Perplexity is taking its biggest swing yet at the place where most office work actually happens: Microsoft’s productivity apps. With its Computer AI agent orchestrator now reaching into Word, Excel, PowerPoint, and Outlook, Perplexity is effectively trying to sit alongside – and sometimes above – Microsoft’s own Copilot as the brain that quietly runs your workday in the background.

If you haven’t been tracking it closely, Computer is the Perplexity product that turns the company’s “answer engine” reputation into something much more ambitious. Instead of just responding to questions in a chat box, Computer acts like a general-purpose digital worker that takes a goal, breaks it down into tasks, and then carries those out across your apps, files, and the web. Under the hood, it orchestrates a whole fleet of AI models – 19 of them – routing each subtask to whatever model is best suited, then running those workloads in parallel until you get a finished result. Think of it less as “one chatbot” and more as a coordinator that manages a team of specialized agents so you don’t have to.

That design becomes much more interesting once it’s allowed to live inside the tools people already use all day. Microsoft 365 Copilot is already embedded in Word, Excel, PowerPoint, Outlook, and Teams, and it has done a good job normalizing the idea of AI assistance inside Office documents and email. But Copilot is still, at its core, an assistant: you ask it for a summary, a draft, a formula, or a slide outline, and it responds within the context of that document or spreadsheet. Computer is aimed at something subtly different: not just helping inside a document, but coordinating entire multi-step workflows that stretch across many documents, inboxes, and data sources at once.

You can already see how Perplexity is thinking about this from its earlier move into Microsoft Teams. Computer for Teams was pitched as a way to bring its AI agent orchestrator “deeper into Microsoft’s product ecosystem,” letting it search the web, work across connected tools, generate content, analyze business data, and build reports and presentations right in the space where teams collaborate. Extending that same orchestration layer into Word, Excel, PowerPoint, and Outlook is the logical next step: instead of having to bounce between chat, browser, and desktop apps, the agent can tap directly into the files and messages where the actual work lives.

Take Excel as a starting point. Microsoft 365 Copilot can already analyze a spreadsheet, suggest charts, and surface trends, and for many people, that’s more than enough. With Computer in the mix, the expectation shifts from “help me understand this sheet” to “go build the entire reporting loop around this data.” In practice, that could look like Computer pulling raw numbers from a CRM or finance system via connectors, cleaning and reshaping the data, generating a multi-tab Excel workbook with pivot tables and charts, and then setting up a weekly refresh cycle that emails the updated report to stakeholders in Outlook – all triggered from a single natural-language prompt.

The story in Word is similar but leans more into long-form content assembly. Perplexity’s roots as an answer engine mean it’s unusually good at real-time research, source citation, and synthesis across the web, which is now being handed off to Computer’s orchestration layer. Imagine kicking off a task like “Create a 12-page whitepaper on the state of AI in healthcare, with real-world case studies, recent regulatory changes, and charts summarizing investment trends.” Computer can spin up sub-agents to scour recent articles and reports online, analyze open data sets, draft sections in Word, generate and insert charts from Excel, and cycle through revisions based on your feedback – often running for hours in the background. Instead of treating Word as a blank page, Computer treats it as one endpoint in a larger pipeline that starts with a goal and ends with a polished deliverable.

PowerPoint sits at the intersection of those two worlds: structured data and narrative. Copilot today is pretty good at turning a Word document or a set of bullet points into a slide deck. Computer’s pitch is to take a broader brief – say, “Prepare a quarterly business review deck for the sales team” – and run the full play. It can pull numbers into Excel, generate charts, draft narrative sections in Word, and then assemble the actual slides in PowerPoint, complete with speaker notes and follow-up tasks, using multiple AI models tuned for reasoning, code, visuals, and layout. In that sense, PowerPoint isn’t just where you “ask for slides”; it’s where you see the outcome of a long-running workflow that may have touched a dozen other tools.

Outlook is arguably where the “digital worker” idea feels most concrete. Email is messy, repetitive, and often the glue between all the other work you do. Perplexity describes Computer as being able to create assets, manage emails, conduct research, and run tasks in the background, with scheduled or condition-based triggers. Once it has a foothold in Outlook, that could mean Computer quietly triaging your inbox, drafting responses, scheduling meetings, and following up on action items it previously generated in Word, Excel, or PowerPoint. You might start a task like “Handle everything related to this RFP,” at which point Computer could read the brief, coordinate internal questions over Teams, assemble a proposal in Word, build pricing models in Excel, craft a presentation in PowerPoint, and keep the email thread going in Outlook – without you acting as the bottleneck at every step.

All of this is possible because Computer isn’t limited to a single model or a single interface. Perplexity launched it as a multi-model agent platform that can coordinate 19 different AI models, routing parts of a task to whichever engine is best for reasoning, long-context recall, research, code, visuals, or other specialties. It also leans heavily on connectors and a secure cloud sandbox so it can interact with real apps and data while keeping enterprise controls around access and logging. Instead of you juggling different AI tools – one for writing, one for charts, one for research – Computer wraps them into a single orchestration layer that sits on top of your existing Microsoft stack.

That puts Perplexity in a fascinating competitive posture relative to Microsoft’s own AI story. Microsoft has gone all-in on the “Copilot” brand, with Microsoft 365 Copilot embedded across Word, Excel, PowerPoint, Outlook, and Teams, and Microsoft Copilot (formerly Bing Chat) acting as a general-purpose AI companion for questions, content generation, and image creation. Copilot is deeply integrated and benefits from first-party access to your documents and organizational graph, but it usually behaves like a single assistant per app or chat window. Perplexity, by contrast, is betting that the real value is in the orchestration layer that can pick and choose among many models, run workflows for hours or even months, and treat the Microsoft suite as one big surface it can operate on rather than a set of separate endpoints.

There’s also a pricing and positioning angle here. Perplexity Computer has been launched as a premium feature aimed at power users and enterprises, with reporting that it’s currently available to Perplexity Pro and Max subscribers, including substantial usage credits and controls for model selection and spending caps. That puts it squarely in the “high-end digital worker” bucket rather than “free add-on assistant,” which is where Microsoft Copilot and other consumer-facing tools often sit. For organizations that already pay for Microsoft 365 and maybe Copilot on top, the question will be whether Computer’s multi-model orchestration and cross-app autonomy actually translate into enough saved time or new capabilities to justify another premium AI subscription layered onto the stack.

From a safety and governance perspective, Perplexity is clearly aware that “an AI that can click around your apps and run for months” is the kind of sentence that makes CISOs nervous. That’s why Computer is framed as running in a secure sandbox with policy controls, logging, and a deliberate separation between the orchestration layer and underlying customer data. The move into the Microsoft productivity suite will likely intensify scrutiny around how those controls work in practice: Who can start long-running workflows? How do you audit what the agent did inside a Word document or Outlook inbox? How do you prevent a misconfigured prompt from triggering a cascade of unwanted emails or file changes? Those aren’t abstract questions once you let an autonomous system natively operate inside your critical business tools.