The buzziest news was Microsoft, NVIDIA, and Anthropic have announced new strategic partnerships aimed at enhancing access to Anthropic’s Claude AI model. Anthropic plans to scale its Claude AI on Microsoft Azure, supported by NVIDIA’s technology, and has committed to purchasing $30 billion in Azure compute capacity, with plans to expand up to one gigawatt. This partnership will allow Microsoft Foundry customers to access various versions of Claude, including Claude Sonnet 4.5 and Claude Opus 4.1. Notably, Anthropic will be the only frontier model available across all three major cloud platforms. Microsoft and NVIDIA will invest up to $10 billion and $5 billion, respectively, in Anthropic, further solidifying their collaboration in AI technology development. Despite these partnerships, Amazon remains Anthropic’s primary cloud provider.
Microsoft has released Microsoft Agent 365, a new management platform designed for overseeing AI agents, which is now available in early access through the Microsoft 365 Admin Center. Key features include the Microsoft Entra registry for managing agents with unique IDs, risk-based access policies, advanced security features, and performance measurement tools. The platform supports both Microsoft-built agents created in Copilot Studio and third-party AI agents from partners like Adobe and Databricks, as well as open-source agents on GitHub. This release represents a shift towards enterprise readiness for AI agents, allowing organizations to manage them in a unified system.
Microsoft has announced new AI capabilities for Windows 11, integrating AI agents into the taskbar. These agents can automate tasks such as researching data and accessing files while users work on other projects. The feature, called “Ask Copilot,” combines local file search with AI capabilities, allowing users to interact with Microsoft 365 Copilot directly from the taskbar. The AI agents operate in a separate workspace for security, and the feature is designed to be optional for users who prefer not to engage with AI functionalities. No pricing details were provided, and the rollout is part of a broader initiative to evolve Windows into an “agentic OS.”\
Microsoft is introducing additional free AI features to its Office apps, including Outlook, Word, Excel, and PowerPoint, set to roll out in early 2026. These features include an enhanced version of Copilot Chat that can access content across an entire Outlook inbox, improving email management and meeting scheduling without requiring a paid subscription. The Agent Mode feature, which automates document and spreadsheet creation, will also be available for all Microsoft 365 subscribers, allowing users to utilize either Anthropic or OpenAI reasoning models. Microsoft plans to preview these features for subscribers by March 2026 and is launching a new Microsoft 365 Copilot Business plan aimed at small and midsize businesses for $21 per month.
Why do we care?
Microsoft is pushing AI agents from “fun demo” into “first-class enterprise citizen.” The key shift is governance. When you give an AI system an identity, permissions, telemetry, and lifecycle controls, you’re saying, “This thing is going to be part of normal operations.” And that’s a big deal. That’s where my interest in Agent 365 really hits. That’s a management platform for agents.
But there’s a bigger layer underneath the Anthropic partnership that’s easy to miss. Anthropic committing $30 billion in Azure compute — scaling up toward a full gigawatt — signals something essential: AI is becoming less about the model and more about the compute you can afford. A handful of hyperscalers now control access to the horsepower required to run frontier models. That means costs will keep rising, and strategy is increasingly dictated by whoever controls the power plants. And that’s assuming this all happens… while it’s another podcast, ask about model costs, data center growth, energy needs, and the like.
And that’s before we get to compliance.
The minute an AI agent has its own identity, it’s in the audit universe. Regulated clients are going to expect you to manage agent access, agent activity logs, and the separation of duties for software that acts like a worker. If an agent misclassifies a document, touches sensitive data, or triggers the wrong workflow, you will be asked to produce proof — logs, policies, and controls — the same way you do for user activity today.
Most MSPs don’t have that muscle built yet.
But there’s another shift happening underneath the surface: AI feature creep. Microsoft is about to give customers more AI for free inside Outlook, Word, Excel, and PowerPoint. And when AI shows up inside the apps they already use, clients won’t treat it like a premium add-on — they’ll treat it like table stakes. That raises expectations without raising budgets. They’ll assume it’s safe because Microsoft built it, they’ll expect it to “just work,” and when it doesn’t, they’ll call you. Suddenly you’re responsible for AI literacy, AI troubleshooting, and AI guardrails for tools you didn’t deploy and can’t fully control. This is where MSPs get squeezed: free AI raises the floor, not the ceiling, and your job becomes managing the expectations that come with it.
And with multiple models in the mix — Anthropic, OpenAI, whoever else — this gets messy fast. You need a model selection framework, you need governance policies, and you need language in your contracts to handle AI behavior. Because if an agent makes a bad decision, someone’s going to look at you.
This is your signal to formalize AI agent readiness as a service. Treat agents like employees. Manage their identity, permissions, and behavior. Build the standards now, while everyone else is still reacting to the shiny Ignite announcements. Is Agent 365 the answer? Figuring that out is key.
This is the point where AI stops being hype and starts becoming something we actually have to run.