The problem with (Enterprise, 365) is that it's critically underpowered for precisely the kind of knowledge work that the 365 power users need it for. Summarizing emails or Word docs, or creating a ppt from an email, are essentially parlor tricks. The value is in the long context that encodes your specific business practice...except that copilot flakes hard when the window gets long, and can only be fed a handful of knowledge sources.

Copilot also has no semantic knowledge of Microsoft-but-not-365 things like PowerBI dashboards, so your average cubicle jockey, who doesnt have access to the data model, can't, say ask it to pull last month's sales figures. That's a giant hole in capability.

The agent discoverability is non existent, so teams end up repeating work. It doesnt have a skills.md type of model that allows for easy, non-admin, extensibility. And most fatal of all IMHO is that it does not play nicely with other non-Microsoft things out there. This may be as much a fault of Salesforce and Atlassian etc as Microsoft's, but Copilot cant see into and manipulate objects in those products (certainly not without admin-level approval, and even then, not very much).

I feel that Microsoft are losing the game here by not allowing a genuine democratization of what the tools can do. Excel is impregnable, not just because society is built on it, but because even the lowliest office grunt is allowed to use all of it. If Copilot can't be used and wired to everything that a knowledge worker uses without requiring IT- or 365-admin approval, then what will happen is that Shadow AI will show up and workers will start bringing thwir own AI tools into work, with ugly consequences for security and data exfiltration.

My crystal ball says that Microsoft know that CP cant do it all, so they will pitch Cowork, as a sort of workaround, at which point some companies will balk at the price and park CP in a dark corner while the real work gets done in Google Workspace or Slack or whatever other environment gives them more of what they need.

I have to say the only part of copilot in m365 I really like is cowork. We have a preview version of it and it's great to spar with it and give it recurring tasks to do. Finally something that actually actively helps.

But that's billed separately on a token basis and cost really adds up.

The other stuff is just not so useful to me. I use it a bit to search for emails and teams chats but that's also cause the search in outlook and teams is so bad. It's not that I'm against AI; I use perplexity a lot personally, at work it's not allowed though. Copilot is just not performing well.

I can imagine people don't really use it.

Work gave me a subscription so I use Opus in M365 Copilot rather than paying for my own Claude/ChatGPT/etc. It's a bit quirky in that it won't accept a lot of file formats, but appending .txt to the end works. I also use https://github.com/simonw/files-to-prompt to combine everything into one file which also helps.

The 4.5% statistic is not attributed. I doubt its validity. This is a bad article.

It's just an entertainment product, not a frontier coding LLM or real capable AI. They say it's for entertainment too in their docs. What do they expect? It's probably directly competing with netflix, xbox etc.

"Less than 5% of Microsoft’s half a billion commercial Microsoft 365 customers pay for Copilot features"

Seems to me that 4.5% of 500 million isn't necessarily terrible. What should the rate be?

And they'll keep laying off more divisions until it's down to 0.5%!

Thank you for this lovely bit of schadenfreude!

[flagged]