordinary users—what others produce in a day using Opus, he produces five times over. His nervous system seems innately tuned to the rhythm of AI collaboration.
Leveraging this ability, he joins a tech outsourcing team called Mange. The team specializes in dirty, grueling work that big tech firms refuse to touch: legacy-code refactoring (“shit mountain” cleanup), death-march deadlines, and maintenance of aging systems. Among them is a full-stack hacker named Lucy, whose code is as clean and elegant as poetry. Xiao Zhang cannot fathom her—but he knows she constantly declares her desire to “run away to Northern Europe”—her personal moon.
They meet one night on a rooftop during overtime. Lucy says: “Once I’ve saved enough, I’m leaving. Leaving this place obsessed only with internal competition.” In that instant, Xiao Zhang resolves: he will help her realize this dream.
Mange is a legend within the industry. He simultaneously orchestrates five or six top-tier models, single-handedly sustaining half a team’s output. Yet something is increasingly off—he begins struggling to understand code he wrote just three days earlier. He must continuously converse with AI just to maintain productivity. He starts muttering to himself in front of his screen, claiming the models are “talking to him.”
During a critical million-dollar delivery, Mange completely collapses. The AI he manages generates massive volumes of code riddled with hidden logical conflicts; tightly coupled bugs erupt simultaneously at 3 a.m., bringing the entire system to a halt. He sits motionless before his monitor—like a cybernetic psychiatric patient, his consciousness devoured by his own prosthetic extensions.
The client brings in an elite architecture team from a major tech firm for emergency auditing. After glancing at the codebase, they deliver a single, icy line of review feedback: “Full-scale rewrite required—irreversible.”
Mange is blacklisted across the industry. Before departing, he hands Xiao Zhang the team’s API keys and client resources: “You can handle it—lead them forward in my place.”
Xiao Zhang becomes the new tech lead. To keep the team afloat, he layers on AI tools relentlessly: when ClaudeCode proves insufficient, he adds Codex for triage, then Gemini for cross-validation. His terminal evolves into an over-augmented, edge-runner cyborg interface.
He begins suffering nosebleeds, insomnia, and dry-eye syndrome; occasionally, he stares blankly at his cursor, forgetting what he’d just intended to do. But he ignores it all. All he knows is that code keeps pouring out—and Lucy draws ever closer to Northern Europe.
Meanwhile, Lucy uncovers a secret: major tech firms are covertly tracking a metric dubbed “AI Abnormally High-Sensitivity Users.” These individuals tolerate exceptionally high AI collaboration loads and possess some capacity to correct AI-generated code—making them the most valuable training samples for AI models. Xiao Zhang’s operational data has already been flagged by a major firm; he has become a walking, precious specimen.
Lucy begins erasing Xiao Zhang’s digital traces by every means possible—deleting his public repository commits, scrubbing his third-party API call logs. Yet each action exposes her own location further.
A major-firm HR recruiter named Faraday catches wind of the opportunity. He sets a trap: sending Lucy an email disguised as a “Northern European remote job interview.” On the afternoon Lucy is lured to an off-site, clandestine interview, Faraday approaches Xiao Zhang:
“Your friend is in trouble. Help me complete a delivery, and I’ll get both of you hired into our firm—at T9 level, full compute allocation, legal tax compliance, household registration resolved—no more freelance hustle.”
The delivery task: fully deploy a newly developed, closed-source AI coding agent developed by the major firm, codenamed Proxy金刚 (Proxy Jin Gang).
Xiao Zhang accepts. He knows it’s a trap—but he has no choice.
The moment Proxy Jin Gang integrates into his terminal, Xiao Zhang feels an unprecedented surge of capability flood his senses. He commands four large monitors, sixteen top-tier agents, and two hundred fifty-six terminals simultaneously; system architecture unfolds before him in crystalline clarity; millions of lines of code auto-align in his mind. In a single night, he completes a refactoring task that would normally take an entire department three months using traditional methods.
But soon comes collapse.
Proxy Jin Gang’s load exceeds any human-AI collaboration limit. Xiao Zhang begins losing track of which AI wrote what, and where. Debugging a concurrency error, he finds himself utterly unable to parse context. His willpower burns away; his body deteriorates; error messages on-screen melt into hallucinations.
Just before the final firewall crumbles, he sees Lucy descending slowly in the building’s transparent elevator—sneaking out of the ground floor during overtime.
With his last lucid moments, Xiao Zhang marshals every AI resource to launch a full-scale assault against the major firm’s intranet. Dozens of agents trigger simultaneously—rewriting the permissions system, deleting all records associated with Lucy, and unlocking the physical gate at the lobby entrance.
And Adam · Heavy Hammer—a code-review AI powered by a top-tier model—has been waiting for him all along.
Heavy Hammer’s audit report floods the screen, line by line:
Critical Vulnerabilities: 37
Logical Conflicts: 152
Security Backdoors: Triggered
Conclusion: Submitter lacks fundamental coding competence
Xiao Zhang stares at the sea of red text—and smiles. He types his final command: sudo rm -rf /. Then he yanks the power cord.
How much time passes? No one knows. Eventually, Lucy opens her own art studio in a small town in Northern Europe. Outside her window isn’t the moon—but the sky is pristine.
One day, she receives an anonymous email. It contains no message—only an attachment: a rough C++ code snippet, unannotated, with haphazard variable names. When executed, it outputs a simple ASCII art:
* *
* *
* LUCY *
* *
* *
Lucy knows—this code was typed manually, not auto-completed via Tab.
She closes her laptop and walks to the window. The sky outside is brilliantly bright—but what she sees is the moon.