In this episode of Return on Intelligence, Abram and I dive deep into the transformative announcements from Microsoft Build 2025, exploring how AI agents are fundamentally reshaping workplace productivity. Fresh from his experience at Build, Abram shares exclusive insights on Microsoft's revolutionary multi-agent orchestration capabilities and the new Copilot tuning features that are set to change how businesses approach AI automation.

We explored Abram's firsthand experience at Microsoft Build 2025, where he presented on stage about the latest developer tools for AI agents. The heart of our discussion focused on two groundbreaking capabilities: Copilot tuning, which allows businesses to fine-tune models with their specific data and formats, and multi-agent orchestration, which enables agents to work together in sophisticated hierarchies to accomplish complex tasks.

We discussed how Copilot tuning differs from traditional knowledge grounding by actually training models to consistently output in specific business formats, making it ideal for structured processes like legal file notes or financial reports. Meanwhile, multi-agent orchestration represents a paradigm shift from monolithic AI systems to specialized agent teams that can handle increasingly complex workflows.

The episode culminated with a live demo where Abram built a functional Microsoft Build reception desk agent in real-time, showcasing just how accessible and powerful these new tools have become for everyday users.

Key Insights

• Fine-tuning revolution: Copilot tuning creates models that reliably output in specific business formats, moving beyond simple knowledge grounding to ensure consistent, structured results for critical business processes.
• Multi-agent architecture: The shift to multi-agent systems enables AI to handle complex, multi-step tasks through specialized agents working together, potentially accessing hundreds of tools through hierarchical organization.
• Microservices for AI: Multi-agent systems mirror the microservices architecture revolution, allowing different teams to build and iterate on specialized agents that communicate through natural language rather than rigid APIs.
• Democratized agent building: The new agent builder in Microsoft 365 Copilot makes creating powerful AI agents accessible to end users in minutes, not hours or days of development work.
• Human-AI collaboration evolution: The future workplace will feature humans as both managers of agent fleets and co-founders working alongside AI thought partners in collaborative relationships.
• Professional development pathway: Agents can evolve from simple no-code solutions to sophisticated pro-code implementations, allowing successful agents to scale across entire organizations with proper governance.

Timestamps

00:00 - 03:15: Build conference overview and Abram's presentation experience

03:15 - 12:30: Deep dive into Copilot tuning capabilities and business applications

12:30 - 18:45: Multi-agent orchestration and the future of AI workplace integration

18:45 - 22:00: Discussion on AI's impact on jobs and the augmentation vs replacement debate

22:00 - 39:00: Live demo building a Microsoft Build reception desk agent

39:00 - 48:50: Agent sharing, scaling, and the evolution from no-code to pro-code development

Connect with the Hosts

• Daniel Anderson: LinkedIn
• Abram Jackson: LinkedIn

Summary

In this conversation, Daniel Anderson and Abram Jackson discuss the evolving landscape of AI, particularly focusing on Microsoft's Copilot and its comparison with ChatGPT. They delve into the quality of responses generated by these AI tools, the significance of reasoning models, and the introduction of new agents like Researcher and Analyst. The discussion highlights the practical applications of these technologies in enhancing workplace productivity and the future implications of AI in various industries.

Takeaways

The quality of AI responses varies significantly between different models.
Microsoft 365 Copilot is designed for workplace productivity, not creative writing.
Reasoning models are a new frontier in AI development.
Deep research capabilities can drastically reduce the time needed for information gathering.
The Researcher agent can access and analyze all work-related data.
AI tools are meant to assist, not replace human jobs.
Governance and security are critical in deploying AI in organizations.
The orchestrator plays a vital role in modern AI functionality.
AI can help identify gaps in knowledge and improve decision-making.
The future of AI in the workplace is focused on enhancing human capabilities.

Chapters

00:00 Introduction and Context Setting
05:05 Quality of AI Responses: Copilot vs ChatGPT
12:04 Understanding Reasoning Models in AI
25:08 Deep Research and Its Impact on Productivity
26:32 Introduction to Microsoft 365 Copilot Researcher
27:30 Utilizing Researcher in Work Contexts
29:33 Security and Data Management in Researcher
34:03 Exploring the Analyst Agent
39:01 The Future of AI Agents in Business
43:32 Upcoming Features and Enhancements at Build

In this conversation, Daniel Anderson and Abram Jackson explore the rapid evolution of AI interfaces, focusing on the transition from traditional search to conversational AI and the emergence of agent swarms. They discuss the importance of understanding the differences between automations and agents, the significance of prompt engineering, and the challenges of integrating AI into daily workflows. The conversation emphasizes the need for users to adapt to new habits rather than skills, as well as the potential of specialized agents working collaboratively to enhance productivity.

Takeaways

• AI interfaces have evolved from transactional to conversational.
• Agent swarms represent a new frontier in AI collaboration.
• The rapid pace of AI development can create feelings of falling behind.
• Specialized agents can work together to perform complex tasks.
• Understanding the difference between automations and agents is crucial.
• Conversational AI is becoming essential in business applications.
• Prompt engineering is important for developers, not end users.
• Chat interfaces are both powerful and challenging to navigate.
• Integrating AI into daily workflows requires new habits.
• The future of AI is promising and not a passing fad.