Microsoft’s recent Ignite conference served as the platform for a barrage of new products and features with a major focus on AI, agents, and Copilots.
It was a lot to digest but we provided a series of reports as the rollouts were taking place. An online event hosted by Microsoft this week, however, delivered a wide range of additional product details, as well as insights on how and where the new features can benefit customers. The event featured the insights of Dewain Robinson, principal solution architect at Microsoft and guest analyst for this website.
In this report, I’ll lay out a half dozen advances that were covered, as well as insights directly from Robinson on why they’re important.
M365 Copilot Agent Builder, Workflow Agents
The company re-christened the former Copilot Lite and now positions the tech for end users to build declarative agents, with augmented capabilities. Now called M365 Copilot Agent Builder, the platform is able to leverage data from Teams meetings, calendar information and other productivity apps, as well as details from individuals’ work profiles and org charts, which are now available as knowledge in Agent Builder in the process of building agents.
“We wanted to make it where you can do things like use Office skills to be able to generate Word, Excel, and PowerPoint documents,” Robinson said. “This was something you couldn’t do on your own custom agents” previously. This feature is now generally available.
New Workflow Agents enable customers to create autonomous agents that execute worfklows. Robinson used the example of a user building an agent to check for emails about software bugs, then automatically reply to those message as appropriate. “This is really awesome for personal productivity and enablement,” he said. Worfklow Agents capability became available in November through the Frontier program, and it will be rolled out more broadly.
Better Insights from SharePoint
Microsoft also improved the Retrieval Augmented Generation (RAG) capabilities within Copilot Studio when it comes to tapping and analyzing data stored in SharePoint. A user would ask questions about data stored in SharePoint via RAG.
Robinson demonstrated a before-and-after scenario where SharePoint was being queried for medical insurance data. Before the RAG enhancements for SharePoint, there was significant unhelpful information (“it didn’t really answer the question”), but that output has been refined and enhanced. “This is a very much requested item for a lot of customers; I think this is probably going to be one of the most popular new features,” Robinson said.
Agent Evaluations
With the proliferation of agents taking place, business and tech leaders need visibility into issues including the accuracy of the agents’ work, their quality of responses, user adoption, and much more. That’s where the new Custom Evaluations come in.
With Custom Evaluations, Microsoft is enabling users and developers to build and run structured tests against any of the agents they launch and measure those agents’ level of success. They can also import tests and conduct regression tests if needed. This insight can be used to iterate on an agent and its functionality.
Copilot Studio previously had a testing capability but now using those capabilities no longer requires installing a dedicated testing kit, Robinson said, adding, “Evaluations are going to be super key for you to improve” an agent’s performance.
AI Model Flexibility
Significant new options have been added in terms of choosing AI models. In a nutshell, when a new model is delivered by OpenAI or Anthropic, that will be available quickly within Copilot Studio. This new flexibility will result in changes in the Copilot Studio orchestration and response generation functions to support those new models.
But that also means customers can’t bring their own model to bear; instead, Microsoft is curating the options in order to ensure effective orchestration and efficient operations; mini-models or custom models would not perform well in this context.
Developer Functions
A new Code Interpreter feature in Copilot Studio lets developers generate Python code as part of a prompt they’re building within an agent. A developer could use Code Interpreter, for example, to analyze data and display a chart about the data while using Python. “Behind the scenes, it’s allowing that agent to go write small pieces of Python code that it will then execute to be able to answer the question and allow you to derive information,” he said.
Before Code Interpreter, if the data wasn’t available in Copilot Studio, this level of analysis would not have been possible.
Previous Ignite Analysis:
- Ignite 2025: The Next Evolution of Copilot and AI Agents
- Renowned Futurists Map Out the Next Era of Enterprise AI Success
- Microsoft Deepens MCP Support With New Power Apps, Dataverse Servers
- With Agent 365 and Security Tools, Microsoft Equips Customers to Govern AI Agent Estates
- Agent 365: Microsoft’s ‘HR for AI Agents’
AI Agent & Copilot Summit is an AI-first event to define opportunities, impact, and outcomes with Microsoft Copilot and agents. Building on its 2025 success, the 2026 event takes place March 17-19 in San Diego. Get more details.
Tom Smith
Analyst
Cloud Wars, Agent and Copilot
Related Posts
