Fortunately, they had no large model of their own
John arrived at Ryan's house soon after.
When Ryan went to open the door for John, his wife Emily was coming downstairs. She waved to John first, then lowered her voice and said, "If you want to talk, keep your voices down; the kids are asleep."
Ryan nodded.
He led John to the study to get the computer, and the two went to a room in the basement.
As soon as they sat down, John spoke up, "You mean that China company that developed an automated workflow product?"
There was still a hint of disbelief in his voice.
Ryan nodded, then checked the time; it was already 10:50.
"I'll do another demonstration during the meeting later, then you can see it then," Ryan said.
"Maybe the tool will malfunction when you demonstrate it later," John joked.
Anyone in this industry has had this experience: often when creating a software prototype, everything works fine in private, but during the on-site demonstration, unexpected bugs pop up.
"I sincerely hope it crashes." Ryan didn't laugh; he looked at John intently.
John's smile also faded.
He understood Ryan's meaning: if the demonstration went wrong, it meant that the new product was at most on par with Atlas—threatening, but not fatal.
But if we can consistently reach the level Ryan mentioned on the phone earlier...
Then Ne. us would be in great danger.
The online meeting started quickly.
When CEO Ryan and CT0 John were together, the employees present sensed something was amiss from the very beginning of the meeting. Product manager Sarah and the engineering team leader had their own guesses about the meeting's content, while the research team leader and marketing manager, who were called in at the last minute, were completely bewildered.
The marketing manager is currently having a meeting on his phone on the street outside the bar.
Ryan ignored everyone's reactions and quickly explained the situation.
"There's a company in China that opened up beta testing of a product a few hours ago, and its direction is highly overlapping with NeusAtlas," Ryan said, opening screen sharing and scrolling through the page as he spoke. "This is their product landing page, and this is the product description."
As Michael, the marketing manager, walked down the street, trying to find a quiet place, he wondered to himself, "Was this plagiarism? Why didn't they call the legal team into the meeting? What do they want with me?"
"And then, here's their product." Ryan said, opening the terminal on his computer and adding a new feature to his toy project.
He deliberately didn't try to replicate the previous function; perhaps Ryan secretly hoped that this time the operation would fail.
However, no.
The Riding the Wind ran very smoothly, analyzing the codebase, planning tasks, generating code, and completing unit tests.
Finally, a summary was generated, detailing the changes made during this task and the underlying product logic.
Furthermore, because it was within the same codebase and this wasn't the first time it had been run, the time spent analyzing the codebase was greatly reduced, and the entire modification only took about three minutes.
After the demonstration, Ryan remained silent.
The other people in the meeting remained silent.
After a while, CTO John spoke up, "This level of completion is much higher than Atlas."
After John finished speaking, the atmosphere in the online meeting became even more tense.
Evan, the research team leader, was the first to change his expression. He was one of the earliest researchers on the Atlas project, and he had signed off on all the training and iterations over the past six months.
"Wait, Ryan, can I watch the demo again?" Evan frowned, muting his phone. "I think there was something wrong with the demo just now..."
"The demo was something I just did live, not a video," Ryan interrupted him. "I've done it twice already, this is the third time. Do you think it's just luck?"
Evan didn't reply, and then he muted himself again.
The online meeting fell silent again.
Ryan didn't let the silence last long; he turned off screen sharing, and everyone's cameras instantly filled the big screen.
"I didn't call you all here in the middle of the night to tell you we lost," Ryan paused, then continued, "I called you here because we need to win, and we only have time to start tonight."
On the streets of the Bay Area, Michael had stopped. There were two homeless people's tents in the distance, but he didn't intend to go any closer.
"Sarah," Ryan called out, "How many users does the closed beta of Riding the Wind currently cover?"
Sarah had already checked and quickly replied, "They're all China users, only one hundred people. There's absolutely no overseas exposure."
Is it being discussed on Hacker News?
"No."
"On X, within our developer community?"
"No."
Ryan nodded.
"In other words, in our world, the Riding the Wind does not currently exist."
He scanned the cameras around everyone.
"You all understand the rules of the game in Silicon Valley —whoever seizes the initiative first becomes synonymous with this field," Ryan said, then emphasized again, "It's not about who does it first, it's about who makes the world remember them first."
John gave a soft "hmm" from the side.
He didn't need Ryan to explain that.
Dropbo wasn't the first cloud storage, Slack wasn't the first enterprise real-time communication tool, and OpenAI wasn't even the first to create a Transformer.
But these names are now synonymous with their respective tracks.
"Atlas was originally scheduled for a Q1 release," Ryan said, "But now it has to be brought forward."
He turned to look at John: "Around us."
John's expression was a little stiff. He knew the progress of the Atlas project better than anyone else. Q1 was a time that he and his team had gritted their teeth and fought for.
Four weeks means the team will be working non-stop, it means he must cut off all temptations to "optimize it a little more", and it means he will release based on known defects.
But he was a man who had seen battle, and he knew that he couldn't say "no" at this moment.
"Four weeks," John repeated, "I'll try."
"It's not a trial," Ryan said, "It's a must."
John paused for a moment, then nodded. "Okay."
Evan tried to interject again. "But Atlas's current accuracy rate is..."
"Evan." Ryan looked at him. "Starting tonight, the engineering team will trace every prompt word coming in from Riding the Wind. Riding the Wind is a command-line tool; every model call, every file read, and every prompt word input is a public event. I need a reproduction report within forty-eight hours."
"----You mean copy?" Evan frowned.
"I don't like Mark," Ryan said, "but he's right about one thing—'Don't be too proud to copy.'"
This quote is very famous in the Bay Area. It was said by Mark Zuckerberg, the founder of Meta, at an internal employee meeting when facing the impact of Snapchat. Since then, from copying Snapchat to copying TikTok, it has been repeatedly quoted within the company and has become synonymous with "mature companies in Silicon Valley are not afraid to learn from startups".
Then, Ryan's tone softened as he said, "You don't need to be too pessimistic. The China company that developed this tool is our partner, and I know them very well."
"They don't have their own Large Model."
He paused, his gaze sweeping over John.
"Riding the Wind's core capabilities—intent understanding, Task Orchestration, and technical decision-making—all rely on the Mercury Model. They use our API. Their entire product is built on top of our model."
After these words were spoken, everyone in the online meeting breathed a sigh of relief.
Yes, Neus started out as a model-making company, so there's nothing shameful about not being able to outperform others in terms of product quality.
Didn't they also rely on our model to create this tool?
Ryan turned to Michael on the screen. "Michael? Michael?"
"Yes, yes, I'm here!" Michael suddenly realized, and after searching for a while, he finally found the mute button.
"Tomorrow morning, take your team and contact all the tech KOLs, podcasters, and active HackerNews accounts you can reach. On the day Atlas is released, I want the whole world to see it."
3
What is the budget?
No upper limit.
Michael closed his eyes on the street; he roughly knew how many vitamin injections he would need to go to the hospital for over the next four weeks.
"There's one more thing," Ryan said, "Absolutely do not mention that China company in the press release."
Michael certainly understood what Ryan meant.
By not actively mentioning the competitor, we are not giving them free exposure. Let Atlas appear as a "Pioneer" and let the Western world conclude that "Neus created the first truly intelligent development workflow." By the time someone digs up Riding the Wind's Chinese materials a few months later, it will be too late.
"Understood," Michael said.
After assigning the tasks, Ryan said his last words in the meeting, "Don't worry, we will win."
But these words did not bring any comfort to the attendees. They still couldn't understand how they had been surpassed by a team from a small company in China —one that was even using their own Large Model at the lower levels.
The meeting ended just after midnight.
John looked at Ryan and reiterated his point. "AtIas is far from ready."
Ryan looked John in the eye and gave his answer, "The market isn't ready either."
Opportunities won't wait for you to be ready.