Microsoft is tearing up the single-AI rulebook. Its Copilot assistant is no longer married to one language model: the company has quietly begun routing user requests through multiple AI engines—including Anthropic’s Claude—depending on the task, according to official documentation and recent reports by technology columnist Paul Thurrott. The move, which adds automated model selection to Word, Excel, and PowerPoint in some regions, transforms Copilot from a standard chatbot into an orchestration layer for an expanding family of artificial intelligences. Meanwhile, down in Cupertino, another narrative is unfolding with equal urgency: Apple’s new budget MacBook Neo is selling faster than anticipated, stretching delivery times to three weeks and raising fresh questions about whether the Mac can finally break into double-digit PC market share.

What Actually Changed Inside Copilot

The most concrete change is visible inside Microsoft 365. In supported regions, Copilot in Word, Excel, and PowerPoint can now tap Anthropic models alongside OpenAI’s GPT series. Thurrott’s reporting highlights that Microsoft’s own documentation now describes Copilot as a “multi-provider environment” where administrators can govern which models are used. That’s not merely a backend tweak; it signals a strategic decision to stop treating AI as a single-model product.

Microsoft has also updated Copilot Studio, its agent-building toolkit, to let developers select specific models—including external ones—for the bots they create. This means an enterprise might design a customer-service agent powered by a model fine-tuned for compliance, while a creative-writing assistant uses a different, more imaginative engine. Under the hood, the system increasingly auto-routes queries to the right intelligence without exposing the sausage-making to users.

At the same time, Microsoft is developing its own line of foundational AI models. Rather than a single “MAI” competitor to GPT, it’s building task-specific engines: MAI-Image-2 for visual work, MAI-Voice-1 for speech, and MAI-Transcribe-1 for audio transcription. These are destined to power first-party Copilot features, filling gaps that generic large language models struggle with.

The user-facing impact of this orchestration push is still taking shape. Microsoft has been tinkering with how it presents model choices inside Copilot interfaces—in some cases removing selectors entirely so the system simply chooses on its own. That’s a bet on simplicity over transparency, and it’s bound to irritate power users who like to fine-tune.

What It Means for You (By Audience)

For everyday Windows users, the ideal outcome is invisible: you ask Copilot to draft a report, summarize an email thread, or analyze spreadsheet data, and it just does it better than before, without you ever seeing a model name. Early signs suggest the combination of models can improve context handling or creative output, but experience will vary by task. The main risk is confusion if quality yo-yos because the orchestration layer misfires.

Power users and enthusiasts who have enjoyed tweaking model selectors may find the loss of control frustrating. However, Microsoft appears to be leaving the option for manual selection in some Copilot experiences, at least for now. If you rely on specific model quirks—say, Claude’s nuanced writing style versus GPT’s broader knowledge base—you might need to test whether automated routing picks the right tool for your particular workflow.

IT administrators gain new levers but also new headaches. The Copilot admin center now lets organizations set policies around third-party model usage, including Anthropic. Data residency, compliance, and cost controls become more complex when a single prompt could get dispatched to a model hosted by a different company. Admins should review service plans and consider whether to enable these multi-model capabilities immediately or wait for clearer safeguards. Microsoft’s documentation is a must-read before flipping any switches.

Developers and citizen makers using Copilot Studio can start building agents that exploit model diversity. The option to pick Anthropic’s Claude for certain natural-language tasks could accelerate agent quality for specific niches—but it also means you’ll need to plan for a multi-vendor AI stack, similar to how you’d plan a cloud architecture today.

The MacBook Neo Factor

While Microsoft rewires Copilot, Apple is making a classic hardware-and-price play. The MacBook Neo, a deliberately stripped-down laptop aimed at the $700–$900 sweet spot, has seen surprisingly strong demand. As Thurrott noted on the MacBreak Weekly podcast, delivery times for the Neo have stretched to two to three weeks, suggesting supply is struggling to keep up with preorders and early sales. That’s a rare shortage for an Apple Mac, typically a supply-chain master.

Apple’s gambit is straightforward: lower the barrier to entry far enough that the “halo effect” from iPhones and iPads converts more buyers into Mac owners. The existing MacBook Air line already starts at $999 ($899 for education), but the Neo undercuts even that by using older A-series chips and a minimalist design. Rumor has it a future iteration could jump to an A19 Pro and 12 GB of RAM, smoothing over the current 8 GB bottleneck. For now, the Neo is a volume play, not a performance champ.

IDC data cited by Thurrott puts Apple’s 2025 PC unit sales at 25.6 million, good for about 9 percent global share—and a growth rate of 11.1 percent year over year. Dell, sitting at number three with 14.4 percent share on 41.1 million units, is growing more slowly (5.1 percent). If Apple keeps accelerating while Dell’s commercial-heavy portfolio hits a refresh lull, the market-share math starts to shift. A three-point swing in either direction—say, Dell falling to 11 percent as Apple climbs to 12—would reshuffle the top-three vendor rankings, even though that delta represents millions of units.

How We Got Here

Microsoft’s pivot toward orchestration isn’t random. Over the past three years, the company has swung from an exclusive OpenAI partnership to a more promiscuous model strategy. The original Bing Chat bet hard on GPT-4, then Copilot extended that exclusivity into the Office apps. But as Anthropic, Google, and open-source models matured, it became clear that no one engine excels at everything. Microsoft’s DNA leans platform: just as Windows succeeded by supporting thousands of hardware configurations and third-party software, Copilot is becoming a platform where the best model wins per job.

The chaotic rollout of AI features—Copilot icons appearing everywhere, then being pulled back after customer complaints—has masked a coherent long-term architecture. Thurrott’s analysis frames this evolution as Microsoft “trying not to do the usual Microsoft thing” of overbuilding a platform before shipping anything. This time, it shipped Copilot fast and is now retrofitting a platform underneath. The result is messy, but the direction is clear.

Apple’s market-share stagnation, meanwhile, is a decades-old riddle. The Mac has been excellent for 20 years—since Mac OS X Tiger—yet never cracked double-digit global share. The reasons are structural: enterprise procurement favors the Windows ecosystem’s compatibility, manageability, and vendor diversification. Consumers are more emotional, but Apple’s premium pricing kept a ceiling on volumes. The MacBook Neo attacks that ceiling head-on, at a moment when component shortages and memory costs are squeezing budget Windows laptops. Add in iPadOS 26 turning iPads into “real computers,” and Apple is building a multi-device assault on the PC market’s lower tier.

What You Should Do Now

If Copilot orchestration is active in your organization, check your Microsoft 365 admin center’s Copilot settings. Confirm which models are allowed, understand data routing, and communicate any compliance implications to your security team. If you’re an individual user, you don’t need to do anything—the transition is gradual—but it’s wise to pay attention to output quality. If you notice Copilot responses shift in tone or accuracy, Microsoft’s feedback button is the primary lever you have to influence routing improvements.

On the hardware side, the MacBook Neo’s success doesn’t demand immediate action from Windows users, but it’s a reminder that price competition is heating up. If you’re in the market for a budget laptop, keep an eye on Windows PC makers’ responses: Dell, HP, and Lenovo may respond with sharper promotions, especially as they clear out older stock ahead of back-to-school season. The component shortage that Thurrott references could temporarily inflate Windows laptop prices; if you need a machine soon, compare carefully rather than assuming Apple is the pricier option.

Developers experimenting with Copilot Studio should start testing agent behavior with multiple models. Anthropic’s strength in nuanced text generation might give your internal tools an edge, but model-aware design will pay off as the orchestration layer matures.

Outlook

Microsoft will continue to blend its own foundational models into Copilot over the next six months, gradually reducing dependency on external AI providers for core functions. The user experience should become more seamless—model selectors will likely disappear from mainstream Copilot surfaces, and the assistant will feel more like a single entity even as it draws on a dozen engines. The risk is that the company repeats the “ribbonization” over-complication that plagued earlier Office redesigns; if the AI layer remains mystifying, user trust could erode.

Apple, meanwhile, is in an uncharacteristic position: chasing volume. If the MacBook Neo becomes a permanent fixture with refreshed specs, and if the component shortage persists, the Mac could finally break the 10 percent market-share barrier this year. That won’t dethrone Windows, but it will force PC makers to defend their turf more aggressively—likely by leaning on AI features that differentiate their hardware, which in turn will accelerate Microsoft’s Copilot integration.

The true battle is not about whose silicon is faster or whose model is bigger; it’s about which company can hide complexity better. Apple has always excelled at making the complex feel simple; Microsoft is now trying to do the same with AI. For the rest of us, that means smarter software, sharper competition, and—if we’re lucky—fewer model-selector dropdowns to click.