Jiang Yixin stared at the data that had just been displayed on the screen, feeling her stomach churning.
It's probably because I ate too many steamed buns and fried dough sticks for breakfast.
She was unwilling to admit that she was nervous.
Soon, the first product she, as a product manager, led from idea to implementation will be released.
Yes, the programmed Agent, after about two weeks of internal refinement, is ready to begin small-scale testing.
Actually, this wasn't the first product Jiang Yixin made.
After joining the source Source Code, she followed up on the development of Open Things, and even named Open Things herself; later, after switching to product development, she was also deeply involved in Open Things and the Kaiwu Overseas Edition.
However, it was Han Luyi and Su Niannian who initiated those plans; she mainly focused on perfecting and implementing them.
The Agent project is different. This project was an idea proposed by Jiang Yixin herself. The prototype, interaction method, and early prototype were all created by Jiang Yixin and Qian Xiaole.
If I had to put it in a way, it was like her first child.
Now that this child is going out to be judged, it's understandable that he feels uneasy.
She glanced to the side subconsciously; Qian Xiaole's workstation was empty, she hadn't arrived yet—she almost never came to the office before ten o'clock.
Jiang Yixin reviewed the data again:
It was used internally for two weeks, with a total of 27 engineers involved.
The number of code merges created with the assistance of programming Agent increased by 64 percent compared to the two weeks before launch.
Of these, 52% were code branches that were first drafted entirely by the Agent, reviewed by engineers, and then merged.
The rollback rate was only 3.1%.
These numbers, when put together, are actually quite impressive.
It's even frighteningly beautiful.
But Jiang Yixin was still worried, so she clicked on the more detailed task breakdown.
The most noticeable improvements are seen in simple tasks: UI component adjustments, field renaming, interface parameter adaptation, test case completion, and log format modification.
These tasks used to require developers to spend their time on them. They were tedious, disruptive, and required special scheduling. A single work order could take up half a day of an engineer's time.
Now, using Agents, an engineer can do five in a morning.
The developer writes the requirements into the code, and the system can automatically view the project structure, find relevant documents, plan tasks, divide them into sub-tasks, generate changes, run tests, and fix them again if they fail.
The adoption rate for this part of the task is astonishingly high; the Agent has essentially become a "low-level chores cleaning robot."
This is also where the product's value is best demonstrated.
However, when it comes to complex tasks, the data doesn't look so good.
Large-scale refactoring across multiple modules, involving modifications to historical business logic, migration of old code with long dependency contexts, and tasks with inherently ambiguous requirements, clearly strains the Agent.
These are problems that cannot be solved by iterative product development itself; they represent the limits of the capabilities of the two Large Models underlying the product.
The code generation capabilities of the Heavenly Craft model are limited, and the Mercury Model has also reached its bottleneck.
Jiang Yixin believes that at this stage, further internal refinement is no longer very useful; it's time to take it to the market and receive feedback.
She opened the report document and added a line at the end, "The current version is suitable as an automated execution tool for low to medium complexity development tasks. Unsupervised approval is not recommended for high-complexity cross-module refactoring."
She had just finished writing when she suddenly felt someone tap her on the shoulder.
Jiang Yixin turned around and saw that it was Li Ting. She was holding a Starbucks coffee cup in her left hand and a paper bag in her right.
"What are you looking at so intently? You look really serious," Li Ting said. "I brought you a large latte, and another one is for Lele."
Li Ting placed the paper bag on the table as she spoke.
"Sister Ting." Jiang Yixin together and bowed to Li Ting, then took out a cup of coffee from the paper bag and took a sip.
Li Ting sat down at her workstation and then slid her chair over.
Li Ting's workstation is right next to Jiang Yixin's.
She glanced at Jiang Yixin's screen and said, "Aren't these data pretty good? Aren't you satisfied?"
"Simple tasks are fine, but I can't avoid complex ones," Jiang Yixin said.
Li Ting turned to look at her. "President Jiang, aren't your expectations of yourself too high?"
Jiang Yixin paused for a moment before saying, "I don't know, I'm just a little nervous."
"Don't be nervous." Qian Xiaole's carefree voice came from behind, "Our product is sure to be a huge hit. It might not be as popular as Open Things, but it'll at least be on par with BugKiller."
As she spoke, she walked over and sat down on the other side of Jiang Yixin, glanced at the paper bag on the table, and asked, "For me?"
Li Ting nodded, Qian Xiaole took out the coffee, took a sip, and said, "Thank you!"
Then she said to Jiang Yixin, "Don't worry, all programmers will love Agents. This is the best programming tool I've ever used! Anyway, I never want to write another line of if-else code by hand again."
Li Ting said from the side, "Look, even the heavy users say so, so don't worry about it."
After being comforted by her two colleagues, Jiang Yixin felt much better. Just then, a notification popped up in the upper right corner of the screen: her meeting time with Su Niannian was approaching.
Jiang Yixin went through the data one last time, glanced at the report document, then got up, picked up her laptop, and walked towards Su Niannian's office.
Li Ting also saw it and gave Jiang Yixin a thumbs up, saying, "Keep it up!"
"President Su?" Jiang Yixin knocked on the office door.
"Please come in." Su Niannian said as she tidied up the documents scattered on the table.
Jiang Yixin pushed open the door and went in, seeing Su Niannian organizing documents. She glanced at the scattered materials on the table out of the corner of her eye.
The book includes "Small Business Toolkit," "Open Things: Tutorial for the Children's Entertainment Industry," and "Open Things: Tutorial for the Beauty and Hairdressing Industry," as well as some materials for distributors.
Su Niannian had already mentioned this in a previous meeting: this year, Open Things's focus is on penetrating specific niche industries and scenarios, starting with children's entertainment and beauty salons.
After the surge in traffic in Q4 of 2026, the demand for website building by Thousand Sails merchants decreased, and the number of calls to the Open Things white-label API dropped significantly.
In Haicheng, the Source Code company set up its own ground sales team as a pilot project and began competing with Thousand Sails for users. This is not the domain of an AI company, but rather the approach of an internet company. Although they were promoting AI tools, AI could not help with the promotion process. Merchants and customers had to be negotiated one by one, which was a lot of hard work.
Su Niannian has been busy with this matter lately.
After quickly organizing the documents, Su Niannian rubbed her temples and looked up at Jiang Yixin.
"Is it about Agents?"
"Yes, President Su," Jiang Yixin said, "I would like to apply to open the first round of closed beta testing starting tomorrow."
As she spoke, Jiang Yixin took out her computer, placed it on the desktop, and pushed it in front of Su Niannian. The computer displayed backend data and the report document she had written.
"This is the data we've been using internally so far. Please take a look."
Then Jiang Yixin reported the data item by item.
After listening to Jiang Yixin's report, Su Niannian thought for a moment and asked a question, "Of the thirty-person development team, only twenty-seven people are using it. Have you asked the three people who aren't using it?"