Why Microsoft Copilot Keeps Disappointing your Team with Graham Rosenberg

Nathan Taylor and Graham Rosenberg explore the next evolution of work. They dive deep into the experimental features currently transforming the Microsoft ecosystem. Moving beyond the hype of simple chat interfaces, they discuss the application of AI agents that take action on your behalf, from automating service desk tickets to triaging overloaded email inboxes.

Graham shares his firsthand experience with Copilot Cowork and WorkIQ, explaining how these tools learn your specific brand voice and organizational context. The conversation highlights why breaking legacy habits, like reflexive Googling, is necessary to build the muscle memory required for an AI-driven future. By focusing on governance and specialized sub-agents, businesses can minimize technical errors while maximizing the speed of their operations.

What You’ll Learn:

How to enable the Copilot Frontier program and scope it for early adopters.
The massive productivity gains of using Claude Opus within Microsoft Excel.
Strategies for building small, specialized agents to avoid AI hallucinations.
The difference between personal Copilots and business-level AI agents.
Why quarterly training is more effective than one-off AI workshops.
How Microsoft Agent 365 will help organizations govern their digital personas.

About the Guest:

Graham Rosenberg is the Director of Intelligence and Automation at Sourcepass, leading the divisions for Business Intelligence, AI, and Robotic Process Automation. With a background in RPA starting in 2019, he now focuses on integrating agentic AI into internal workflows and client operations to drive scalable, secure outcomes.

Episode Highlights:

[00:09:07] Defining the Copilot Frontier Program

Graham explains that Frontier provides early access to experimental Microsoft 365 features before they hit the general public. It allows admins to test new capabilities like multi-model choices and advanced agents in a controlled security group.

[00:11:15] Claude Opus in Excel as a Game Changer

Previously, Copilot in Excel often struggled with complex data tasks or large files. Graham describes how enabling the Anthropic model via Frontier transformed Excel into a superpower capable of analyzing half a million rows without crashing the computer.

[00:22:26] Multi-App Productivity with Copilot Cowork

This feature allows AI to move across multiple apps to perform coordinated tasks, such as triaging a cluttered inbox. Nathan and Graham discuss how it can categorize mission-critical emails and even draft responses based on previous meeting transcripts.

[00:27:57] WorkIQ and Organizational Intelligence

Nathan highlights WorkIQ as a favorite feature because it allows Copilot to learn specific brand tones and messaging. This intelligence layer ensures that generated outlines and documents reflect the actual culture and history of the business.

[00:33:00] Specialized Agents vs. Monolithic Systems

Rather than building one giant AI to handle everything, Graham advocates for creating sub-agents with a narrow focus. This strategy minimizes the risk of hallucinations and ensures high-quality output for specific tasks like ticket categorization.

[00:40:11] The Champion Strategy for AI Success

AI technology moves too fast for a single annual training session to be effective. Graham suggests identifying internal champions who test Frontier features and lead recurring monthly or quarterly education sessions for the rest of the team.

Episode Resources:

If you are ready to move your organization beyond basic chatbots and start building a digital workforce of AI agents reach out to our team to discuss your strategy here: https://sourcepassmcoe.com/demystifying-microsoft-contact

Graham Rosenberg on LinkedIn

Nathan Taylor on LinkedIn

Quotes

"I look at it as two lanes. One is like personal AI, like Copilots, your ChatGPTs, your Claudes, like individuals interacting with these chatbots to make themselves more efficient. But then there's the other side of the agentic side and agents that do that same thing but for your business, not for individuals."
"I always tell friends, coworkers, anyone that AI is not the solution. It's a tool in your tool belt, and you should use it as such. It's like an executive assistant that's way smarter than you."
"Keeping the context just as small as possible will get you the best results. You don't realize that you're chatting with a dozen different agents behind the scenes because the framework takes care of all that for you."
"You really have to get in there and do it yourself. Figure out what it's good at, what it's not good at, and how to prompt it correctly. Until you play with it, you don't know what it can and can't do."
"By 2030, if you are a SaaS vendor and you still require individuals to log in to your platform through a web browser, you're behind. You need to be thinking of how can I build my app for agents to interact with my data."