Microsoft CPO Aparna Chennapragada on AI at work: 100M Copilot users, reasoning models shifting enterprise behavior

Summary

Microsoft has crossed 100 million Copilot users across commercial and consumer segments, and more than 90% of Fortune 500 companies are now using some form of its AI products. Aparna Chennapragada, Microsoft's Chief Product Officer for AI at Work, frames the current moment as analogous to the first wave of mobile apps, where the initial instinct is to bolt a chat interface onto existing tools before deeper integration follows.

The more telling signal is what Chennapragada calls the 'frontier firms' divide. A subset of enterprise customers has moved beyond treating Copilot as an off-the-shelf chatbot purchase and is actively building, buying, and partnering around AI. Those firms, she argues, are extracting materially more value. The inflection point she identifies is January–February 2025, when reasoning models came online and triggered a clear behavioral shift among this cohort, including at Microsoft itself.

Gross margin pressure from reasoning models is an acknowledged reality. Notion's margins reportedly dropped from roughly 90% to 80% as inference costs rose, a dynamic Chennapragada implicitly validates by cautioning that deploying frontier reasoning models for routine tasks is economically wasteful. Her framing: using GPT-5 in thinking mode for a simple workflow is like 'taking a flying car to a grocery store.'

On the product architecture question, Chennapragada is direct that both sidebar chat and deep in-product integration are necessary, not competing, strategies. She points to Equals Copilot, announced approximately two weeks before this segment, as an early example of embedding assistance directly into workflows rather than alongside them. The Office suite, including Word, Excel, PowerPoint, Outlook, and Teams, already operates as a platform with a plug-in ecosystem, and she draws a clear line between functionality broad enough to be a native feature versus vertical use cases, such as specialized accounting tools, that belong in add-ins.

Chennapragada identifies three structural shifts redefining how Microsoft builds software. First, the anthropomorphic quality of natural language interfaces means model upgrades register emotionally with users in ways a menu change never did. Second, release cadences have fundamentally changed, requiring behavioral evals and gradual rollouts rather than versioned release notes. Third, team composition is blurring, with the traditional engineer-designer-PM triangle giving way to a structure of generalists, model specialists, and go-to-market, with far less separation between the roles.

On LinkedIn, she notes that Ryan Roslansky is now more front-and-center on the Office integration side, and she positions many AI-enhanced professional networking ideas, such as persistent interest-based alerting and intelligent contact management, as concepts that were simply 'unfundable' before sufficient model capability existed. The implication is that Microsoft views LinkedIn as fertile ground for AI-native features that were previously impractical, not outside its strategic scope.