Zhao Wenyuan sat in his office in Jingan District, with two screens in front of him. CodeSafe's backend was on the left, the forum on the right, and a cup of iced-over Americano sat between them.
Ever since the "Laid-Off Programmer" blog and videos had exploded in popularity, CodeSafe's average daily new registrations had fallen from 150 to just over 90. The trend looked ugly.
The direction of community discussion had changed as well. Two weeks ago, people were still asking, "CodeSafe vs BugKiller, which is better?" Now the hottest thread on the front page was titled, "A Laid-Off Employee's Product Crushes a Stanford PhD."
Technical comparisons had become story comparisons.
And when it came to stories, Zhao Wenyuan had no chance of winning.
He recalled Wang Zhiyuan's words. "No rush. But remember, opportunities don't wait."
Three days had passed, and he still hadn't replied.
Access to 150,000 enterprise clients and Dingsheng Cloud's brand endorsement sounded like a highway straight to the sky.
But Zhao Wenyuan had seen too many endings like this in Silicon Valley. A small company took a major corporation's distribution channels. In the first year, it was a partner. In the second, a core supplier. By the third, the founder's name had vanished from the official website.
He had quit Google and returned home to start a business, not to work for another giant corporation.
Zhao Wenyuan switched to CodeSafe's public beta data dashboard.
Two weeks into the beta, user approval stood at 82%. Project detection accuracy led the industry benchmark, and three mid-sized companies were discussing procurement.
The data wasn't bad. What it lacked was a narrative.
He opened his saved beta feedback and scrolled for five minutes. One frontend team said, "CodeSafe caught two P0s before deployment." An independent developer said, "For the first time, code review doesn't feel like a meaningless formality."
It was a good product.
Zhao Wenyuan closed the chat logs and opened the social media post editor.
He typed.
Then deleted it.
He typed again.
Then deleted it again.
May 11, 2:00 p.m.
Zhao Wenyuan's post appeared on the tech forum, Twitter, and Weibo. The exact same words, unchanged.
"NullPointer once said in Detection Rate Is Everything, 'You'll know whether a tool is good after running it once.'
I agree.
So I propose an open, transparent, community-supervised head-to-head evaluation: CodeSafe vs BugKiller. Same codebases, same evaluation criteria, livestreamed from start to finish, with all data made public.
Let the products speak!"
The tech community immediately exploded.
"Holy shit, Dr. Zhao is challenging NullPointer one-on-one."
"The best showdown in tech this year. No contest."
"Independent developer vs funded team, laid-off P7 vs Stanford PhD."
"Finally, this is a proper duel."
"His pride won't let him surrender. Respect."
The post surged to number one on the trending list within twenty minutes.
When Su Niannian sent over a screenshot, Han Luyi was eating Old Zhou's new chicken cutlet rice at Bug Café. He set down his chopsticks and opened it.
"Did you see it? Did you see it? Did you see it?"
"He used your own words to issue the challenge."
"He delivered himself right to your door."
Han Luyi picked up his chopsticks, grabbed a piece of chicken cutlet, and typed with his left hand.
"Accept."
"Just one word?"
"If he wants to compete, then we compete."
An editor from 36-Degree Technology contacted Han Luyi and volunteered to host the livestreamed evaluation.
Han Luyi agreed, and Zhao Wenyuan's side agreed as well.
Someone spammed the comments. "36 Degrees is going to make a killing off the traffic from hosting this."
May 12. The community voted to select the test projects.
Using the NullPointer account, Han Luyi posted a suggestion:
"For fairness, the test plan should cover codebases of different sizes. Users' projects come in all scales, after all. I suggest three rounds: small projects, around 30,000 lines; medium projects, 100,000 to 200,000 lines; and enterprise-scale large projects, over 300,000 lines. Same batch of codebases, same evaluation criteria."
The comments below were filled with praise.
"Support tiered testing. This is the scientific approach."
"NullPointer's thinking is on another level."
"This proposal already gives him the upper hand. If you propose a fair plan, the other side has no choice but to agree."
Zhao Wenyuan replied, "Completely agreed. Covering different scales is the only way to make it convincing."
Someone raised a crucial question below the post. "Both sides might have pretrained on public projects. What would that even prove? They need unseen code for it to be fair."
Old Zhao from 36-Degree Technology stepped in to coordinate. Three corporate sponsors each donated a private project that was about to be open-sourced, none of which either side could possibly have encountered before. The code would be sealed by a third-party notary before the livestream and unsealed only when the broadcast began.
The final plan was set—
Round 1: A startup's Go microservice framework, approximately 30,000 lines.
Round 2: A mid-sized SaaS company's Python data processing platform, approximately 120,000 lines.
Round 3: A publicly listed company's Java enterprise core system, approximately 350,000 lines.
36-Degree Technology released an announcement that night.
"Public Livestreamed Head-to-Head Evaluation of AI Code Review Tools, May 14 at 7:00 p.m. The three testing codebases have been donated by corporate sponsors and have never been accessed by either side. They will be sealed by a third-party notary and unsealed during the livestream. Real-time data will be public, with community supervision throughout."
May 13. The number of reservations exceeded 30,000.
The tech forum opened a dedicated thread: "CodeSafe vs BugKiller, who are you betting on?"
The poll results were delicate: 55% bet on BugKiller, 45% on CodeSafe. The gap wasn't wide.
"BugKiller has a better community reputation, but CodeSafe has a full team, enterprise solutions, and a PhD endorsement."
"NullPointer is taking on fifteen people alone? Are you serious?"
"Don't focus on team size. Last time, a user said CodeSafe missed three bugs."
"One-off case."
"Go read the comments. He isn't the only one."
More than a dozen tech media outlets published analytical articles predicting the outcome from every angle. Ma Xiaofei also wanted to make a prediction video, but Han Luyi stopped him with a single message.
"Don't post anything before the match. I just proposed a fair testing plan, and if you immediately release a prediction video, the top comment will be, 'NullPointer's team is manipulating public opinion before the match.'"
Ma Xiaofei replied, "So I just hold it in?"
Han Luyi said, "Hold it in. Once the results are out, edit the highlights. That traffic will be ten times what a prediction video gets."
Ma Xiaofei nodded.
"Good point. I'll prepare the script first."
May 14.
Su Niannian sat at her workstation at Beaver Technology, staring at a blank weekly operations report template. She had been looking at it for twenty minutes.
On her first day back from annual leave, the newly appointed director had transferred her to operations support. Her access to product documents had been shut off, she could no longer enter the data dashboards, and she was no longer invited to technical reviews. They called it, "You can transition through operations for now."
At two in the afternoon, a message popped up in the product group. The AI code review project she had previously led needed to draft a user interview plan that week. The person who had taken over tagged the whole group and asked who had a spreadsheet of user contacts.
Su Niannian had it on her computer. She had spent three months doing user research, adding every contact one by one herself.
She typed out a line, then deleted it.
No one had tagged her.
She minimized the weekly report window and opened BugKiller's system monitoring page. The competition codebases were sealed away and unavailable, but BugKiller's own condition had to be flawless. She remotely ran the internal benchmark once: detection engine response, concurrent processing, stability of parsers for every language. All green.
She sent Han Luyi a message. "The system's fine. I just have no idea what those three sealed projects are like. Total mystery boxes."
Han Luyi replied quickly. "We've controlled everything we can. Leave the rest to the product."
Su Niannian looked at the message, then glanced at the blank weekly operations report template again.
At five thirty, she got off work on time. Her laptop was in her backpack.
May 14, 6:55 p.m.
36-Degree Technology's livestream began.
The screen was split down the center. CodeSafe's logo was on the left, white lettering against a blue background; BugKiller's was on the right, red lettering against black.
The viewer count jumped from 8,000 to 12,000, then from 12,000 to 18,000. It was still climbing.
The livestream comments were already flooding in.
"They're here, they're here."
"Don't chicken out, Doctor."
"Go, NullPointer!"
"I've been waiting since three this afternoon."
"Nobody talk. Just wait for the results."
36-Degree Technology's studio wasn't large. Two rows of seats faced each other, three meters apart. CodeSafe was on the left, BugKiller on the right. Four cameras hung overhead, while a large screen at the front displayed real-time data.
Zhao Wenyuan sat in the left row, wearing a flawlessly pressed white shirt. Only the top button of his collar was undone. He nodded toward the camera.
Five meters away, Han Luyi sat down in the right row. He wore a white hoodie with "Debug the World" printed across the chest. He closed every unnecessary window, leaving only a terminal and BugKiller's admin dashboard on his screen.
Open Sight.
The calm was real, and so was the nervousness. He had come here with the mindset of staking everything on this one chance.
Han Luyi withdrew his gaze.
Five minutes before the opening, Zhao Wenyuan stood and walked over, extending his hand.
"NullPointer, I've heard a lot about you."
Han Luyi stood and shook his hand. "Dr. Zhao, your product is good."
Han Luyi wasn't merely being polite. Zhao Wenyuan's direction wasn't wrong; he had simply chosen the wrong time to fight a battle he shouldn't have fought.
Zhao Wenyuan smiled. "Likewise."
After shaking hands, they returned to their seats. The cameras captured the moment, and the livestream comments erupted again.
The viewer count surpassed 20,000.
The host stepped in front of the large screen.
"Good evening, everyone. Welcome to 36-Degree Technology's 'Public Head-to-Head Evaluation of AI Code Review Tools' livestream."
"The rules were announced beforehand. Tonight, CodeSafe and BugKiller will conduct three rounds of testing under identical conditions using identical codebases. All data will be displayed in real time, with community supervision throughout."
"The first test project is a Go microservice framework, approximately 30,000 lines of code."
"Now—"
"Begin!"
Before you continue