Most people use Copilot to draft emails and never go further. Here’s what changes once you push past the chat box — Teams, Word, Excel, and building your first agent in Copilot Studio.
This is post two in the Leveling Up series. The overview covered the habits that apply to any AI tool — prompting, context, iteration, knowing the limits. This one is about how those habits actually show up when the tool is Copilot.
At Kwik Trip, Copilot is our enterprise AI tool — it runs inside our Microsoft 365 tenant, which means governance and data handling I’ve written about in Data Privacy in AI Projects apply automatically. That’s the context everything below sits inside.
Where Most People Stop
Most Copilot users I talk to use it one way: open the chat pane, ask it to draft something, copy the output into an email. That’s real value — it’s just the smallest slice of what’s there.
Copilot isn’t one tool. It’s embedded across Teams, Outlook, Word, PowerPoint, and Excel, and it can be extended into custom agents through Copilot Studio. Each surface does something different, and most people never leave the first one they tried.
Where It Lives, and What Changes
In Teams, Copilot can summarize a meeting you missed, pull action items out of a chat thread, or catch you up on a channel’s activity since you were last in it. The value here is speed — it’s the difference between reading a 45-minute recording transcript and getting the three decisions that actually got made.
In Outlook, it drafts replies and summarizes long threads — genuinely useful when you’re catching up after PTO and a thread has forty replies you don’t have time to read individually.
In Word, it can draft from an outline, rewrite for a different audience, or summarize a long document you didn’t write. I use this for turning rough meeting notes into something that reads like a document instead of a stream of consciousness.
In Excel, it can explain a formula you inherited, generate a formula from a plain-language description, or help you spot a pattern in a dataset you’re staring at. This one surprises people the most — it’s not just a chat box bolted onto a spreadsheet, it can reason about what’s actually in your columns.
This is the one I reach for the most, specifically for the Excel skills I don’t use often enough to keep in memory — a nested formula I need twice a year, or conditional formatting rules I always have to relearn. That used to mean a web search and some trial and error. Now I just tell Copilot what I’m trying to accomplish — “highlight this column red if it’s over budget and green if it’s under” — and it writes the formula or sets up the formatting rule directly. I don’t have to know the syntax. I have to know what I want the spreadsheet to do.
In PowerPoint, it drafts a deck outline from a Word document or prompt, which is a faster starting point than a blank slide — though it still needs a human pass for anything going in front of an audience that matters.
None of these are exotic. They’re the same habits from the overview post — being specific, giving context, iterating — applied to whichever app you happen to be in. The mistake is treating Copilot as a single chat window instead of a capability that follows you into the tool you’re already using.
The Enterprise Constraint Is a Feature, Not Just a Limit
I’ve written before about the data governance side of this — Data Privacy in AI Projects and Your AI Usage Is Being Logged both cover it in more depth. The short version for this post: Copilot for M365 doesn’t train on your organizational data, and your prompts stay inside your tenant.
That constraint is usually framed as a limitation — and in the sense that you can’t just paste anything into it, it is. But it’s also what makes Copilot usable for the kind of PM work that consumer tools are off the table for: project charters with vendor terms, stakeholder communications with names attached, anything with a whiff of confidentiality. If your organization has Copilot, that enterprise wrapper is the reason you can actually use it on real work instead of only sanitized hypotheticals.
Building Something: Copilot Studio
Everything above is still using Copilot the way Microsoft ships it. The next level is building with it.
I built a Copilot agent grounded on our SharePoint lessons-learned and project-charter libraries — I wrote the full story here — and the short version is: it answers questions using your organization’s actual documents, not general knowledge. Hand it a new project charter, and it surfaces the lessons learned from comparable past projects without you having to know which ones to go looking for.
That agent didn’t require development experience. It required a data connection to the right SharePoint libraries (your IT or M365 admin team sets that up) and a clear idea of what question you wanted answered repeatedly. If there’s a piece of institutional knowledge your team asks about often enough that you’ve answered it from memory more than twice, that’s usually a sign it’s worth grounding an agent on it instead.
Memory Beats Repeating Yourself
Copilot has a memory feature, and it’s the piece I’d point to first if someone asked what actually saves time day to day. I use it mainly so I don’t have to re-explain how I want responses formatted every single session — once it knows I want a certain structure for status updates or bullet points over dense prose, I stop having to restate that in every prompt. Multiply that across a week of prompts and it adds up.
I’ve also used Pages a handful of times — useful for a document you’re building up across multiple sessions rather than a one-off answer — but it’s occasional for me, not a habit. Memory is the one doing the real work.
The Habits, Applied
Tying this back to the four habits from the overview post:
Prompting in Copilot means being explicit about which surface you’re using it for — “summarize this Teams thread for someone who missed the meeting” gets a different, better answer than a generic “summarize this.”
Context is where the M365 integration actually pays off — Copilot can already see the document, email, or spreadsheet you have open, so you don’t have to paste it in. Memory extends that further: format preferences and other standing context persist across sessions, so you’re not re-establishing them every time. What you still have to supply is everything Copilot can’t remember or see on its own: the audience, the constraint that matters, the thing that happened in a meeting it wasn’t in.
Iteration looks the same as anywhere else — the first draft of a status report or a deck outline is a starting point, not a deliverable.
Knowing the limits matters more here, not less, because the enterprise wrapper can create false confidence. Data governance handles where your prompts go. It doesn’t make Copilot’s answers correct. A grounded agent like the lessons-learned one is only as good as the documents it’s grounded on — garbage in, garbage out still applies inside a compliant tenant.
What’s Next
Next in this series: Leveling Up Claude — projects, memory, and what changes when you’re working with a much longer context window than Copilot gives you.
Links & References
This Series:
Related Monday Business Posts:
- I Built a Copilot Agent to Pull Lessons Learned
- Data Privacy in AI Projects
- Your AI Usage Is Being Logged
- A PM’s Guide to Choosing an AI Tool
Most people never leave the chat pane. The habits are the same wherever you apply them — Copilot just has more places to apply them than people realize.