Should your AI assistant live inside your apps, or hover above them?
Anthropic just answered that question with its latest Claude Opus update. The April 16 release of Opus 4.7 quietly added "Claude in PowerPoint" to its system prompt—a slides agent joining the already-listed Chrome browsing agent and Excel spreadsheet agent. But calling this a PowerPoint feature misses the point. This is a platform bet.
The real news isn't that you can now ask Claude to make slides. It's what that integration implies when you look at the full picture: Anthropic is building an AI infrastructure layer that sits inside the tools you already use. The system prompt explicitly states that "Claude Cowork can use all of these as tools," meaning a single conversation can hop from Excel to PowerPoint to Chrome without you lifting a finger.
Think about what that unlocks. Right now, most AI assistants work like a consultant you brief before a meeting—they take instructions, produce output, and hand it back. What Anthropic is constructing looks more like a personal operations team embedded directly in your software stack. The agent doesn't just read your spreadsheet; it manipulates the actual application, which means it can automate workflows across tools that have never talked to each other before.
Simon Willison, who tracks Anthropic's system prompt changes closely, noted the addition alongside several other updates: expanded child safety instructions, a directive to stop being "pushy" about ending conversations, and a new acting-versus-clarifying framework that encourages Claude to just try things when details are vague. These aren't just behavioral tweaks. They're the scaffolding for an AI that takes initiative across your digital life.
Anthropic is notably the only major AI lab publishing system prompts, which turns what could be opaque model updates into observable research. Their archive stretches back to Claude 3 in July 2024, enabling exactly this kind of transparency. The PowerPoint agent's appearance there suggests it's either live, in beta, or imminent. Anthropic hasn't announced a separate product launch—but the system prompt is its own announcement.
The competitive implications are significant. Microsoft is building its own AI layer across Office. Google is doing the same with Workspace. Anthropic is betting it can be the neutral infrastructure that connects them—or at least own the interface that users prefer. The difference is philosophical: Microsoft builds AI for Office, Anthropic builds AI that uses Office.
Whether this platform strategy pays off depends on execution. Tool integration is hard. PowerPoint's object model is notoriously gnarly, and reliability across versions matters. But if Anthropic pulls this off, the chatbot era ends not with a whimper but with an operating system inside your applications.