Welcome to "The AI Economy," a weekly newsletter by Ken Yeung on how AI is influencing business, work, society, and technology. Subscribe now to stay ahead with expert insights and curated updates—delivered straight to your inbox.
Microsoft is introducing a new way for developers to create smarter Copilots. The company announced it is adding deep reasoning capabilities to Copilot Studio, the platform designed to help enterprise customers build custom bots tailored to their specific workflows. This update, available today, also includes the addition of agent flows.
“These features represent the latest milestones in a wave of AI innovation, marking significant advancements in how you can build and manage agents. These features showcase the breadth of agents you can create when you combine enterprise business data, access to advanced reasoning models, and workflows,” Charles Lamanna, Microsoft’s corporate vice president for business and industry Copilot, wrote on LinkedIn. “With deep reasoning, agents can perform complex tasks and make more accurate decisions. And agent flows bring AI workflows into Copilot Studio that enable agents to follow a predefined sequence of actions, ensuring consistent results for structured tasks.”
Subscribe to The AI Economy
Build Smarter Agents in Copilot Studio
What is deep reasoning? It’s an advanced form of cognitive processing beyond simple pattern recognition or data retrieval. With it, AI can understand complex relationships, draw inferences, solve problems, handle abstract concepts, and apply contextual understanding.
When applied to AI agents, deep reasoning enables bots to undertake complicated tasks that require detailed analysis, methodical thinking, and nuanced understanding. In the case of Microsoft’s announcement, developers using Copilot Studio can apply advanced reasoning models like OpenAI’s o1 to their agent and connect it to their enterprise database.
Helping companies build smarter agents is a clear advantage, but adding reasoning capabilities also reinforces Microsoft’s push to stay ahead in an increasingly competitive landscape.
Microsoft says deep reasoning is available for both conversational and autonomous systems, and developers don’t need to do anything to enable it. Lamanna states that it’s taken care of in the background. Furthermore, what’s included is instruction analysis (analyzing instructions and input data to identify complexity factors or potential ambiguities), contextual understanding (assessing scope to identify tasks too complex for basic models), and decision-making (choosing the right tool and activating deep reasoning for suitable tasks that need advanced thinking).
Developers have full control over their agents, allowing them to disable deep reasoning if desired and specify which steps don’t require advanced processing.
Build Automation into Your Copilot
Along with deep reasoning in Copilot Studio, Microsoft has added agent flows, AI-powered automation. Starting March 31, this feature enables developers to create structured, rule-based workflows incorporating AI actions. Lamanna claims it will automate any task you want.
“Agent flows follow predefined and deterministic paths, executing tasks with impressive speed and consistency,” he writes. “They are designed to handle predictable and repetitive scenarios like document processing, routine financial approvals, and compliance tasks. By blending structured automation with strategic AI actions, they simplify complex decision points such as conditions, exceptions, and loops.”
Agent flows can be easily created using natural language instructions or a graphical user interface (GUI). By being low-code, Microsoft enables anyone to create agents within Copilot Studio easily.
Deep Reasoning Comes For Microsoft 365 Copilot
Microsoft isn’t finished adding deep reasoning to its products. In addition to Copilot Studio, the company is releasing two reasoning agents it bills as “first-of-its-kind” for Microsoft 365 Copilot. With Researcher and Analyst, users can better analyze the copious amounts of information spread across their workspace, from emails to meetings, chats, the web, and more.
With Researcher, this agent is designed to tackle the complex, multi-step research needed at work. It leverages OpenAI’s deep research model and M365 Copilot’s advanced orchestration and deep search capabilities. There are also third-party connectors so users can integrate data from ServiceNow, Salesforce, Confluence, and other platforms into M365 Copilot.
As for Analyst, Microsoft wants you to think of it as a data scientist. It’s built on OpenAI’s o3-mini reasoning model and is built to handle advanced data analysis, so it can quickly extract insights from any raw data you provide. It uses chain-of-thought reasoning to progress through problems, slowly refining its thinking, and promises to deliver an answer that should be close to a human being’s thinking.
The Researcher and Analyst reasoning agents for M365 will be rolling out in April.
Featured Image: A Microsoft sign displayed at the company's Redmond, Washington campus on May 20, 2024. Photo credit: Ken Yeung
Subscribe to “The AI Economy”
Exploring AI’s impact on business, work, society, and technology.
Leave a Reply
You must be logged in to post a comment.