Google once went to antitrust regulators to pry open Microsoft’s desktop search. Nearly 20 years later, hidden code shows it building an AI layer across Windows itself.
By [Ryan Merket](/author/ryan-merket)
· Published
RUNTIMEWIRE INVESTIGATION — Scoop
Original reporting by RuntimeWire, based on reverse engineering, documents.
Why it matters #
Microsoft owns Windows. Google owns the consumer reflex to “Google it.” AI gives Google a chance to extend that habit from the web into files, apps and whatever is happening on a user’s screen.
Reporting record #
Finding
RuntimeWire established that a Google Windows executable compiled on September 6, 2026 contains unreleased infrastructure for continuous Search Live conversations, mouse- and window-triggered AI, File Explorer integrations, selectable tools and models, structured screen extraction, and an Agent Task mode.
How we verified
Methods: reverse engineering, documents.
The strongest evidence is the examined executable's PE timestamp of 2026-09-06 17:33:27 UTC, SHA-256 hash 2c7c0bff4ac5ce6cff16c2cc067b365f29a6c7d6ce81e671023e12d7307cb50a, feature flags, implementation names, Windows integration paths, tool and model identifiers, and Chromium strings describing a system-wide Search with Chrome surface.
RuntimeWire examined and reverse-engineered the Google Windows executable, including its PE metadata, feature flags, code and interface identifiers, error states, Windows event handling, shell integration paths, tool and model catalogs, and content-extraction structures, then compared those findings with Google's public Windows and mobile product documentation and Chromium strings.
Tested versions: Google for Windows / WGA version 1.0.21.1.
Reproduction
RuntimeWire partially reproduced the finding.
Reverse engineer the latest google.exe binary from September 6th, 2026
File hashes
SHA-256 hash 2c7c0bff4ac5ce6cff16c2cc067b365f29a6c7d6ce81e671023e12d7307cb50a : google.exe
Company response
The company was not contacted before publication.
Nearly 20 years ago, Google complained to U.S. antitrust regulators that Microsoft was using Windows Vista to disadvantage Google Desktop Search. The Justice Department intervened, and in 2007 Microsoft agreed to let users and PC makers choose a competing desktop-search product and have Windows respect that choice.
Now Google is back inside Windows, with ambitions that appear to extend well beyond desktop search.
RuntimeWire examined a Google Windows executable compiled September 6th, 2026 that contains unreleased infrastructure for continuous Search Live conversations, mouse-triggered AI tied to a specific window, File Explorer actions, selectable AI tools and models, richer screen understanding and an Agent Task mode.
Google's Windows app still presents itself primarily as a fast way to search from anywhere on a PC. The code underneath points toward something closer to an AI layer that can follow users across the operating system.
Compiled code reflects development work rather than a release commitment, and Google could alter or abandon any of these features before they reach users. Taken together, however, the pieces show Google building more ways to summon its assistant from wherever the user already is while bringing the user's immediate context into the interaction.
That creates a strategic problem for Microsoft. Google does not need to replace Windows to become important at the operating-system layer. It needs to become the service people instinctively reach for when they want to find, understand, create or eventually do something.
Point at a window and talk to Google
One of the clearest unreleased features would let users trigger a live AI conversation about a specific Windows application with a mouse gesture. The exact final interaction cannot be determined from the build, but the architecture points toward a simple idea: point at a window, press a mouse button and start talking to Google about it.
That matters because context remains one of the biggest sources of friction in consumer AI. Users still have to open an assistant, describe what they are viewing, attach a file or deliberately share a screen. Google appears to be working toward making the act of invoking its assistant identify the context at the same time.
The build also contains a broader Search Live implementation for Windows, with support for continuous voice conversations and visual input. That suggests Google is adapting the live experience it already offers on mobile for use across the desktop.
The consumer promise is simple: less setup before the AI can help. Instead of opening Google, selecting a window and explaining what is happening, a user could summon Google in place and begin the conversation from there.
Google is moving outside the Google app
The same pattern appears elsewhere in the executable. Google is building integrations that would let users invoke its AI from Windows surfaces rather than requiring them to return to the floating search box.
The build includes infrastructure that could place an "Ask Google about this file" command inside File Explorer for images and PDFs. It also points toward richer understanding of shared windows through their text, documents and other structured content rather than relying entirely on screenshots.
The individual features are modest. Their common direction matters more: Google wants to meet users at the file, window or system surface where the question already exists.
A PDF would not need to be dragged into an AI app. A window would not need to be manually reintroduced as context. A Windows setting would not necessarily have to be found through Microsoft's own search. Google is building toward a model where its assistant is available from the object itself.
Alt+Space is becoming an AI launcher
Google's hidden tools make the broader ambition clearer. The Windows build can accommodate multiple AI modes, including Deep Search, Canvas, image generation, Deep Browse and Agent Task, as well as different Gemini models.
Those names could change before release, but the architecture shows Google preparing its Windows interface to move among specialized AI tools rather than remaining a single-purpose search box.
That changes what Alt+Space can become.
Today the shortcut opens Google so a user can find something. The emerging product could make the same gesture an entry point for research, content creation, image generation, live assistance and task-oriented AI.
Search gives Google a strong distribution wedge because the behavior is already familiar. The company can add capabilities behind an existing consumer habit instead of teaching users an entirely new interaction.
Then there is Agent Task
The most consequential unreleased mode is Agent Task. The Windows app treats it as a selectable AI tool, establishing that Google is preparing the interface to host an agent-oriented experience.
RuntimeWire did not find evidence that this build can autonomously click, type or operate native Windows applications, so describing it as full computer control would go beyond what the executable supports.
The surrounding work is still worth watching. The build points toward better understanding of webpages, documents and windows as structured content rather than flat images. That kind of context could eventually help an agent distinguish controls, forms, text and relationships on a screen.
For now, Agent Task is best read as a direction marker. Google is preparing a Windows interface that can host an agent-oriented mode while improving how much context the system can understand. Microsoft already has most of the ingredients
That is what makes Google's push awkward for Microsoft. Copilot on Windows already supports voice, file search, screenshots, Windows Settings assistance and system-level invocation, while Copilot Vision can inspect shared applications and guide users through tasks.
Microsoft is hardly missing the technology. The vulnerability is that the consumer experience remains in motion.
Copilot has appeared as a sidebar, a standalone app, a taskbar destination, a keyboard key and a compact overlay. On August 14th, Microsoft temporarily disabled Ask Copilot on the taskbar for Windows Insiders while it reworked the experience. Notebooks, Connectors and Vision also became temporarily unavailable for some testers during the preview.
Earlier in the year, Microsoft removed generic Ask Copilot buttons from Photos and Snipping Tool as part of an effort to reduce unnecessary entry points and make AI integration across Windows more deliberate.
Those choices may be sensible product cleanup. They still create a contrast with Google, which appears to be experimenting with more ways to summon its assistant from a keyboard, a mouse, File Explorer and whatever is already on the screen.
Google's wedge is much easier to explain: press Alt+Space and Google it.
From there, the company can expand the meaning of that interaction gradually. Web search becomes local and Drive search. Screen context becomes live voice. Specialized tools appear inside the same interface. Agent Task waits in the menu. Microsoft has spent years teaching consumers what Copilot is supposed to be. Google can begin with a verb consumers already understand.
This fight is almost 20 years old
There is a historical irony to Google trying to become a search and AI layer across Windows because the company once fought Microsoft over almost the same territory.
In 2007, Google complained to U.S. antitrust regulators that Windows Vista's built-in desktop search put Google Desktop Search at a disadvantage. Microsoft ultimately agreed to changes that allowed third-party desktop-search products to register as the Windows default.
That dispute was part of a larger distribution struggle. Google had also objected to Microsoft's handling of search inside Internet Explorer, arguing that control over Windows and the browser could steer users toward Microsoft's own services.
Then came Chrome
Google launched Chrome in September 2008, initially on Windows, creating its own route to users instead of depending on Microsoft's browser. Chrome eventually became one of Google's most consequential answers to Microsoft's distribution advantage.
The pattern is familiar: Microsoft owns the operating system, and Google builds another layer on top of it that consumers choose to use. AI creates an opening for Google to do that again.
There is another antitrust reversal in the background. In 2025, a federal court barred Google from entering or maintaining certain exclusive distribution contracts involving Google Search, Chrome, Google Assistant and Gemini after the Justice Department won remedies in its search-monopoly case.
There is no evidence that Google's Windows app was built in response to those remedies. The strategic value of direct consumer adoption is nevertheless clear: if Google can build an AI experience people deliberately summon, it relies less on being installed or designated as the default by someone else.
Twenty years ago, Google wanted regulators to ensure Microsoft could not use Windows to lock it out of desktop search. Now Google is trying to become a preferred search and AI layer inside Windows on its own terms.
The next fight is over intent
The old boundaries between web search, desktop search, document retrieval, application assistance and operating-system help are beginning to collapse. All of them increasingly fit inside the same conversational interface.
Google does not need to replace Windows Search, the Start menu or File Explorer. It needs to capture the moment when the user decides they want help.
Historically, Microsoft owned much of that local intent through Windows itself, while Google dominated the web query. AI makes those categories less distinct. A single prompt can ask about a webpage, a local file, an application, a PDF, something visible on screen or a task the user wants completed.
If asking Google becomes the easiest way to move across those categories, Microsoft can retain control of Windows while losing some of the interface through which people increasingly operate it. Chrome was Google's answer the last time Microsoft controlled a critical gateway to users. AI may become the next one.
Microsoft can keep the Start button. Google is building toward becoming the button people actually press.