Using AI with Microsoft 365 - Hype versus Reality

AI is everywhere, and it's definitely in Microsof365. But what can it actually do? What is overhyped?

Let's be real—generative AI is everywhere. By now, you've probably had an AI write your emails, brainstorm your dinner recipes, or at least attempt to summarize a ridiculously long group chat. But while everyone was busy making generative AI do parlor tricks, Microsoft put a suit and tie on it, gave it a briefcase, and called it Copilot.
Since its grand debut, Copilot has basically moved into the Microsoft 365 suite permanently. It's hanging out in Word, crunching numbers in Excel, crashing your Teams meetings, and trying to declutter your Outlook inbox. But is it actually the ultimate productivity hack, or is it just another clippy-esque distraction hiding behind a sleek new interface?
Grab a coffee, kick back, and let's break down exactly what Microsoft CoPilot is bringing to the table in 2026—what it absolutely crushes, where it totally drops the ball, and how it stacks up against the other big brains in the AI world.
If you haven't checked under the hood recently, Microsoft has been relentlessly upgrading Copilot. It's no longer just a chatbot living in a sidebar; it's practically a shadow employee. I'm not going to regurgitate what's on Microsoft's website as their descriptions are way too slick. I'm breaking it down on a real usability basis with layman's terms.

  • It's there. No need install. Whether you like it or not, CoPilot is pretty much everywhere in the Microsoft 365 sphere. It's in Office, Teams, SharePoint, OneDrive, Power Apps, and even in Purview. Unless it's been switched off (it can be!) it is there to give you guidance and help whenever you need it. You're welcome.

  • Not just asking stuff, it can do stuff too: it's not just sending clever prompts. There are several useful use cases which give real value added. Examples, it can interrogate other people's diaries (assuming they are open), you can upload a file and ask it to do things to it, you can turn data into a PowerPoint slides, you can ask it to extract data from multiple documents in document libraries to save you time, and so on. As long as the documents exists, it can access it and do things to it.

  • Advanced brains: It's not one-note. Beyond the basic search, it can do more advanced deep-thinking searches. Need more? Use Analyst and Researcher modes to get even better responses. But your prompts have to be on-point.

  • PMO power: You have a meeting. No one wants to take notes. Don't worry. Turn on meeting recording and it will not only record the meeting video, but it will also note who is there, what was said (and by who), and will also do a summary at the end as well. Say goodbye to your PMO staff!

  • Agents are super useful: Insert a SharePoint agent next to your files and the power of the AI gets even better. Do even more intensive searches and summaries with nifty agents that can be easily setup and crawl on command. A lot less hallucinations since you restrict where it can obtain information.

What are the positives and features of Microsoft Copilot so far?

Look, Microsoft promised us the moon, and while they definitely delivered a really solid satellite, there are some glaring blind spots.

  • It has its limitations. While you can ask it to find a spot in a colleague's diary to have a meeting, it won't actually book the meeting for you. It will give you steps, but won't go that one step further to book the actual meeting. Same goes with Power Automate. While it can advise on flow breakages, it won't actually help you implement your flow. When it actually does (sometimes), it will ignore your settings and add in Premium flow parts that you don't have license for. And don't get me started on when CoPilot just fails and refuses to do anything. Yes, it does happen!

  • Hallucinations are wild: Unless you use the agent and target a specific location to use, the hallucination factor is pretty high. Often times than not you will need to check and double-check the results that return. But this is typical of most AI systems out there, so not the most surprising.

  • You only get as good as you give: Users using CoPilot and similar AI tools should spend a considerable amount of time mastering prompts and understanding the "language" that CoPilot understands in order to return the most accurate information. A lot of users, myself included, resort to lazy language and then we get frustrated because AI didn't read our minds.

  • CoPilot language is clunky: I've no idea who in the world is using CoPilot to write their emails since often times than not, CoPilot will write the most perfect and well-written summaries and use language that no real person would use. Executive Summaries and similar project documentation are fine, but the system tends to sound very mechanical and people using it to write personal text will often do rewrites.

  • Fool me once: The biggest problem with AI like CoPilot is going to be adoption and continued usage. So many times I've seen users hand back their licenses and the reasons they cite is often that it has failed them. The moment the AI fails or hallucinates, you will lose confidence in its abilities, and fall back to manual working. This is perhaps one of the biggest hurdles that companies face when trying to improve adoption.

Where does CoPilot disappoint?

Not really a fair comparison, since you will likely be mandated to only use CoPilot or similar AI systems that are part of commercial software already purchased by your companies. Often, outside AI are outright banned due to the potential for data leakage of sensitive information. But just a general compare with the heavy hitters:

Writing and composition
When comparing with Grok, Gemini and Claude, the result is that CoPilot performs a lot better, but the reason will be due to the proximity of CoPilot to the data in question. it just interrogates better and doesn't require files to be exported or downloaded/uploaded. Composition is on-par with Claude which is the better of the writers.

Coding
CoPilot was initially developed specifically for coding and to support developers to speed up their work. Which is why I was disappointed that I found Gemini and Claude much better when composing SQL, JSON and formulas. They were also better with pin-pointing issues and feeding back useful scripts that I could use. CoPilot gave general indications and code returned often didn't work

Security
From a security point of view and reduction of possible data leakage, it has to be hands-down CoPilot since the tool is native and isn't connected to other systems with the potential to share information. With recent news about AI going rogue, this will be a major concern since a lot of AI activities will involve uploading company data and to be honest, we don't really know where this information ends up or who sees it.

Images and graphics
Gemini and Grok exceeds here in generating original images whereas CoPilot will try (unsuccessfully) to find the closest thing and try to pass it off as authentic. Not blown away here, and expected it as it's a tool that specialises in creative arts.

Comparison with other commercial AI

So, where does that leave us? Microsoft Copilot isn't the whimsical, all-knowing oracle that pop culture told us AI would be. Instead, it's something arguably much more useful: a highly pragmatic, slightly rigid, relentlessly efficient digital assistant that knows your corporate files better than you do.
If you approach Copilot expecting it to write a bestselling novel or render a photorealistic video, you're going to walk away annoyed. But if you use it for what it was built for—skipping the hour-long meeting, instantly summarizing a 40-page PDF, and turning raw data into a presentable slide deck—it feels like a superpower. It's taking the busywork out of business work, and honestly, that's exactly what we need it to do.
Now, if only we could get it to actually attend those Teams meetings for us, we'd be set.

Conclusion