E175 ‘Shaping the Future of AI Agents ‘ with Microsoft’s Jennifer Geraghty Ryan

Today’s guest is Jennifer Geraghty Ryan, Principal Product Manager at Microsoft Ireland. As AI agents emerge as some of the most exciting innovations, Jennifer – who was an AI Person of the Year finalist at the 2024 AI Awards – delves into her journey and the groundbreaking work that her team is doing with AI agents and Copilot Studio.

Jennifer began her journey at Microsoft in 2022, working on supply chain products before transitioning to an ambitious AI project. This initiative led to the creation of an intelligent agent that aids customer success teams in managing customer interactions. The innovation marked a significant transformation from using AI as a simple helper to more agentic roles that are triggered by data changes in downstream systems. Her team integrated these functionalities into Copilot Studio, creating a platform that enables companies to build their own agents.

In May 2024, during the Microsoft Build conference, CEO Satya Nadella announced that the project was entering private preview. Jennifer has since been leading the early access program for Copilot agents, collaborating closely with early adopters to refine and develop their first AI-driven agents. Excitingly, during the recent AI Day in London, it was announced that they would be moving to public preview. This launch highlights successful collaborations with clients like Pets at Home, which developed an agent detecting potential fraud, and McKinsey, which streamlined customer engagement intake.

The podcast also underscores the remarkable scale and impact of Jennifer’s work. With agents built on the Copilot Studio platform, both Microsoft and its customers benefit from extending their Copilots, making Jennifer’s role an instrumental piece of this evolving AI landscape. Her insights provide a glimpse into how AI can transition from a support role to being at the core of business processes.

However, developing AI agents comes with its own set of unique challenges. Jennifer discusses the unexpected ease with which some use cases were handled, such as during a collaboration with Thomson Reuters. Their agent, designed for due diligence, exceeded expectations despite the lack of custom prompt engineering. 

However, other instances required more complex solutions, particularly when dealing with AI’s non-deterministic nature. Testing AI presents distinct obstacles due to its evolving learning and adaptability. Jennifer explains the complexity of applying traditional software testing methods, noting the importance of having human oversight to ensure functionality is performing as intended.

Transparency is a key pillar in Jennifer’s approach. Ensuring users understand which parts of the output derive from AI and which are based on reference materials is crucial. Copilot Studio provides a detailed view of the agent’s activity, allowing creators to see reasoning and decisions in real-time, contributing to a framework for ethical AI development.

The idea that human managers might soon be overseeing AI agents is shaping new career paths. Jennifer anticipates that managing AI agents will become a common responsibility, requiring human oversight for tasks agents cannot perform independently.

Jennifer’s work exemplifies the potential of AI to revolutionise workplaces across the globe. By integrating AI at the core of business applications, companies not only enhance productivity but also redefine how tasks are approached and completed.