>_TheQuery
← All Articles

Microsoft Killed Its AI Sidekick. Game Or Strategy.

By Addy · May 12, 2026

On May 5, 2026, Asha Sharma posted on X. She had been Xbox CEO for three months, inheriting a division under pressure and a product that gaming writers had already treated as a solution looking for a problem at launch.

The product was Gaming Copilot. The post was its obituary.

"As part of this shift, you'll see us begin to retire features that don't align with where we're headed. We will begin winding down Copilot on mobile and will stop development of Copilot on console."

The full lifecycle of Gaming Copilot - from announcement in March 2025 to cancellation in May 2026 - spanned fourteen months. It never left beta. It never shipped on a console. It was described by its own users as redundant, by gaming writers as a tool built for a problem nobody had, and by internal Microsoft sources as part of a broader pattern they could not say out loud in public.

The Xbox cancellation is the most visible data point in a story that is much larger than one product. To understand what actually happened, you have to go back further than Gaming Copilot. You have to go back to the moment Microsoft decided that AI was not a product. It was an identity.

The Plan Was Total Occupation

In 2023, Microsoft made a bet. It invested $13 billion in OpenAI, secured the most significant AI partnership in corporate history, and decided that the way to win the AI era was not to build one great AI product. It was to make every Microsoft product an AI product.

The strategy was called Copilot. The brand would be the thread. Every surface Microsoft owned - Windows, Office, Teams, Xbox, Bing, GitHub, Edge, Notepad, Snipping Tool, Photos, Widgets - would carry the name and the logo and the capability. Copilot would be to Microsoft's AI era what the Windows logo was to its computing era: the mark of presence, the signal that you were in the right place.

Within eighteen months, the Copilot brand had been applied to nine distinct products. Copilot for Microsoft 365. GitHub Copilot. Copilot for Security. Copilot for Xbox. Copilot in Windows. Copilot in Bing. Copilot Pro. Copilot Studio. Copilot Plus PC. Each was a real product with real engineering behind it. None of them shared a coherent user experience. All of them shared a name.

The ambition was unmistakable. Microsoft was not trying to build an AI feature. It was trying to build an AI company out of an existing company, by renaming everything it already owned and calling the result a transformation.

It did not work. The evidence arrived quietly, from multiple directions, before the Xbox cancellation made it impossible to ignore.

The Receipts

The numbers, when they came, were not flattering.

As of early 2026, roughly 3.3% of Microsoft's Microsoft 365 and Office 365 base was paying for Copilot. Not 3% of Windows users. 3% of the audience Microsoft was already selling productivity software to. Microsoft had reportedly invested more than $60 million in television advertising for Copilot in 2025 alone. The result was still single-digit penetration.

GitHub Copilot was the exception - roughly 4.7 million paying subscribers, genuine developer utility, real word-of-mouth. The rest of the portfolio did not replicate that result.

The internal evidence was worse than the external evidence. A Microsoft employee leaked a sentiment that the public data confirmed: "There is a delusion on our marketing side where literally everything has been renamed to have Copilot in it." A high-ranking executive separately called most Copilot AI tools "gimmicky." These were not disgruntled outsiders. These were people watching the strategy execute from inside the building.

The enterprise customers were sending their own signal. IT administrators began disabling Copilot entirely through group policies - not because the technology was broken, but because managing it was too complicated. Features appeared inside applications that users had not requested. Buttons arrived in Notepad. Logos appeared in Snipping Tool. An always-present colorful icon sat in the top-right corner of applications that people had used for decades and expected to behave predictably.

For IT teams, the reduced visible surface was itself the fix.

The feature that reduced support tickets was the feature's removal.

The Retreat Nobody Would Name

In March 2026, Pavan Davuluri, President of Windows and Devices, published what he called "Our commitment to Windows quality." It was framed as a quality initiative. What it was, translated from corporate: we went too far, and we are taking it back.

The commitments were specific. Microsoft would reduce "unnecessary Copilot entry points, starting with apps like Snipping Tool, Photos, Widgets, and Notepad." The Copilot logo disappeared from Notepad's toolbar. The generative AI functions stayed - they were renamed "Writing Tools" and moved to a right-click menu. Same capability. No logo. No name. No constant reminder that you were being observed by an AI product.

The Snipping Tool went further. The Copilot integration was removed entirely. Not rebranded. Gone.

Microsoft also rolled out a new enterprise policy - RemoveMicrosoftCopilotApp - that allowed IT administrators to uninstall the Copilot app from managed Windows 11 devices entirely. The company had spent two years integrating Copilot into the operating system. It then shipped a tool to help enterprises remove it.

The moment that revealed the internal reality most clearly happened the day Asha Sharma killed Gaming Copilot on Xbox. Jacob Andreou, Microsoft's Executive Vice President of Copilot - the person whose entire role was the Copilot brand - quote-reposted her announcement with the following: "It's critical that we remove Copilot from places where it doesn't live up to its promise."

Then he deleted the post.

The deletion is the most honest moment in the entire Copilot story. The internal mandate to clean up the mess was real enough that the EVP of Copilot publicly endorsed removing Copilot from Xbox. The corporate optics of admitting that were sensitive enough that the post disappeared within hours. Both things were true simultaneously. The company knew. It just could not say it.

Google Played the Same Script

While Microsoft was removing Copilot from Notepad, Google was reporting 750 million monthly active users for Gemini.

That number requires context. Gemini is embedded in Search, Gmail, Android, Google Workspace, Docs, Sheets, Slides, Drive, Chrome, and the Pixel hardware line. Google says Gemini Enterprise has sold more than 8 million paid seats across more than 2,800 companies, while direct API use of its first-party models has climbed to more than 16 billion tokens per minute.

The strategy Google used to get there is the same strategy Microsoft attempted. Embed AI in every product. Put it where users already are. Make the brand ubiquitous. There is no meaningful difference between "Copilot in everything" and "Gemini in everything" as a strategic statement.

The outcomes are not the same.

Microsoft's low single-digit Microsoft 365 Copilot penetration against a reported $60 million advertising push. Google's 750 million Gemini app users and broad enterprise seat growth. One company retreating from consumer surfaces. One company riding one of the fastest consumer AI adoption curves in the market.

The question is why the same script produced opposite results. The answer is not the model quality. It is not the brand. It is not the advertising. It is the surface.

Why Google Won and Microsoft Did Not

Google embedded Gemini into Search. Gmail. Android. YouTube.

These are not applications people open occasionally. They are applications people open reflexively, dozens of times per day, without thinking about it. Search is the front door of the internet for most of the world. Gmail is where professional communication lives. Android runs on billions of active devices. YouTube is where people go when they have seven minutes between meetings.

Gemini did not ask users to change a behavior. It was present when users arrived at behaviors they already had. The AI Overview appeared at the top of a search result the user was already conducting. The Gemini suggestion in Gmail appeared while the user was already writing an email. The on-device Gemini on Android was there when the user long-pressed the home button - a gesture they were already making.

Nobody adopted Gemini. They encountered it while doing something else. Encounter at sufficient frequency becomes familiarity. Familiarity becomes habit. The 750 million users did not decide to use an AI product. They were already using Google products, and Google had made those products AI products.

Microsoft embedded Copilot into Notepad.

Notepad is an application people open to paste something temporarily, write three lines, and close. It is the digital equivalent of a Post-it note. Nobody has a Notepad habit that generates forty daily opens. Nobody thinks about Notepad between uses. Nobody was looking for an AI product when they opened Notepad to paste a URL.

The colorful Copilot logo that sat in the top-right corner of Notepad was not annoying because Copilot was bad. It was annoying because it was present in a context where no user had expressed a need, on a surface with insufficient daily gravity to generate familiarity through repetition, on a product where the expectation was simplicity and the delivery was an AI assistant nobody asked for.

Xbox is the same story. Gaming has AI applications - Automatic Super Resolution, which improves image quality and performance in the background, is one that Sharma specifically cited as AI done right. Background rendering enhancement is invisible. It makes the game look better and the user does not have to do anything. Gaming Copilot was the opposite: a chatbot the user had to address, on a surface where the user had come to play a game, solving a problem they could already solve by searching the internet.

One surface had daily gravity. The other did not. Google understood this intuitively because its surfaces are the gravity. Microsoft forgot that its surfaces were not.

The Lesson That Predates AI

The dynamic that determined Copilot's fate is not an AI story. It is a distribution story that has played out across every technology wave.

Microsoft Office became the dominant productivity suite not because it was the best word processor or the best spreadsheet. It became dominant because it was bundled with Windows, which ran on most of the world's computers, which people used every day. The distribution surface had gravity. The product arrived inside it.

Internet Explorer held over 90% of the browser market not because it was the best browser. It was bundled with Windows. Users did not choose Internet Explorer. They encountered it when they turned on their computer. Google Chrome displaced it by making the installation frictionless enough that users who had a reason to switch actually switched - and then Search, Gmail, and YouTube gave Chrome daily gravity that IE never had.

Google Search has dominant global market share not because it is that much better than Bing. It has it because it was the default on Android, on Chrome, on most browsers worldwide, embedded in the surface where people already were when they wanted to find something.

The pattern is consistent. Products that succeed by distribution succeed because they arrive inside surfaces with enough daily gravity that encounter becomes habit. Products that fail by distribution fail because they are placed on surfaces where users arrive with a specific expectation and the AI product is a non-sequitur relative to that expectation.

Gemini arrived inside Google Search. The expectation when a user opens Google Search is: I want to find something. Gemini helped them find it. The AI matched the surface's gravity.

Copilot arrived inside Notepad. The expectation when a user opens Notepad is: I want to write a few lines of plain text. Copilot offered to help them write differently. The AI contradicted the surface's expectation.

The technology was not the variable. The surface was.

Where Microsoft Goes From Here

The retreat from consumer-facing Copilot is not a retreat from AI. Every source inside and outside Microsoft is clear on this. The AI capabilities remain in Notepad - they are called Writing Tools now. The Copilot brand is being moved to enterprise, where Microsoft 365 Copilot targets the back-office workflows - document automation, meeting summarization, multi-step agentic tasks across Outlook and Teams - that actually match the surface's daily gravity.

Microsoft 365 Copilot now supports third-party models including Claude Opus 4.7. The M365 E7 subscription tier at $99 per user per month bundles Copilot as part of an agent-centered enterprise offering. GitHub Copilot, with 4.7 million paying subscribers, continues to grow. Azure AI services are growing at rates that make the consumer Copilot disappointment look like a footnote in the quarterly report.

The strategy moving forward is a divergence: consumer surfaces get invisible AI, enterprise surfaces get branded Copilot. Automatic Super Resolution on Xbox. Writing Tools in Notepad. Agentic workflows in Teams. The AI that users experience as a better product rather than as a product they are expected to use.

This is the lesson Microsoft learned the hard way that Google understood instinctively: users do not want a co-pilot. They want to fly the plane themselves. When the AI helps them fly it better without drawing attention to itself, they accept it. When the AI asks to be acknowledged, they resent it.

The gaming community's reaction to the Gaming Copilot cancellation was relief. Not mourning. Not nostalgia. Relief that the chatbot was gone and the game was still there.

That reaction is the clearest data point in the entire story. Fourteen months of engineering, beta testing, and brand investment, and the most common user response to its cancellation was that they had not noticed it was there and were glad it would not be in the future.

The best AI is the one you do not know you are using. Microsoft spent two years learning this. Google started there.

Sources:

Previously on TheQuery: OpenAI's Deployment Company Is About Execution, Not Hype and The Defenders Just Got Better Tools. So Did Everyone Else.