Coding Is Becoming AI’s Digital Execution Force
Wen Le, reporting from Aofeisi
Qoder has taken on a new form.
And this time, the style is noticeably different.
As soon as its brand-new desktop app launched, it started raising a pet right on your desktop.
You can chat with it by voice in real time, poke it whenever you want it to get to work, and even watch it open a browser to check its own results as it goes…
Wow—an actual workplace buddy~
Of course, Qoder is still the same highly capable Qoder.
Over the past year, it has continued to evolve in Agentic Coding. From understanding code repositories and planning development tasks to writing code, running tests, and completing long-running tasks on its own, it now serves 6 million users and 100,000 enterprise customers worldwide.
This time, though, it has moved the Coding capabilities it has built up out of the IDE.
Now, when an idea strikes, you no longer have to open a project, create a workspace, and then figure out how to implement it. You can just chat with the desktop pet—
Qoder, beyond coding. So just how far beyond coding has it gone this time?
Let’s try it out.
Chatting Your Way to a Finished Product
For our first test, let’s “not be programmers.”
This time, we’ll play the role of a regional manager overseeing five branches of a chain coffee shop. Every day, besides keeping an eye on revenue, order volume, and best-selling products, we also have to manage employee schedules.
So why not build a system for monitoring business performance and managing staff?
The problem is, the store manager doesn’t know Coding.
No problem—we can simply call up Qoder’s desktop pet and start chatting.
If we don’t explain everything the first time, we can fill in the details gradually.
Once the requirements were more or less aligned, Qoder didn’t immediately dive in and start writing code.
Instead, it first produced a plan that reorganized the requirements scattered across several rounds of conversation.
It broke down the employee interface, the functions needed on the manager side, the data to display on the business dashboard, and the interactions between pages into a series of concrete tasks.
At this point, we can more or less step aside. This is when Qoder really gets to work.
Creating the project, designing the data structures, generating simulated business data, building the pages, writing the interaction logic, and filling visualizations with metrics such as revenue, order volume, and average order value.
Before long, a surprisingly convincing coffee shop management system was up and running.
On the employee side, staff can view their shifts and working hours and submit shift-change requests.
Switch to the manager side, and you can see an overview of every branch’s operations: how many orders were sold and how much money was made, all at a glance.
The weekly schedule is laid out clearly too, showing which employees are working at which branch and on what shifts. The shift-change request submitted by that employee is also waiting for approval~
As soon as it’s approved here, the status changes in real time on the employee side as well.
We didn’t write a single line of code from start to finish, yet a requirement that originally existed only as a verbal description really became a tool you can click through, view, and operate!
But one thing needs to be made clear: the fact that we didn’t do any Coding doesn’t mean Qoder didn’t.
When we reveal the execution process behind Qoder, it’s clear that it wrote every bit of the code.
And what’s more, Qoder can test the features it writes itself.
Through Computer Use and Browser Use, it can operate inside real software environments to verify whether the generated code works as expected in practice. When it finds a problem, it can trace back, modify the code, and complete the validation loop again.
After that first test, it was undeniably satisfying.
Still, building something from scratch is more or less the comfort zone of a Coding Agent.
With Vibe Coding having become so popular over the past two years, we’ve already seen plenty of examples of creating webpages, games, and even small applications with a single sentence.
To really see whether Qoder’s Coding fundamentals are solid, we had to throw it into someone else’s pile of code.
For the second test, we went straight to a real project.
This time, we found an existing 3D solar system project.
The original project was already quite complete: the Sun and all eight planets were present, the planets moved along their respective orbits, the view could be zoomed and rotated, and you could click different planets to view information about them.
This time, we wouldn’t ask Qoder to rewrite everything from scratch. We’d have it modify a project someone else had already built.
We also gave it a visually specific requirement:
Add an “interstellar travel mode” to the solar system. For example, when traveling from Earth to Mars, an interstellar route should appear, along with a small spacecraft traveling along it.
Add a travel panel on the side to show the current progress, flight status, and estimated arrival time. Ideally, the camera should move along with the spacecraft as well.
It sounds like a one-sentence task, but implementing it is not as simple as dropping a button onto the page.
Qoder first had to understand the existing Repo: what technology stack the project used, how the Sun and planets were generated, where the orbital animations were controlled, how the scene was rendered, and what managed the camera…
It couldn’t simply bulldoze the logic someone else had already built. It had to find the right places to integrate the new travel mode.
Once it had a solid grasp of the project context and produced a modification plan, Qoder finally started making changes.
It added the components required for travel mode, modified the existing scene logic, inserted the spacecraft into the 3D world, and handled the flight route, animation states, progress updates, and camera control. The changes began spreading across multiple files.
After modifying the code, Qoder once again “conscientiously” began validating its own work by operating in a real environment.
Once automatic approval mode was enabled, we could hand over the entire validation process to it and focus only on accepting the final result.
Next, we restarted the project ourselves and tried out the interstellar journey.
After selecting two planets, a Bézier route appeared in the solar system, where previously the planets had simply been orbiting on their own.
Click launch, and the spacecraft sets off. The progress bar advances, while the camera begins following it through the solar system.
This is what development work looks like in the real world. More often than not, programmers take over projects that have already been running for months or even years.
That’s why a tool that can understand someone else’s project and continue developing it is a genuine necessity.
Qoder also supports more than 40 connectors, over 70 plugins, and more than 20,000 skills, allowing work systems such as code repositories, project management tools, cloud services, and internal tools to be integrated into AI workflows.
Of course, these capabilities didn’t appear out of nowhere when the desktop app launched. They are built on the technical foundation Qoder developed over the past year in the field of Agentic Coding.
First comes context understanding.
Whether it’s a vague verbal request or an existing project repository it needs to take over, Qoder can understand the current state of the project and identify what the user is truly trying to achieve.
On that basis, Goal and Plan further break the objective down into an executable path.
Alibaba’s in-house Agent Harness organizes context, the file system, the terminal, and tool calls, allowing each model decision to translate into actual Coding execution—and enabling the process to continue based on new runtime results.
Finally, it validates the result through testing and hands-on capabilities such as Computer Use and Browser Use. When it finds a problem, it returns to the execution stage above and makes further modifications.
The seemingly effortless experience of “chatting your way to a finished product” is, when broken down to the underlying level, actually a highly complete execution chain:
Understand → Plan → Code → Run → Validate → Correct
Only after this entire loop was running smoothly could Qoder’s underlying Coding capabilities be brought to a wider audience.
Coding Is Becoming AI’s Digital Execution Force
Everyone is probably wondering too: Qoder is clearly a Coding Agent, so how did it end up taking on a wider and wider range of tasks?
To understand that, we need to start with the way we work today.
Think about it carefully: a large portion of many people’s work has already moved into the digital world.
Webpages, data, documents, software, databases, code repositories, business systems…
Although they come in all kinds of names and categories, at the end of the day, we spend every day working with these digital assets.
To build a webpage, you modify frontend code. To process a batch of data, you may need to run a script. To build an internal tool, you need to create files and implement features. To automate a workflow, you have to call different software programs and tools and connect steps that were originally scattered across different places.
These tasks may seem unrelated, but they’re actually all doing the same thing:
Operating on information and assets in the digital world.
How did people handle these tasks in the past?
The most obvious answer is with a mouse and keyboard: clicking buttons, dragging files, editing spreadsheets, copying and pasting.
But if the software’s built-in features weren’t enough, and you wanted to create something new or have a computer process batches of work automatically, you had to write code.
That’s where Coding’s power lies:
It can create entirely new software from nothing, process tens of thousands of data records at once, connect different systems, and automate tasks that would otherwise require dozens of manual clicks.
In other words, Coding provides the ability to create, transform, and connect the digital world.
Seen this way, it’s not so strange that Coding Agents are taking on an ever-wider range of tasks.
Because that’s exactly the problem AI Agents are facing today.
Modern large models can think, analyze, and plan solutions. But being able to come up with a solution doesn’t mean the task is finished.
Between coming up with a plan and actually bringing it to completion, there is still one step left: “taking action.”
Coding happens to fill that gap.
In the age of Agents, Coding is no longer merely a specialized skill used by programmers to develop software. It is also becoming a form of execution that allows AI to enter the digital world.
Coding No Longer Belongs Only to Programmers
Looking back at the changes in AI Coding over the past few years, the trend has been clear all along:
AI writes more and more, while humans write less and less.
But as Agents begin taking over more and more of the implementation process, another interesting change is emerging:
As more Coding is carried out by Agents, do the people invoking Coding capabilities still need to know how to Code themselves?
That’s what Qoder is doing this time: putting this Coding execution capability within reach of more people.
Of course, this doesn’t mean everyone will have to change careers and become a programmer.
How a product should be defined, whether the experience is truly good, how a business process should work, and whether an idea is worth pursuing—these decisions that genuinely determine the outcome will still come from product managers, designers, and operations teams.
Professional developers won’t disappear either.
The more complex the software engineering task, the more it requires genuine expertise in architecture, engineering quality, critical technical decisions, and final review.
What is really changing is the layer between “having an idea” and “implementing it.”
Over the past year, Qoder has refined this Coding execution capability through real-world use by 6 million users and 100,000 enterprise customers worldwide.
Now, it is putting that capability into the hands of more people.
So the next time a thought suddenly pops into your head—“Maybe I should build XXX”—
Don’t rush to find a programmer.
Call up the desktop pet and chat with it for a bit. You might just find that, somewhere along the way, it really does become a finished product.
Qoder, beyond coding—
Where Coding can take you is no longer limited to a programmer’s IDE.