Is Microsoft's Straightforward Agent Story Enough To Create More Fans?

Forbes

09-04-2025

Microsoft provided a three-fer announcement last week with respect to its strategy for AI agents. These capabilities are all available in various forms from competitors already; see, for example, the articles I've written on Salesforce and UiPath. So it wasn't any breakthrough functionality that caught my eye.
What stuck out for me instead was how each part of Microsoft's announcement was presented in a unified and sensible way. From a customer perspective, it was clear enough where a businessperson could say, 'I'd like to hear more.' It also helped that Microsoft explained and demonstrated each part without using endless terminal windows with multicolored text. Instead, Microsoft was able to leverage its visual development capabilities in Copilot Studio AI (as in the image below).
(Note: Microsoft is an advisory client of my firm, Moor Insights & Strategy.)
Microsoft announced new capabilities to support different types of agent applications — agentic workflows, deep reasoning workflows and autonomous agents. These cover a wide swath of agent use cases. While some platforms may specialize in one type of these or another, Microsoft is going broad and building from existing features in its Copilot Studio AI development tool. Let's break these down.
As the name suggests, an agentic workflow provides a path for an agent to follow to get to a desired end result. In this case, the AI value-add is using language to take inputs from a previous step, interpret them, provide an answer and move to the next step per the workflow rules. This is a very common use case, but not everyone has been very good at making workflow development intuitive. Microsoft has done well with these enhancements, including the ability to use natural language to create, modify and test the workflow, which is presented to the author via an easy-to-understand UI.
Deep reasoning models are relatively new and represent a change in how a user interacts with natural language models. These models are more expensive computationally and financially because they take more compute time to 'think' through a particular request and give a more detailed answer. So, there is a tradeoff between costs and complexity from an agent perspective when comparing deep reasoning versus agentic workflows. Microsoft's demo used the very good example of a deep reasoning agent that can develop a response to a complex and unstructured RFP, which would not have been possible using an agentic workflow.
Microsoft has also deployed a solution for autonomous agents, which are agents that respond to some specified business trigger. A trigger is a piece of code that is on the lookout for certain business events. Once initiated, the trigger collects information and creates an event-specific prompt called a payload. An example would be the addition of an entry to a database or a document to a Sharepoint folder. The trigger sees the event, creates the payload and runs. These actions seem like a very natural fit for agentic workflows to achieve a high degree of automation for rote tasks needing no human intervention.
From a developer's point of view, Microsoft has a good and predictable reputation. The company tends to have strong tooling capabilities and excellent integration with its other products. It's reasonable to say that Microsoft follows the market versus pushing the envelope — but it does a reliable job. There are notable exceptions where Microsoft has been out in front of the industry, for example in its partnership with OpenAI or its major push to get Copilot established in 2023. But in the case of agents, what I see is something more in line with its historical reputation of fast following. For example, the workflow tooling that ServiceNow announced last year is similar to what Microsoft is announcing now. And we have already seen deep reasoning agents from Anthropic in the last few months.
That is not a bad thing. When considering agents, you need to consider that a big part of AI monetization will come from the model and where the data for it is stored. This explains Microsoft aggressively laying down an AI foundation with Copilot (and, by the way, investing heavily in data with offerings like Microsoft Fabric). Agents will become important consumers and facilitators of models and data repositories. So, in the case of agents, Microsoft can afford to look to others and follow quickly.
I recently published a piece on how Google aimed squarely at the customer of its Customer Experience Suite with its latest product announcement. I get a similar feeling from this Microsoft announcement, so kudos to Microsoft for delivering a clear and understandable perspective on agents and AI that should be easy for enterprises to grasp. That said, I do have a few comments and suggestions as I conclude.
First, it would be interesting to see if the user could get some automated guidance about whether an agentic workflow approach or a deep reasoning approach would be better. I mean 'better' from the perspectives of both accuracy and economics. The idea would be that if I used natural language to start creating an agent, Copilot could say which type of agent would be a better fit — but also at what cost. For example, 'This workflow is not very detailed, so a reasoning agent works better, but each time you run this agent it may cost up to three times as much.'
Second, it may be hard to create a trigger ecosystem, and that is a genuine concern. Microsoft already has 50 triggers available, and (unsurprisingly) they cater toward the Microsoft product line. So, for instance, there are triggers for Outlook, Sharepoint and Dataverse. However, we are already starting to see momentum for open standards like MCP, which could someday support a similar capability. This may be an area where an open triggering standard could be beneficial.
Finally, the demos for this iteration tended to focus on personal productivity. That might be fine for now, but I think that in the future Microsoft could leverage other parts of its portfolio and demonstrate some very powerful capabilities, especially in B2B. I'd love to see something that would further differentiate Microsoft solutions such as Dynamics 365 or Defender from their respective competitors.