What started as a simple question about Copilot licensing quickly turned into another lab
experiment.
I wanted to know whether the Microsoft 365 licensing already available in my lab tenant included any
meaningful Copilot capabilities. Rather than spending hours reading licensing documentation, I decided to
take my usual approach: create something and see what happened.
The journey started in Microsoft 365 Chat and eventually led me into Copilot Studio.
At first, I wasn't even sure whether I had access. Copilot Studio presented options for creating
conversational experiences but also displayed reminders about upgrading licenses and starting trials. It
wasn't immediately obvious where the limitations were.
Rather than guessing, I created an AI Lab Team and started clicking buttons.
Within a few minutes, Copilot Studio had provisioned a Microsoft Teams-backed Power Platform environment and
created a Dataverse instance. That discovery was my first surprise. Even without starting a trial, there was
far more functionality available than I expected.
My next surprise came almost immediately.
When I opened the test panel and began experimenting, the results weren't particularly impressive. No custom
topics existed yet, so there was very little conversational behavior to test.
That was the moment I stopped thinking about AI and started thinking about architecture.
As I explored further, I discovered that the real work wasn't happening at a chatbot level. The
conversational behavior was being built through topics, triggers, variables, conditions, and responses. Once
I realized that, Copilot Studio began feeling surprisingly familiar. The experience resembled Power Automate
workflows and Power Platform development more than a traditional chatbot platform.
The first exercise was simple: create a custom topic, define trigger phrases, and return a message.
After that worked, I added a question that asked users which Microsoft technology interested them most. The
response was stored in a variable and then returned to the user.
That may not sound particularly exciting, but it represented an important shift. The conversation was no
longer limited to static responses. The topic could capture information from the user and use that
information later in the conversation.
Then came branching logic.
Depending on whether the user entered SharePoint, Entra ID, or Azure, the conversation would follow a
different path and return a different response.
Somewhere along the way, I found myself looking for a switch statement.
At first, I created separate conditions before discovering that Copilot Studio supports multiple branches
within a single condition block. The result felt very similar to a traditional switch-case structure,
complete with a default path for anything that didn't match the defined conditions.
That discovery led me to another interesting feature: formula-based conditions.
Although I didn't spend much time with formulas during this session, seeing the underlying expressions
revealed that the visual branching tools were ultimately evaluating Boolean logic behind the scenes. The
low-code interface was beginning to look much more developer-friendly than I originally expected.
The biggest lesson so far has been that Copilot Studio is not simply an AI chatbot platform.
At its core, it feels like a Power Platform solution that happens to include conversational interfaces.
Topics, variables, conditions, and Dataverse felt far more important to understanding the platform than any
AI functionality I had encountered so far.
Even before exploring premium features, knowledge sources, tools, advanced integrations, or AI-powered
responses, I had already learned about environments, Dataverse, topics, variables, conditions, branching
logic, and conversational design.
Eventually I found the first hard boundary.
When I attempted to add knowledge sources, Copilot Studio informed me that additional licensing or a trial
would be required. That answered my original question better than any licensing guide could have.
Without purchasing anything or starting a trial, I was able to provision an environment, explore Dataverse,
build custom topics, capture user input, store variables, create branching conversation logic, and learn the
foundations of the platform.
Not bad for a project that started with a simple question about licensing.
The trial hasn't started yet.
Which usually means the real rabbit hole has barely even started. 🐦🐦