Skip to main content
  • Home
  • Library
  • Read
    • Popular
    • Charts
    • Browse
    • Categories
    • Wikis
    • Glossary
  • Request
  • Activity
  • Leaderboard
  • Help
Our Apps
  • DaoSearch
  • Narratorr
Social
  • Discord
  • Patreon
The AI Wave Tycoon's Rise
Chapter 147

Since it's so smart, let's call it an Agent

1,522 words 5 min read 2 views 1mo ago Source ​
Skip to the end of the chapter
Prev
Next

Qian Xiaole and Li Ting were taken aback by Jiang Yixin's reaction.

"What's wrong?" Li Ting asked her.

Jiang Yixin didn't answer immediately; her chopsticks hovered in mid-air, her eyes still fixed on Qian Xiaole.

Say that again.

Qian Xiaole: "What kind of TV drama scene is this? After I say this, are you going to say, 'No, it's the previous sentence'? I've said several sentences while eating a sandwich."

"He kept clicking 'yes' without even looking."

Qian Xiaole said, "Isn't that obvious? It's pointless to look at it; I'll just look at the test."

"Yes." Jiang Yixin, emphasizing each word, "Since we're not going to look at it anyway, why should we let people order it?"

Qian Xiaole was about to take a bite of the sandwich, but her Zhang Kai mouth remained there without taking a bite.

Jiang Yixin put down her chopsticks. "The Mercury Model stops and asks you questions at every step. Its design is based on the assumption that 'a person needs to review things within the loop'—the so-called 'man in the loop' in technology. But if a person doesn't look at it at all, or doesn't need to look at it at all, then the only purpose of these confirmations is to trap people inside."

"Huh?" Li Ting looked at this and then at that.

She's in operations; the technical details are like foreign languages to her.

"Planning tasks, writing code, running tests, fixing bugs," Jiang Yixin, as if talking to herself. "It can already do the whole process that we developers do every day at our workstations, every single step."

"Who's 'us' with you? You're a product manager, a class enemy!" Qian Xiaole retorted reflexively, but then added, "However... my daily work really is just this: scanning project structure, writing code, running tests, fixing bugs—it's all become routine. This AI is still asking questions like this, it's pointless."

"What are you talking about?" Li Ting pressed.

Jiang Yixin did not answer immediately.

For the past few days, she had been immersed in the documentation of the Mercury Model, trying out the product, and researching "what this thing can do," but the phrase "just keep clicking yes" made her step out of the researcher's perspective.

What if we remove the person from the ring?

Define the task, and then let the AI complete the task directly.

What if we integrated this capability into our own codebase?

Product development requires approval, but internal tools don't.

Isn't this exactly what Boss Han has been promoting—using AI?

Not only use it, but use it extensively and extensively.

"We can make it with our own technology," Jiang Yixin said.

Qian Xiaole put down the sandwich and sat up straight. "You mean, use Heavenly Craft?"

"Heavenly Craft handles code generation and execution, integrating with BugKiller's detection engine. There's no readily available model for task planning and management, but Neus provided us with a free Mercury Model API, so we'll use their API first."

Qian Xiaole stared at her for two seconds. "Are you free this afternoon?"

That's exactly what I was going to ask you.

Li Ting first looked at Jiang Yixin on her left, then turned to look at Qian Xiaole on her right, and sighed, "So, are we still having milk tea?"

After the meal, Jiang Yixin and Qian Xiaole each grabbed a cup of milk tea and occupied a small meeting room.

A simple architecture diagram was drawn on the whiteboard: the user inputs natural language commands, the Mercury Model performs intent understanding, transforming the vague description of the task into a completion standard and many executable sub-tasks, and then handing it over to the Heavenly Craft model for code generation and execution. If the BugKiller detects a bug, it returns it to the Mercury Model, and the above steps are repeated until the task completion standard is met. This is no different from Open Things, the only difference being that the new architecture has a listening process.

What does that mean?

Previously, Open Things would receive an instruction, generate an app, and that would be the end of the process. If the user didn't like it and suggested modifications, Open Things would send the previously generated app along with the new instruction. Each round had a beginning and an end.

However, once a monitoring process is established, it never ends; as long as the computer is on, the process will continue to run.

You can give it a long task, check its status at any time, interrupt it, or give it new instructions at any time. If the previous Open Things method was a soldier, it would attack wherever you pointed.

The new architecture model becomes a commander that can direct soldiers and report to users at any time.

It sounds interesting, but it's a bit difficult to do.

After discussing it, the two began to divide the work.

"I'll have Heavenly Craft write the glue code," Qian Xiaole said. "I'll write the calling logic, and AI will write the parsing logic. We'll work together—no, I'll work with AI." And what about me?

"You just need to stand on the sidelines, clapping and cheering."

Jiang Yixin kicked her chair. "I'm a veteran developer too, okay?"

The two started working, and the only sounds in the meeting room were the clatter of keyboards and occasional discussions.

Qian Xiaole is indeed fast at writing code, or rather, she's fast at directing AI. She almost never types more than ten lines of code in the editor by hand; she lets the AI generate all the repetitive structures, only modifying the key logic and API calls herself. Jiang Yixin watched from the side, feeling like she was watching someone play a real-time strategy game: the amount of operation wasn't large, but every move was precise.

Does coding involve micromanagement?

Two hours later, a very rough but functional demo version was displayed on the screen.

Jiang Yixin used a portion of non-sensitive code from the front-end project as test material, and then typed a line of text in the input box:

"Add One dark mode toggle to the settings page."

Then I clicked "Start".

The log stream on the right side of the screen started scrolling.

Project structure scan: Vue3 + TypeScript was identified, along with CSS variables used for styling and global state records. Relevant files were located: SettingsPage. vue, theme. ts, and variables. css. Planned changes: Add a Toggle component, expand CSS variables to support dual themes, and write user preferences to localStorage. Coding begins.

The code has been committed to the feature/dark-mode branch.

Test suite run: Pass.

BugKiller scan: Oissues.

A preview URL has been generated:

The whole thing takes about thirty seconds.

Qian Xiaole sat down as well, the two heads huddled together, staring intently at the log.

"It ran the test itself?"

"Um."

"Is BugKiller also invoked by itself?"

"Um."

Jiang Yixin clicked on the preview link.

A moon icon has been added to the top right corner.

She nodded.

The entire page was rendered in a dark color scheme, with font color, background, card borders, and button states all using dark variables, and not a single element seemed out of place.

"I'll write this myself," Qian Xiaole's voice was a little hoarse. "From reading the code to finishing it and passing the self-test, it would take at least one or two hours. This only takes thirty seconds?"

The AI understood what "dark mode" meant in this specific project. It didn't just change the background from #fff to #000; it also involved modifying many component CSS variables, adapting component libraries, and handling user preference storage.

It understands the context of the project and provides a customized solution, not a general one.

More importantly, it handles the entire process, from understanding to execution, all on its own. Jiang Yixin and Qian Xiaole simply pressed the start button.

The two looked at each other, both seeing shock in each other's eyes.

"This is... amazing!" Qian Xiaole murmured. "Shouldn't we give it a name?"

Jiang Yixin on her hand. "It's so intelligent, let's call it..."

"Agent?"

When I came out of the meeting room, it was already completely dark outside the window.

Jiang Yixin organized the screen recording and screenshots of the test she had just taken, opened Lark, and found Su Niannian's chat window.

Since transitioning to product management, her direct supervisor changed from Han Luyi to Su Niannian, and she still understood the principle of not reporting directly to superiors. Message title: <Agent, a New Internal Tool>.

She sped up the short video she had just recorded for the demonstration; it only took about 10 seconds to watch.

She carefully considered the main text several times, condensing the technical details into three paragraphs, focusing on the comparison of effects: for the same requirement, it would take several hours manually, but only thirty seconds with the new tool.

Send.

Jiang Yixin started tidying up her desk to leave work, while Qian Xiaole had already left; she never worked overtime.

A new message suddenly popped up on Lark.

It's Su Niannian.

And it wasn't a text message, it was a voice message.

Jiang Yixin clicked on it, and Su Niannian's urgent voice came from the speaker:

"Xiao Jiang, are you still at the company? Don't leave yet, I came to check on you."

End of Chapter
Prev
NovelSummaries
Next

Explore the wiki

Han Luyi
Su Niannian
Jiang Yixin

Comments

No comments yet.
AboutStatsAPIChangelogStatusContactPrivacyTermsDMCA
© 2026 RandomTranslator·v3.26·RSS