Microsoft CEO Satya Nadella wants your PC to become “an infinite software factory” in the final blow to the app store concept, but I fear companies won’t bite the AI apple.
- Microsoft CEO Satya Nadella says Copilot Code and local AI make every Windows PC an infinite software factory without spending cloud tokens
- Instead of killing app stores, Microsoft is moving app hosting to Copilot, while IT leaders are wary of citizen-created tools.
- Microsoft’s own engineers say that local inference does not disconnect sessions and routing will ship later this month.
Microsoft CEO Satya Nadella has a new pitch for the Windows PC and takes aim at one of the most lucrative ideas in consumer technology.
In a post on
He wrote this idea straight out and said, “Your PC becomes an infinite software factory.”
Latest videos ofTechnologyRadar
Today marks a new chapter for Windows, as we bring unlimited intelligence to every desktop and every home, and make every PC a place where agents can work securely on your behalf. Some highlights of what we announced: • MAI-Code-1.1 Flash: 137B parameter encoding model with 256K context window, which is now optimized to run on your PC. • GitHub Copilot now pushes work to local models like MAI-Code-1.1 Flash, helping projects cost much less without sacrificing quality. • With Hybrid Intelligence, Copilot can now take action directly on the PC and keep work confidential on your device. • And with Code in Copilot, you can basically create any software you need on your desktop, without spending cloud tokens, and it’s just fantastic. You’re no longer limited to what’s in an app store! Your PC becomes an infinite software factory. • Security is key to all of this, so we’re also bringing Windows and Agent 365 together so agents can work within secure boundaries on the device, including MXC, a local sandbox for running agents. Windows becomes your secure agent box! • All of this comes to life on a new generation of devices, like the Surface Laptop Ultra, powered by NVIDIA RTX Spark. I can’t wait to see what you build with all of this.October 7, 2026
From the Gates factory to your desk?
The narrative is a surprising line from the head of a company that sells more software than almost anyone.
But read Microsoft’s engineering documentation for the same release, and the infinite factory starts to look pretty finite: It works better on a $5,899 laptop, it still reaches the cloud, and the companies Microsoft most needs to convince have spent the past year telling researchers they’re worried about exactly this kind of thing.
More importantly, the factory metaphor is not new to Nadella: in a In his 2025 staff memo, he recalled that Bill Gates had envisioned “not just a software company, but a software factory,” and argued that the age of AI demanded a rethink.
At a recent customer meeting, the CEO said Microsoft had progressed from code support through agent coding “to software factory, all in a year,” and now wants customers to open their own software factories, spending more with Microsoft along the way.
The October 7, 2026 release builds on this ambition: A Windows Experience blog post by Windows and Devices head Pavan Davuluri describes hybrid intelligence, where Copilot reads local files with permission, acts on Windows, and transfers work to device models.
Copilot’s Code experience creates an app, dashboard, or workflow from a plain-language description, and on Windows, Microsoft says it can produce native apps from a single message. Jared Spataro, director of marketing for Microsoft’s AI at Work, argued that creating these types of tools is becoming “a skill as basic as writing a memo or modeling a budget.”
The underlying engine is MAI-Code-1.1 Flash, Microsoft’s internal 137 billion-parameter encoding model, which according to Nadella’s post is now optimized to run on a PC. There’s a problem, though: Even though Microsoft describes it as a situation where consumers are finding that their AI needs “exceed what their cloud budgets can support.”
Microsoft’s own engineers are more informed about what the process entails, saying that GitHub Copilot’s automatic mode decides on a task-by-task basis whether the work runs locally or in the cloud, and a title of the diagram is telling: “local inference does not disconnect the session.”
That route won’t ship until late October 2026; Copilot’s hybrid features will arrive on Copilot+ PCs in the coming months, with Code rolling out first through Microsoft’s Frontier early access program. This is interesting, but somewhat predictable, because Nadella’s approach would, as stated, cannibalize Microsoft’s token-based business model.
An approach that redirects the most difficult tasks to the Azure cloud makes a lot more sense by preserving compute wherever possible while putting Microsoft cloud computing at the center of things.
Hardware is another factor one might be overlooking here: According to the command-line post, the on-device version of MAI-Code-1.1 Flash is quantized up to 53GB, roughly an 80% reduction from the cloud model, and maxed out at 75.5GB of memory in full context on Microsoft’s test machine, a Surface Laptop Ultra built on Nvidia’s RTX Spark.
That’s important because eight Surface Laptop Ultra configurations are at play; only one has more than 64GB of memory, according to TweakTown: the $5,899 128GB model, which costs $3,300 more than the base model and isn’t yet available to purchase, meaning potential users can’t really test how well it does compared to cloud-based solutions at a time when all performance-related numbers come from Microsoft.
The death of modern enterprise software? Not yet
Despite the hardware limitations at play, it could signal a looming death knell for most basic enterprise-grade software solutions, even as more ambitious ports have already been in play: developer Brandon Thomas, for example, just released seven open source alternatives to Adobe Creative Cloud that it says it created with Anthropic’s Claude Opus 5.5, declaring that “the software is over.”
Its own roadmap, however, places the Photoshop replacement at between a quarter and a third of what a professional needs for daily work. Creating software has become cheap. Creating software that anyone can trust hasn’t.
The biggest problem will be enterprise IT, which I suspect will keep its doors closed for now; Companies don’t just buy features; They buy someone to call when it fails, a patch schedule, an audit trail, and a vendor who takes responsibility.
Getting an open source or AI-assisted application that someone designed without any responsibility that could break things, create security risks, and may not include future software updates that fix any of the problems.
Survey data points in the same direction: A June 2026 Retool study of 307 CTOs, CIOs and CISOs found that 93% were concerned about vibration-coded tools reaching production, although Retool sells an internal tooling platform presented as the governed alternative.
A Harris Poll for Dataiku survey of 600 CIOs found that 81% were concerned that citizen-built AI could expose sensitive data, and more than half had already discovered rogue AI in their organizations.
Meanwhile, Gartner predicts that more than 40% of agent AI projects will be scrapped by the end of 2027, with analyst Anushree Verma describing many as “mostly hype-driven and often misapplied.”
Microsoft’s own release materials quietly concede the same, and even take steps to ensure that agents are not run fraudulently while building software or solutions on the device.
Command-line publishing supports that Copilot’s built-in file tools are controlled by policies within the application rather than isolated by the operating system, and GitHub requires enterprise and enterprise administrators to enable a policy before staff can use MAI-Code-1.1 Flash, while Microsoft’s runtime containers, now widely available, allow IT to define which agents can read, modify, or execute.
The real test for Nadella’s factory is not individuals with a $5,899 laptop and a tolerance for alpha software, but rather companies willing to take the risk and trust Microsoft’s safeguards. Until then, one could assume that the AI factory narrative will continue to fall well short of its advertised potential.
Follow TechRadar on Google News and add us as a preferred source to receive news, reviews and opinions from our experts in your feeds.
Post Comment