My team manages complex programs across different products and functions, and we already use AI regularly for meeting summaries, documentation, research and meeting preparation. I am interested in whether some of the productivity we are seeing individually can translate into useful team practices.
Meeting preparation is one area where this has been particularly useful. Copilot can search across SharePoint, Teams, email, Confluence and meeting history and help bring together previous discussions, decisions and open items. It can save a fair amount of time that would otherwise be spent looking through multiple sources.
But we have also seen an interesting limitation.
While preparing for a discussion, Copilot may surface something that was discussed internally but never actually made it to the client. The information itself isn't necessarily wrong. What is missing is the context around what happened after that conversation. Was it just an idea? Was it agreed upon? Was it communicated? Is it still relevant?
That distinction matters in Program Management. Having access to more information doesn't automatically mean all of that information belongs in the next discussion.
Experiences like this are also helping shape a broader experiment with my team. We have been building a prompt library, not because there is a shortage of prompts, but to bring some structure to what we are learning across programs. I want to understand which approaches are useful enough to reuse and where we may be creating unnecessary prompting overhead.
As we work through this, I keep coming back to a few questions:
What part is repeatable?
In our meeting preparation, Copilot has been useful for finding and bringing together information from multiple sources. That is work we don't necessarily need to repeat manually.
What context is still required?
The Program Manager knows the history around the information: what was exploratory, what was decided, what reached the client and what has changed since then.
Where is human judgment important?
Copilot can help build the initial picture. The Program Manager still determines what is relevant, what needs validation and what should shape the conversation.
Where is the remaining friction?
If people repeatedly provide similar instructions, assemble similar context or make similar corrections, there may be an opportunity to make that part easier.
This last question is also changing how I think about the prompt library itself. Sometimes a reusable prompt may be enough. In other cases, the better answer may be a template, better access to context, or eventually something built into the workflow.
We are still working through this. I don't know yet which patterns will translate well across different programs, and I don't expect every AI interaction to become a team practice. For now, I am finding it more useful to start with what is already working, understand why it works, and then decide what is worth making easier to reuse.