AI Is Beginning to Compete for Control of the Computer: Why Desktop Software Suddenly Matters Again

Just as software was moved to the cloud, AI is now rushing to build desktop clients one after another.
But this is not desktop software coming back.
What is truly happening is:

AI is beginning to compete for control of the computer.

Look at the recent moves by several major Chinese tech companies, and it becomes very interesting.

Tencent's WorkBuddy is no longer just a chatbot – it is evolving toward an "AI workbench": connecting WeChat, WeCom, documents, email, meetings, and knowledge bases, and has started launching industry Buddies, Skills, connectors, and enterprise-level Agent management. At the end of September, WorkBuddy Managed Agents entered enterprise whitelist testing.

ByteDance's "Doubao Work" is even more direct.
Launched in August, it quickly added multi-Agent parallelism, Mac computer operation, local Office editing, browser recording and playback, and switching between local and cloud computers. Feishu officially now defines "Doubao Work" as an agent work platform for individuals, teams, and enterprises.

Alibaba's Qwen Office is following the same path.
In September, it continuously updated capabilities such as connectors, security centre, scheduled tasks, and project workspaces. What the desktop client emphasises is no longer "chatting with you," but enabling Agents to truly complete tasks in local working directories.

Even DeepSeek has launched Harness.
The latest desktop version on October 5 is still being continuously updated, and its core idea is "everything is a plugin": processing files, writing code, running tasks, background execution, and even letting users create new plugins themselves.

Putting all of this together, the answer becomes clear.

What AI vendors are truly competing for is not a desktop entry point, but "who can operate the computer on your behalf."

01The Old Software World: Humans Operate Software

The old software world was: humans operate software.
You open Excel and find the data yourself;
open ERP and fill in the form yourself;
open CAD and modify the drawing yourself;
open MES and check the records yourself.

In the Agent era, it may be the other way around:

humans set goals, and AI operates the software.

You only need to say:
"Analyse this batch of equipment data, find anomalies, check historical maintenance records, and then generate a report."
AI finds the files, opens the systems, invokes the tools, and runs the analysis itself.

At this point, the desktop suddenly becomes very important.
Because a large amount of real work is simply not in the cloud.
Files are on the computer, professional software is on the computer, and enterprise internal systems are also on the computer.

02Especially for Industrial Enterprises

Especially for industrial enterprises.
A factory may simultaneously have MES, ERP, CAD, SCADA, testing software, PLC engineering software, and a large number of legacy systems without standard APIs.
In the past, we could only connect these systems bit by bit through interfaces, integration, and manual effort.
But now another path has emerged:

letting AI directly learn to use these software systems.

So the next round of industrial software competition may not be about who builds another bigger MES.
But rather:

who can add a new AI operation layer on top of existing MES, ERP, CAD, and equipment systems.It understands enterprise knowledge, equipment data, and business processes – and can actually get hands-on work done.

03The Real Change: Software Must Also Be Operable by AI

This is the real change behind why desktop software is becoming increasingly important in the AI era:

In the past, software was operated by humans.In the future, software must also be operable by AI.

And whoever first masters this layer of "operating rights" may secure the entry point to the next generation of software.