Microsoft VP: AI agents are the next public cloud for startups

Amanda Silver, a 24-year veteran at Microsoft, argues that the rise of agentic AI represents a watershed moment for startups. Silver, who previously worked on the GitHub Copilot coding assistant and now serves as corporate vice president at Microsoft’s CoreAI division, compares the shift to the historical transition to public cloud computing. In her view, just as the public cloud eliminated the need for startups to invest heavily in physical hardware and real estate, agentic AI will lower the barriers to launching new ventures by automating complex operational tasks. “I see this as being a watershed moment for startups as profound as the move to the public cloud,” Silver said.

Silver’s current work focuses on Foundry, which serves as a unified AI portal for enterprises within Microsoft’s Azure cloud platform. The system is designed to help companies deploy autonomous, multistep agents that can execute sequences of actions. According to Silver, these agentic systems will kind of reduce the overall cost of software operations by handling tasks that traditionally required significant human labor. She notes that multistep agents are already becoming broadly used across various coding tasks. For example, agents can manage live-site operations—the maintenance and management of live software services—by diagnosing and mitigating minor technical incidents, reducing the need to wake up human developers for 24/7 on-call support.

While Silver notes that agentic deployments have not happened quite as fast as expected over the last six months, she emphasizes that the return on investment remains clear. She expects that human-in-the-loop scenarios—workflows requiring human oversight or intervention—will remain very common, particularly for critical operations like legal obligations or production code deployment. As an example of this balance, she points to package return workflows, which can transition to being mostly automated while still retaining human oversight for borderline cases.

Silver highlighted specific efficiency statistics to illustrate how these systems change operational math:

• Codebase maintenance: Multistep agents can reason over codebases to update dependencies, resulting in maybe a 70% or 80% reduction in the time required.
• Workflow automation: In scenarios like package returns, workflows can transition to being 90% automated, leaving only 10% requiring human intervention for borderline cases.

Why it matters

Microsoft is positioning its Foundry system to capture the startup market by lowering operational overhead, signaling a shift toward leaner, agent-driven software development.