The Omnipresent AI Illusion: Microsoft Finally Halts Its Copilot Frenzy

As this industry-wide storm begins to cool down, we can finally sit down and calmly discuss real user needs.

It’s raining outside again. March in Shanghai is pretty much like this; the continuous cold and damp weather always manages to sober you up a bit. I just saw the news that Microsoft announced an indefinite pause on the automatic deployment of Copilot in Windows and Microsoft 365, quietly removing the AI assistants from Notepad and Photos as well. How should I put it? This fanatic hurricane trying to forcefully stuff AI into every corner of their products has finally hit a brick wall.

Some Data Actually Wrote the Ending Long Ago

I previously saw a set of data from an independent research firm. Within Microsoft’s massive base of 440 million paid M365 users, the proportion that actually converted to paid Copilot subscriptions was a pitiful 1.8%. When I first saw this number, I thought it was unbelievable, but upon closer thought, it makes perfect sense.

All along, what the outside world has seen are PR releases from Fortune 500 companies spending hundreds of thousands on AI licenses for their employees. But if you actually talk to the IT administrators who run the enterprise backend, you hear a completely different story. Companies spent top dollar on enterprise licenses only to find that employees either couldn’t use them at all, or the cost of using them was simply too high.

Here is a real scenario that happens every day: A programmer uses Copilot to write a snippet of code. It might indeed save a minute of typing time, but they then have to spend a minute and a half checking whether the code called the wrong library or contains security vulnerabilities. In an enterprise productivity environment that pursues certainty, if a user must double-check every result generated by AI, then this so-called “efficiency boost” is frankly just a new kind of efficiency tax.

Furthermore, data assets in large enterprises are extremely sensitive. If an ordinary employee accidentally feeds an unreleased financial draft or core source code into an auto-connected AI assistant, the resulting compliance risk is something no IT department can bear. So they had no choice but to add layers of firewalls, ultimately castrating a perfectly good intelligent tool into a useless electronic ornament. Faced daily with a mountain of complaint tickets about system lags and update errors, IT administrators collectively pushed back hard, even forcing management to cancel subscriptions. Microsoft’s emergency halt this time is essentially paying the price for the aggressive “I don’t care what you need, I’m cramming AI into everything” strategy of the past year or so.

Who Is Still Superstitious About Dialog Boxes?

Speaking of this, I’m reminded of a particularly intriguing detail in this update—the widely criticized forced system update was officially halted, and the classic, movable taskbar that everyone has used for over a decade announced its return.

This exposes a rather deep logical contradiction: Do we really need a universal dialog box when designing product interactions?

Think about it, what is the point of a software like Notepad? It is meant to handle plain text in a lightweight, instant-open, burden-free manner. When I want to quickly jot down a phone number in it, my core demands are “speed” and “muscle memory.” I don’t need a blinking icon to jump out and ask “How can I help you today?”, nor do I need it to presume to polish that phone number for me. Let alone that widget permanently residing on the edge of the screen—besides eating up RAM, I truly can’t think of any substantial help it offers to ordinary office workers.

Those product managers who constantly preach system-level integration really need to wake up. Forcefully pinning a dialog box to a user’s taskbar is, at best, a piece of irremovable digital psoriasis. Slapping technology directly onto the surface of a product is called a “gimmick”; making technology invisible and genuinely improving efficiency is what we call “refactoring.” Microsoft’s retreat this time proves exactly that, no matter how powerful the underlying models evolve to be, they cannot cross the red line of “respecting user habits and system administrative privileges.” The absolute sense of control a user has over their computer brooks no interference from a presumptuous assistant trying to dictate terms.

Looking at Peers: To Put It Bluntly

Comparing current product forms across the board actually reveals a lot.

Right now, stuffing features into systems or large-scale software generally falls into two schools of thought. One is the radical approach previously taken by Microsoft, where they seemingly want to place a blinking logo in the bottom right corner of every single page, constantly asserting their presence.

The other, such as Apple’s approach in macOS, is relatively much more conservative and restrained. It is mostly embedded in the underlying context; when you need it, you highlight text to invoke it, and when you don’t, it’s as if it doesn’t exist. Or look at a certain leading domestic collaborative office platform—they also experimented with various floating windows in their early days, but later discovered that what everyone uses most is still the forward slash “/” hidden in the blank space of a document to wake it up. Use it and go, with zero disturbance.

To put it bluntly, if your flagship feature relies entirely on forced pop-ups and an unhideable taskbar icon to gain daily active user data, it means the product itself lacks the charm for natural retention. But there’s one issue you must understand: once a feature is forcefully pushed at the system level, the ensuing privacy panic is also system-level. When users feel this assistant is pervasively monitoring every keystroke and every local file, the resulting resistance is irreversible.

Furthermore, beyond the differences in experience, cost is an insurmountable hard threshold. The computing power consumed to broadly roll out computational services of this scale is staggering. If the user end doesn’t buy it, this massive idle spinning of compute is just burning money like crazy. In contrast, those lightweight competing products that focus on thoroughly solving one or two specific scenarios actually live much more comfortably, because they address tangible pain points rather than peddling grandiose but impractical futurism.

Some Things I Think About Sometimes

I sometimes wonder, could the entire industry’s frantic march over the past year or so just be a stress response to new technology?

Because the evolution speed of large models has been too fast—so fast that even tech giants of this massive scale haven’t had time to truly think about what native interaction should look like. To spin a good story during earnings calls, and to prove to the market that they still hold the ticket to the future, the fastest method was to rip open all existing product lines and forcefully cram a language model inside.

Taking a step back, does AI really have to be an anthropomorphic “assistant”? Is it possible that the system intelligence of the future won’t have a UI at all? It might just be underlying gears within the system scheduling resources, managing memory, or even predicting which software you’ll open next. Of course, maybe I’m overthinking it; after all, an anthropomorphic dialog box is currently the easiest shortcut to make the public perceive that “this thing is smart.”

It’s just that after this period of growing pains, perhaps everyone can calm down and recognize reality. This should be considered a fairly important industry inflection point: moving from blindly striking out everywhere to genuine restraint and convergence.

The Coffee Is Cold

The rain seems to have stopped, and this cup of coffee is just about empty (¬_¬). By the way, a quick rant: wouldn’t it have been better to give me back that freely draggable classic taskbar earlier? Yesterday, when I was testing on another computer, I couldn’t drag that centered taskbar no matter how hard I tried, leaving me completely speechless (helplessly throwing hands up).

References:

—— Lyra Celest @ Turbulence τ.