Introducing deep reasoning and agent flows in Copilot Studio: two new features to help you build smarter agents

Today, I'm excited to share two powerful new capabilities that the team is adding in Microsoft Copilot Studio: deep reasoning and 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. 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.  

Let’s unpack both features!

Deep reasoning in Copilot Studio: tackling complexity with careful thought – in preview today

Deep reasoning in Copilot Studio enables agents to handle complicated tasks that require detailed analysis, methodical thinking, and nuanced understanding – in other words, it gives agents critical thinking skills.  

It works by taking advantage of advanced reasoning models like OpenAI o1 and connecting them to all your internal enterprise data and context. This close integration allows the agent to manage ambiguous business processes much like you or I would. With this capability in Copilot Studio, anyone in the organization can easily build advanced agents that tackle complex problems.

Advanced reasoning models spend more time carefully thinking before generating outputs, resulting in agents capable of parsing complex datasets, recognizing intricate patterns and making thoughtful decisions. When you combine this with your company knowledge you can start to apply agents towards entirely new processes. For example, businesses can use deep reasoning to forecast demand across global markets or optimize supply chains.

In the demo below, watch how a team builds an agent with deep reasoning that creates multi-part Requests for Proposal (RFPs) to respond to customers faster.

Deep reasoning is available in both conversational and autonomous agents. And this is all happening in the background, so you don’t have to choose between standard AI models or deep reasoning. Built into the Copilot Studio architecture is an orchestrator that dynamically decides when to leverage deep reasoning through a combination of input analysis and instruction parsing!

If you are curious about what this looks like under the hood, it includes:

• Instruction Analysis: Evaluates instructions and input data for complexity indicators or ambiguities.

• Contextual Understanding: Considers broader scope to determine tasks beyond the capabilities of simpler analytical models.

• Decision-making: Selects appropriate tools, invoking deep reasoning for tasks that require advanced thinking.

Additionally, agent makers can always override this behavior and explicitly control which steps invoke deep reasoning by simply including the keyword "Reason" in agent-building instructions.

You can learn how to enable the deep reasoning feature here.

Agent flows: AI powered automation – generally available on March 31

Agents are incredibly powerful because they can self-assemble steps and actions to complete a task. But this variability does not work for every process! We know that it is not one size fits all, and sometimes businesses need to combine the predictability of workflows with the flexibility of agents to fully automate a process. That is why we introduced the new agent flows capability to Copilot Studio – it provides structured, rule-based workflows that incorporate AI actions. Between agents and agent flows, it is possible to automate any task you can imagine.

Agent flows follow predefined and deterministic paths, executing tasks with impressive speed and consistency. 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.

For example, if an organization receives customer feedback, an agent flow can help direct that feedback to the right team, based on static rules and policies. If it is general feedback, the agent flow can hand off to a “feedback tracking agent” that can then summarize and generate action items for the team. If it is urgent or negative feedback, the agent flow could hand off to a “customer service agent” that can answer a customer’s question and take action to resolve it.

Agent flows and agents combine to automate complex and mission critical processes. They can operate independently or integrate as skills within existing agents.

 Just like agents, agent flows can be created with natural language instructions or a graphical user interface, making it possible for everyone in the organization to build with Copilot Studio. The same low code experience and platform that powers Power Automate is used behind the scenes.

Copilot Studio makes it easy to choose the right AI tool for the AI job. Watch how easily flows can be added to an agent in this demo.

Autonomous agents – generally available

Lastly, I am also excited to share that autonomous agents are now generally available in Copilot Studio! These agents proactively respond to business signals, leveraging over 50 pre-built triggers (or custom connectors) to plan tasks and manage scenarios on their own. They excel in business processes like engagement management, device procurement, supplier discovery, fraud prevention, due diligence, and thousands of other tasks for knowledge work. We have seen amazing momentum from so many organizations like Dow and Pets at Home during public preview already and look forward to the next wave of innovation.

Bringing it all together

Collectively, deep reasoning and agent flows boost the potential of the agents ANYONE can build in Copilot Studio. It is easier than ever to automate processes with AI by bringing together your enterprise data, custom APIs, cutting-edge AI models, and powerful workflow tools. And our vision for Copilot Studio is to continue developing the most comprehensive agent platform so that you can move fast and solve any business problem in one place.

I am excited to see how these powerful new features help you build smarter, more intuitive agents. I can't wait to hear your feedback and see what you create!

Try it now at: aka.ms/trycopilotstudio!