Julie Yack

Why Copilot sometimes shines in the newest products—and struggles a bit with the classics

As Copilot shows up across more and more Microsoft products, I’ve found myself fascinated by the differences in how mature the experience feels depending on where you use it. You’d assume Copilot would be strongest in the most established tools—the ones with decades of refinement behind them. But strangely, the opposite often feels true.

Word and PowerPoint, the foundation of so many careers and companies, sometimes have Copilot experiences that feel tentative. Meanwhile the newer parts of the Power Platform—Dataverse especially—feel like Copilot has been practicing for this moment its entire life.

It took me a while to untangle why that is. And once I did, the pattern became so clear that it actually made me more optimistic about the future of AI-powered work.

Here’s what I realized.

Word, PowerPoint, and the rest of the Office-era products were built to be almost absurdly stable. And they needed to be. For a long time, releases came every two or three years. Businesses depended on those apps behaving the same way, every day, for decades. Entire industries standardized on them. Those tools had to support every formatting quirk and every workflow variation ever discovered by millions of people across millions of documents.

That kind of stability is an engineering achievement. But it also creates a landscape where Copilot must tiptoe. Word’s formatting engine is a masterpiece of flexibility—but that means there’s rarely a single “correct” way to present anything. PowerPoint gives you infinite ways to tell a story visually, which also means infinite ways for Copilot to misread your intent. When Copilot enters these environments, it’s not just helping you write a sentence or place an image. It’s navigating thirty years of people doing whatever they wanted, however they wanted, with tools designed to let them do exactly that.

Now contrast that with Dataverse.

Dataverse was born in a completely different era—one where software updates weekly instead of annually, where APIs and metadata are first-class citizens, and where we assume from day one that people will want automation, integration, and AI-assisted building. Dataverse is structured by design. It’s opinionated in all the right ways. A table is a table. A relationship is a relationship. A field has a type. A rule has a purpose. There’s no ambiguity about what something means. Yes, Dataverse was born out of xRM, which had the older model of releases every few years. But, Dataverse still feels more modern, and orderly.

So when Copilot enters that world, it has an immediate advantage: Dataverse tells Copilot exactly what finished looks like.

If I describe a business scenario, Copilot doesn’t just guess at my intent—it can infer it. It can propose entities, relationships, cardinality, rules, and even potential downstream automation. It’s not doing magic. It’s working in an environment that gives it clarity, boundaries, and semantics. In Dataverse, intent isn’t something Copilot has to interpret between the lines—it’s built into the architecture.

This is where the real paradox reveals itself. It’s not that Copilot is inherently better in newer products. It’s that newer products are built for outcomes, while older products were built for artifacts. Dataverse is designed to help you build something functional. Word is designed to let you express something however you want. AI thrives in places where the goal is explicit. It struggles in places where the goal is “whatever you imagine.”

And there’s nothing wrong with that. It doesn’t mean Copilot is failing. It means it’s learning to collaborate inside environments designed long before AI was part of the story.

If anything, this paradox shows us where the future is headed. We’re not just bolting Copilot onto the tools we’ve always used. We’re quietly, steadily shifting toward products designed to be AI collaborators from day one. The more the architecture makes intent explicit, the more Copilot can help. And the more agile our tools become—not unstable, but intentionally agile—the more Copilot can evolve alongside them.

So yes, Copilot sometimes feels a little tentative in the classics and shockingly confident in the newcomers. But that’s not a problem. It’s an invitation. It gives us a way to understand how software needs to evolve to truly unlock AI’s potential. And it shows that maturity in an AI world isn’t measured by age—it’s measured by clarity.