Copilot Work Mode, Semantic Search, and AB-730 Limits
Explore how Microsoft 365 Copilot switches between Web Mode and Work Mode, how the semantic index and Context IQ improve source-aware answers, and why the AB-730 prompt framework matters. The episode also covers key Copilot limits, including execution timeouts, source caps, and the importance of verifying AI outputs.
Is this your podcast and want to remove this banner? Click here.
Chapter 1
Web vs. Work Mode and the Semantic Index
Simon Carver
Welcome to the show everybody! I'm Simon Carver, here with Lachlan Reed. And Lachlan, I want to start with a scenario that I think anyone who's tried to use AI at work has felt. You're sitting at your desk, you ask the AI to find that one quarterly report your boss sent last Thursday, and instead of pulling up the PDF, it starts giving you a generic essay about corporate finance trends from three years ago. It is just incredibly frustrating.
Lachlan Reed
Oh mate, it's like asking your dog to fetch the morning paper and he comes back with a half-chewed stick from the park. He's trying, but it's not what you needed! That right there is the classic mix-up between Web Mode and Work Mode in Microsoft 365 Copilot, which is actually a massive focus for the AB-730 exam.
Simon Carver
Right, and understanding that distinction is half the battle. Web Mode is basically Copilot acting like a supercharged search engine. It's pulling from public data, and it can even bypass that standard September 2025 knowledge cutoff because it's actively browsing the live web. But Work Mode is a completely different beast. It operates within what Microsoft calls Work IQ.
Lachlan Reed
Yeah, Work IQ is the absolute key here. When you flick it over to Work Mode, you're telling Copilot, "Hey, stop looking at the whole internet and only look at my company's sandbox." We're talking your internal emails, your Teams chats, your OneDrive files, and your calendar. It's secure, and it only sees what you actually have permission to access.
Simon Carver
And the technology that makes Work Mode feel so magical is the semantic index. In the old days, if you wanted to find a file, you had to hope you remembered the exact keyword in the title, right? If the file was called "Project Zebra" and you searched "Project Horse," you got absolutely nothing.
Lachlan Reed
Exactly, keyword search is so literal. But the semantic index doesn't just look at the letters you typed; it understands the conceptual meaning. It maps relationships between words. So, tell me about that "sluggish boat" example you mentioned earlier, Simon.
Simon Carver
Yes! Imagine you're trying to find a specific team meeting where everyone was talking about a project that was moving incredibly slowly, but you can't remember the project name. With the semantic index, you could literally type in "sluggish boat movies," and it might surface a Teams transcript where someone joked that a project was moving as slow as the Titanic. It connects "sluggish boat" to "slow Titanic" because it understands the underlying concept, not just the keywords.
Lachlan Reed
That is wild. It's basically reading between the lines of your own company's data. But let's say you don't want Copilot to scan everything. You want to point its nose directly at one specific file. That's where Context IQ comes in, and specifically, the slash command.
Simon Carver
Oh, the slash command is a game-changer. You just type a forward slash in the prompt box, and a little menu pops up. You can instantly select a specific PowerPoint, a PDF, or even a person's name to narrow down the focus.
Lachlan Reed
Right, instead of saying "summarize the email from Sarah," you type "slash," select Sarah, type "slash," select the email, and Copilot knows exactly which digital paper trail to follow. It saves so much time and stops the AI from wandering off.
Chapter 2
Mastering the Prompting Framework and M365 Limits
Simon Carver
It really does. Now, if you're preparing for the AB-730 exam, you absolutely have to memorize the official four-part prompt engineering framework. Microsoft breaks this down into Goal, Context, Source, and Expectations. And here's the kicker: only one of these is actually mandatory to get a response, and that's the Goal. You have to tell it what to do, like "write an email."
Lachlan Reed
But if you only give it the Goal, you're getting a generic response. The real magic happens when you layer in the other three. Context tells it why you're doing it. Source is where you use that slash command we just talked about to lock down the reference files. And Expectations define the format or tone, like "make it a three-bullet summary."
Simon Carver
And if you leave out that "Source" component? Copilot will default to searching your entire accessible tenant. That means it might pull data from a three-year-old draft instead of the final version you wanted. So, specifying the source is critical for accuracy.
Lachlan Reed
Now, once you get that perfect output, what do you do with it? Copilot Chat actually has some neat features here, like Copilot Pages. It's basically a digital canvas where you can save and edit the AI's responses directly alongside your team, so it doesn't just disappear into the chat history. Plus, you can set up recurring prompts—like asking for a daily news brief on a competitor—and use the Prompt Gallery to share successful prompts with your team.
Simon Carver
That Prompt Gallery is such a time saver for onboarding teams. But we also need to talk about the hard limits, because the exam will definitely test you on these. Copilot isn't infinite. For instance, did you know there's a strict 100-second execution limit? If you ask Copilot to process a task that takes longer than 100 seconds, it'll just time out.
Lachlan Reed
Yep, 100 seconds and it cuts the cord. And there are file count caps too. If you're using Copilot in Word, you can only reference up to 20 sources at a time. In PowerPoint, that cap drops significantly—you can only reference up to 5 sources. If you try to feed it 6 files in PowerPoint, it's going to reject them.
Simon Carver
Exactly. And let's not forget the golden rule of generative AI: always verify. Even with the semantic index and Work IQ, Copilot can still hallucinate or misunderstand context. If it drafts a client email with the wrong pricing because it pulled from an outdated spreadsheet, that's on you, not the AI.
Lachlan Reed
Too right, Simon. Copilot is your co-pilot, not the captain. You still gotta fly the plane. Well, that's a wrap on our quick look at AB-730 and Microsoft 365 Copilot workflows. Best of luck to everyone prepping for the exam, and we'll catch you next time!
Simon Carver
Goodbye everyone!
