Copilot for M365 — six months in. Real talk.

I've been running Microsoft 365 Copilot for about six months now, since it moved into broader commercial availability in early 2024. That's enough time to have moved past the novelty phase and form a genuine opinion.

This is not a feature roundup. Microsoft has plenty of those. This is an honest account of what it's actually changed in my work and where I still find it wanting.

What's genuinely changed

Meeting summarisation is the single most consistently useful feature. Teams meeting recap with Copilot is real, and it works. I routinely leave meetings and have a coherent summary, list of action items, and key decisions within minutes. Not always perfect (Copilot occasionally misattributes who said what, and it struggles with technical jargon in specialist contexts), but accurate enough that it's genuinely saved me time on meeting follow-up.

Drafting in Word has changed my writing process. Not in the way the demos suggest. I'm not typing "write a proposal for X" and using the output verbatim. But having Copilot produce a first draft that I then substantially edit is faster than starting from scratch, and the structural scaffold is usually sound even when the specific content needs work.

Outlook summarisation is useful for long email threads. The "summarise thread" feature is one I use multiple times per week. Threading context that would take me ten minutes to reconstruct is surfaced in seconds. This is directly valuable.

Where it still falls short

Document Q&A is unreliable. The "ask Copilot questions about this document" feature, where you can ask questions about a specific SharePoint document or document set, works inconsistently. For well-structured, clearly written documents, the answers are reasonable. For complex, technical, or poorly structured documents, the answers are often incomplete or subtly wrong. I've stopped relying on it for anything where accuracy matters.

PowerPoint generation is a demo feature. The "create a presentation from this document" capability produces presentations that require substantial rework. The structure is often wrong for the intended audience, the speaker notes are generic, and the visual choices are default Copilot templates that rarely fit the branding or tone of the work. It's a starting point at best.

Search quality is not as improved as expected. One of the promises was that Copilot would make it easier to find things across Microsoft 365. In practice, the "find me documents about X" capability is hit and miss. The underlying search quality issues in Microsoft 365 haven't been solved by adding AI on top.

Six months on

Copilot is worth the cost for heavy knowledge workers: people who generate a lot of written output, who have heavy meeting loads, and who work primarily within the Microsoft 365 surface. For this profile, the productivity gains are real and the $30/month justifies itself.

For users who do more structured work, or who primarily work in applications outside Microsoft 365, or who are not heavy document and email producers, the value is lower and the cost may not justify.

Copilot is a genuinely useful productivity tool for the right profile, not the universal AI transformation it was positioned as. Most software turns out to be more specific in its value than the marketing suggests. The value is real; it's just more targeted than the demos imply.