Windows is still here, but the one sitting in front of the screen is no longer human

For someone with almost zero programming knowledge, who only uses computers for basic office software and gaming, a product like Claude Code has been a mind-blowing experience.

At first, I just wanted to connect Claude to the yfinance database to query historical stock prices, performance, and real-time price information for Hong Kong and US stocks. After a few simple back-and-forth conversations and some copy-pasting, it was basically done. Later, when I noticed extreme anomalies in certain stock indicators, I started thinking about downloading the raw financial reports of the stocks in my portfolio and those I needed to monitor closely, so that Claude could help me input them into the database.

I hadn’t even started using Code by the time I completed these tasks, but Claude suggested through our conversation that I could directly use Code to modify my local server.py file, calculating various metrics based on the original financial reports to supplement and revise yfinance. It wasn’t until I used Code to complete all these operations that I understood why this software receives such high praise across various AI communities, and why its code leak caused such a sensation in the global developer community—

It's just so mindlessly easy to use。

All it gives me is Python code, which I don't understand and don't need to understand. I downloaded the financial reports in PDF format for companies like Meituan and Alibaba myself, put them into a folder, and told Code to process them. It reads the PDFs, extracts tables, builds databases, and occasionally pops up a permission request where I just click 'allow,' and it continues working. At this stage, all I did was download files and grant permissions.

Next, I didn’t even bother opening my browser. I simply said, 'Go scrape the original financial data from the company’s official website yourself,' and Code actually opened the browser on its own, found the IR page, downloaded the PDFs, extracted the tables, created folders on my hard drive organized by company and year, and entered the data into the database, completing a full set of data for the Magnificent Seven US stocks. Occasionally, when there were anti-crawling measures or permission verifications on websites, I had to handle them manually, but...For the most part, I just sat back and watched text scroll in the terminal, occasionally answering a confirmation question and clicking 'allow' with my mouse.。

After all the operations were completed, I suddenly realized something—I hadn’t done anything at all. I didn’t need to open a browser, use Excel, drag and drop files, or even create new folders. All I did was talk to it; if I wanted, I could avoid using the keyboard altogether and just use voice commands instead.

This led me to question whether, after AI achieves full control over computer operations in the future, systems like Windows with graphical user interfaces will still be necessary. The desktop, taskbar, start menu, window switching—all of these were designed for humans to operate computers. But now, the one operating the computer is AI, which doesn’t look at screens, click mice, or drag windows.Windows is still running, but the one sitting in front of the screen is no longer human.。

01 Who’s looking at the screen?

Of course, saying that Windows has no reason to exist is an overly hasty judgment. But what Windows represents—all software designed for 'human operation in front of a screen,' from operating systems to Office to Salesforce to every SaaS product—needs to answer a question after the widespread adoption of AI:

Is the cost users pay daily for this layer of design still worth the price?To understand this calculation, we need to first break down the value of software into three layers.

The bottom layer is the infrastructure: computing power, storage, networking, and deployment environments. This layer doesn't disappear. The middle layer is business logic: how data is processed, workflows operate, and rules are executed—the core value resides here. The topmost layer is the interface: translating the capabilities of the two lower layers into a graphical presentation that humans can understand and interact with. Buttons, menus, tables, charts, drag-and-drop, dropdown boxes—all these are products of the interface layer.

Over the past forty years, the evolution of the software industry has consistently been about thickening the third layer. From DOS command lines to Windows graphical interfaces, and then to desktop software and touchscreen mobile applications, every paradigm shift has lowered operational barriers and improved interface experience. The business logic behind this direction is clear: the better the interface, the more users it attracts, and the more revenue it generates. Based on the subscription model of the entire SaaS industry, every additional person who needs to 'look at the screen and click buttons' brings in extra revenue, essentially pricing the interface layer.

Anyone who has played games understands this. A game breaks down into code, a bunch of if-else statements and data tables. But when the computer runs this code, what appears on the screen is a stunning 3D world, dazzling effects, and polished UI. Most of the GPU's processing power goes into 'rendering for you to see,' and in this sense,Windows is actually no different from a game; it’s just that instead of fighting monsters inside, you’re doing productivity tasks.。

The computational power and energy consumption spent on that semi-transparent frosted glass effect on the desktop, real-time previews during window dragging, and the bouncing icons on the taskbar—that’s the interface tax.

How heavy is this tax? When Windows 11 is idle, memory usage typically ranges between 3 to 6 GB, while a Linux server without a graphical interface only needs a few hundred megabytes even when idle. Even after deducting the overhead from drivers and system services, the additional memory consumed purely due to the graphical interface exceeds 1 to 2 GB—enough for a local AI model to retain context for several thousand more tokens.

On the GPU side, things are even tighter: Windows’ Desktop Window Manager (DWM) continuously consumes GPU resources for interface compositing and animation rendering. Running an 8-billion-parameter local model on a consumer-grade graphics card with 8GB of VRAM is already pushing the limits; if the interface uses a few hundred more megabytes of VRAM, it becomes the difference between being able to run or not.

When the main operator of computers shifts from humans to AI, the direction of software evolution flips for the first time—not continuing to thicken the interface to accommodate more users,but bypassing the interface entirely to let AI directly invoke underlying capabilities.。

This is not speculation; looking at Microsoft's FY2025 figures, it’s clear they are already adapting to and even driving this trend. Annual total revenue of $2.817 trillion is divided among three segments: Intelligent Cloud, centered on Azure, accounts for 38%; Productivity and Business Processes, which includes M365, makes up 43%; and Personal Computing, including Windows, represents 19%.

In FY2025, Microsoft moved Windows Commercial out of the Personal Computing segment and integrated it into the Productivity segment where M365 resides, further reducing Windows' presence as an independent product in its organizational structure. Microsoft is not a victim of AI threats; rather, it is a driver of the collapse in value of the interface layer, actively shifting value from the interface to the data and infrastructure layers.

Microsoft has benefited greatly from the current wave of AI, but the winner is Azure, not Windows. Jensen Huang made a famous statement: 'AI doesn’t reinvent tools; it maximizes efficiency within existing tools.' Many believe this supports the idea that the current software system is irreplaceable, but maximizing efficiency actually means bypassing the interface to directly access underlying capabilities. What AI needs is the data itself and the logic to process it, not the visible spreadsheet you see in Excel.

In other words, computing power and business logic will not disappear, but the premium on the interface layer may no longer hold. If the world’s most successful software company is moving away from the interface layer, there’s no need to say more about this trend.

02 Stripping away the interface, what remains?

The depreciation of the interface tax is clearly outlined in Microsoft's financial report. But the more critical question is: why?

It’s not because AI performs tasks more beautifully; Claude Code looks terrible—just a black terminal with white text. It can replace Excel not because it has a better interface, but because all the computational functions achievable in Excel can be encoded into methodologies and taught to it, enabling AI to execute operations directly at the database level, ten times faster than manually clicking around, with significantly lower error rates when relying on local data for calculations.

The rationale behind software interfaces is that 'humans need them to operate.' When the methodology behind these operations can be encoded into instructions for AI, that rationale disappears.Interfaces have shifted from being 'essential' to 'optional,' turning a core value of the product into a bypassable cost.。

My individual investment research workflow operates like this, and so does a fifty-person sales team. The core methodologies of Salesforce—such as scoring leads, managing pipelines, and forecasting quarterly revenue—are all rules and processes that can be encoded. Once encoded, AI can handle the underlying CRM database directly, rendering the interfaces used by fifty individuals redundant.

Salesforce itself clearly understands this point as its Agentforce has started charging $2 per conversation, Intercom charges $0.99 per issue resolved, and Kustomer has fully transitioned to a hybrid model based on AI usage fees from a subscription basis. In highly standardized business scenarios, the underlying assumption of charging per seat is being replaced by the idea that each methodology only requires one agent to execute.

So which software companies should be worried? The criterion is actually quite simple: what remains after stripping away the interface.

If what remains can be encoded into methodologies taught to AI, then it's just an interface shell. The most typical example is Microsoft's own Office 365—Word's layout, Excel's data processing, and PowerPoint's presentation creation are all methodologies that can be taught to AI. Therefore, Microsoft introduced M365 Copilot with an additional charge of thirty dollars per month, attempting to shift the pricing anchor from the interface to AI.

But if I can directly generate a properly formatted docx using Claude Code or directly process data output, Office transforms from a productivity tool into a file format compatibility layer. It will still exist because docx and xlsx are de facto standards,and presenting work results to clients or colleagues will likely still require these software tools.However, in terms of actual work execution, it will be difficult for Office to regain its previous dominance.

If what remains is unique data that others don't have, that's a different story. Methodologies can be taught to AI, but data cannot. An AI that learns analytical methods without having access to data is like a skilled chef without ingredients. No one cared that Bloomberg terminals had an ugly interface for thirty years because you were paying for their data; Palantir’s interface isn’t its selling point either—governments and the military buy its ability to integrate sensitive data sources.

These companies possess unique data barriers and may actually benefit in the AI era: the cost of the interface layer is reduced by AI, while the data barrier remains intact, potentially increasing profit margins.

As an interesting footnote, the long-term undervaluation of Chinese enterprise SaaS is often explained by the market as being due to weak payment willingness among Chinese enterprises. But viewed from another perspective: China has never developed the habit of paying for desktop software interfaces. Pirated Windows was used for twenty years, and even now, the penetration rate of genuine Office isn’t high. This isn’t stinginess—it’s simply less sensitivity to the interface layer.

If the global software industry is experiencing a collapse in the value of the interface layer, Chinese enterprises, on the other hand, don’t have this sunk cost to absorb—they’ve never paid much of a premium for this layer. Conversely,China has never had its own desktop operating system ecosystem. In the past, this was a weakness, but if the next-generation system is designed for AI rather than humans, it represents a huge opportunity for innovation.。

03Conclusion

AI not only helps you bypass a layer of software interface; it helps you skip the entire operational chain from downloading financial reports, entering data, calculating ratios, to comparing with peers. However, the time saved should be spent on tasks that AI cannot do: going to the front lines, talking to people, or even visiting malls and supermarkets you haven’t been to in a while, observing things that are not in any terminals or databases.

BecauseThe best thing AI gives you is precisely the least valuable thing in investing: consensus.The 'interface tax' that AI saves you should be used to seek non-consensus.

The logic behind investing in software companies is the same. When you buy water, what you should focus on is what components, minerals, and trace elements are in the water, not whether the cup is made of glass or crystal. The interface is the cup, data is the solution—don't pay a premium for the craftsmanship of the crystal cup. The composition of the solution is what truly holds value.

As for judgment, that’s your own sense of taste. AI can’t give it to you, and neither can the interface. $Palantir (PLTR.US)$$Microsoft (MSFT.US)$$Alibaba (BABA.US)$

Disclaimer: This article is intended for learning and communication purposes only and does not constitute investment advice.

239K Views