This was the leap from using AI to building with it. For a long while we only had basic chat — but Copilot agents arrived (ahead of ChatGPT's, for us) and Copilot had matured into something genuinely useful. So we learned to make our own.
What we covered
- Accessing Copilot Studio — where it lives and how to get in.
- Creating an agent — building one start to finish, with the pitfalls to watch for along the way.
- Deploying it — taking an agent from "works for me" to live where others can use it.
- The why and the where — when an agent is the right tool, and where it makes sense to put one.
The demo: a curriculum agent wired to live data
To make it concrete, I showed an agent I actually built and deployed — a chatbot living on our curriculum landing page that answers questions about our training.
Why it's clever
The agent ingests our Smartsheet curriculum data directly. Because it reads from the source, I never have to update it with knowledge — when the data changes, the agent already knows. Learners and managers can just ask:
- What courses are available for a given team member?
- How long does a particular course take?
- What training exists for a specific task or role?
No more digging through a spreadsheet — and no maintenance burden on me.
That last point is the one that landed: an agent connected to source data is self-maintaining. You build it once, and it stays current on its own.
Key takeaways
- Agents are worth it when a task is repeatable and answerable from known data.
- Connect an agent to source data and it never needs manual knowledge updates.
- Deployment placement matters — put the agent where the question gets asked (like the curriculum landing page).
- Know the pitfalls going in: scope it tightly and test before you deploy.
Delivered live to my team and recorded internally for anyone who missed it.