I still think there's a potential niche for 365 Copilot in its boringness.
At work (not a tech company), there's an ongoing, slow, 365 rollout. The people who participate in the rollout are not technical in any way, but they all love it because they're not regular ChatGPT/Claude users, either. In a way, the restricted feature-set of Copilot compared to ChatGPT helps them, they're overwhelmed enough by Copilot. IT loves 365 because it's so risk-averse. No big jumps, no surprises, clearly defined data and risk policies.
I think if they drill down into the boring, slow, predictable, they will capture the market of risk-averse non-tech companies, not people.
I've tried building agents using 365 for our internal documents and they're OK for basic stuff (what's in what document where - max 20 documents only!!), but langchain/RAG/whatever are a million times more powerful.
I am not even sure what is 365 Copilot. But Copilot is included as part of our Office / Teams package, and literally every one is using it in my organisation. And it is perhaps one of the biggest organisations in the country as well.
> IT loves 365 because it's so risk-averse. No big jumps, no surprises,
Umm no, it's the opposite. It's super high-risk right now for us. Microsoft is constantly shifting stuff around leading us to have to constantly change our processes, documentation etc. Often with zero heads-up and often defaults to on. Some incidents:
- They suddenly started a "free promotion" with Sharepoint agents. We don't want to offer that to our users but it just popped in one day and the admin setting defaulted to on so people were already using it before we turned it off. This was a big deal for us.
- Constant rebranding of their product names leading to confusion among users and zero-value documentation and process rewrites for us. Also constantly fiddling with the URLs is sooo annoying.
- Constant changes in features leading to impact to our DPIA. For example copilot chat didn't have history at first. So no data was kept (they also promise they don't store any for training). Suddenly they added that one day, so we had to redo our entire DPIA because it now does suddenly store personal info which it didn't before so a whole lot of overhead comes into scope (data lifecycle, privacy regulations, security, data loss prevention etc). This is exhausting and there is no way to delay these features until we have approved them. Also, it caused our DPIA team to be highly critical after this incident. Because of course: If they did this before, what guarantees that they won't change something worse next month?
- Limited granularity of access controls - a lot is very high-scope on/off style controls. Meaning that if we want to block something we often block unintended features as well.
A lot of these things are definitely 'big jumps' and 'surprises'.
Plenty places seem to treat Microsoft et al as a "force of nature". Employee makes a mistake: terrible. Small vendor breaks something, whoever advocated for them is in hot water. Microsoft fucks something up, same category as if HQ gets leveled by an earthquake: sad, but nobody could've prevented it.
We use it in finance for the reasons above. Microsoft actually just added GPT 5 to Copilot so it’s much better than it used to be. I’ve used it to help write scripts in VBA and it’s really good for my use cases. Before GPT 5 it was worse than useless.
> just added GPT 5 to Copilot so it’s much better than it used to be
Only marginally in general office use. Maybe is has better coding capabilities, but there is no way its the same sized model used on ChatGPT or via. OpenAI API.
This is the impression I get too and I hear it a lot from our users too. Some have provided clear examples of something prompted to their personal ChatGPT sub and the company copilot one, and yes, it is often much lower quality. As it is the same product under the hood, there must be some kind of squeezing going on behind the scenes. And understandably because they're just reselling a third party service and they want their own margin on it.
I also wonder why Microsoft pushes it so hard when they don't even have a competitive LLM themselves. They're basically admitting failure (similar to when they moved Edge to a rebranded Chrome by the way).
I think you answer your own question: because it doesn't matter if they can sell it and still make a profit.
Microsoft has a lot of already existing clients and those clients (mostly) trust them. They don't need to do a lot of convincing or marketing/sales to get them to use just one more thing from them even if it is moderately useful. If they can manage to resell whatever the LLM makers have made and still generate a profit, that's all they really need. They can figure out later how to make their own version if that turns out to be profitable enough.
This is the same thing with the browser; there is not much money to be made from building/maintaining an engine in the end. But packaging an already existing one with their own UI and tools (and ads) that may make them money indirectly is a good enough approach.
The thing that impresses me the most about Microsoft is that they rarely come up with the best solution right away but they always manage to stick to the new stuff and work out over time the useful parts that will make money. It's quite ruthless but in the end it's good business and this is why they manage to stay relevant. Good old, "embrace, extend, extinguish" is quite effective indeed.
Boring doesn’t bring in the billions and billions in stock though
Boring is a short term thing - I think it's most useful to think about things like copilot AI for all windows users as a longer term initiative. Right now, I think MS is just letting people get used to a basic free tier. They will observe how Windows users put it to use, see to what degree of involvement people have with it, and evolve the product (and its pricing) from there.
Is it raking in cash right now? No. But that's just now. MS can and I feel is thinking longer term. Everyone is on a "Crash Dammit" Economist front-cover mode right, trying to will a crash into existence. Microsoft can think longer term about it - after all, we see younger generations diving in to make use of related technologies. Older demographics will, and are, slowly finding out what to make of the tech that is free with an OS install, and MS will see what kind of business to make out of that over time.
The problem with this broader topic is that people are restricting the time frame of the conversation to Right Now. MS doesn't need to worry about that so much, in the grand scheme of things.
Part of the problem is that this "free" stuff is obviously not remaining free. I'm very hesitant to invest my time in it because I don't want to run it into my workflow and then be held by the balls when they make it paid.
I'm definitely not hoping for a crash but I do think it's inevitable. The amount investment is insane and I don't see how this can be recouped. It's a textbook bubble. And the collapse of a bubble always has pretty bad effects on the industry which will impact all of us. This is probably the biggest bubble to blow since the dotcom crisis.
Oh that's true. If AI is a bubble, and if it implodes, then MS might be the only AI place left standing if it can get its product into all of the non-tech companies in the world.
0% chance. The MIT 95% study showed that the tools are good, integration and process is the problem. There's a ton of work for people that can come into a company and set up a bespoke AI control plane/orchestration, the tooling needs to evolve a little bit so this can be a sane, repeatable business though.
We're trying/pushing this now as a small business, and it doesn't really work. Each deployment is/would be far too niche and expensive. No one has really gotten on board, so we've mostly just been doing normal data engineering/orchestration work instead.
Think of it like FDE work. It's cost front-loaded but very profitable on the back end, so early on you want to limit it to lighthouse customers. That being said my suggestion is to build some tooling for now to support your existing work that can be expanded gradually as a substrate for AI service, I expect the tools should mature within 6mo-1year, so you can offer trial AI services to new clients to reassess periodically.
I perceive MS as just being in the "get people using it, accustomed to it, used to it, habituated to using it throughout their day" phase for the default free Win10/11 tier.
That little copilot icon is sitting there in the task tray, clickable and usable. I think people will, to whatever degree makes sense for them, get used to making some use of the basic free tier. Monetization / restrictions requiring some degree of limitations to upsell will eventually come for the currently half-billion or so Win11 users.
I get the point that few are currently paying for it, but there isn't all that much pressure to do so right now. That will come, later.
if the generic public start associating all AI with the useless "Copilot" then Microsoft will succeed in killing off the entire AI industry
Just like using bing by default even if it's outright terrible (to the point of randomly suggesting the suicide hotline if you use it as a calculator), I make it do dumb bullshit for more Microsoft Rewards Points. It pays for xbox game pass for myself and several friends any time we've needed it
I think you'd be surprised at the number of average users who don't think of it as useless. As I've mentioned on comments elsewhere in this thread, remember what we here were talking about a couple years ago: LLM's killing websearch.
> LLM's killing websearch.
Websearch is dying since some more than 10 years. It was stabbed in the back by greed (ads, SEO). LLM is just the nail in the coffin.
Remember Cortana? Microsoft was eroding consumer trust in these chat-style "agent" tools before we even had a name for them.
And clippy before that. It's a running joke since the early 2000's with all the memes. And even this hilarious music video: https://www.youtube.com/watch?v=b4taIpALfAo
Clippy has done a complete volte face when it comes to memes, in 2025. This is because of the right-to-repair people.
"Clippy never wanted to harvest you, turn you into the product, and clandestinely copy your stuff into some language model. Clippy only wanted to help."
> I get the point that few are currently paying for it, but there isn't all that much pressure to do so right now. That will come, late
Yes that is inevitable. But it's exactly why I won't use it. It feels like a scam, the typical drug dealer model. Get them hooked on free stuff, and then milk their addiction.
I almost only use local LLMs now. They're not as good, no. But at least they are mine and nobody will raise the price suddenly. Some pro AI models I just use with API, for the rare instance where local just doesn't cut it. It costs me a few bucks per month that way, not 20 or 30.
The problem is that it doesn’t actually do anything.
It’s “installed” at my company but basically refuses to interact with basic company data like files in sharepoint.
Is it the IT department’s fault for not enabling something? I don’t know, but does that matter?
If the normal search system has access to the data why doesn’t copilot? Surely it would be trivial for Microsoft to run copilot in a way that keeps company data private so that they could actually turn it on and make it useful by default.
> It’s “installed” at my company but basically refuses to interact with basic company data like files in sharepoint.
on my work computer - there's a sep. 365Copilot app that is tied into Teams,Sharepoint, outlook, and I believe our engineering wiki. Probably other stuff I'm not aware of.
I'm honestly shocked how often I use it now.
If I get a random Pipeline failure; I'll copy/paste it into the o365 Copilot app - and it points me to an email I didn't notice ~3 months ago about a new policy change, and then points me to discussion thread I wasn't on ~2 weeks ago about how to get in compliance with direct links to EngWiki 'how to fix..' documentation, and an Teams link to join the breaking teams Office Hours.
Just off a single ~1 sentence prompt and a stack trace
It's kind of amazing.
The part where it gave you access to a thread you were not a part of seems scary to me..
In this case your absence from the thread was probably an oversight, but in general there could be a very good reason for it
The only way I can reasonably interpret it is that it was a discussion on a Teams channel they had access to but weren't involved in.
this; it was a public channel I had access to, but never look at (I'm on... too many to keep up with, fairly large company).
Isn't that just indexed search?
No.
Search is part of this, but that doesn’t necessarily work from an error message.
It doesn’t mean you get the relevant parts of the thread either.
It certainly doesn’t mean you get a populated meeting invite for a relevant team.
> It’s “installed” at my company but basically refuses to interact with basic company data like files in sharepoint.
We make heavy utilization of Copilot Studio lite and full. Lite has quick access to SharePoint/Teams data. Full has access to _any_ data that has a Power Platform connector, REST API, MCP server, or Copilot [Graph] connector, all of which you can build or buy yourself. SAP, SQL, Databricks, you name it, Copilot Studio full can consume it.
It sounds like you don't have an M365 Copilot license but are instead using Copilot 365 Chat (the naming is horrid, absolutely).
If we have “Chat” it doesn’t even seem to have what I would describe as “quick” access. It seems to just do nothing.
Not a very compelling product in terms of tempting companies to buy a license.
You can get trial full licenses if you’re interested in it. Chat isn’t meant to be anything more than the consumer experience with some business data.
A company that’s looking to grow doesn’t wait for customers to proactively start a trial. That would involve IT departments and approvals and all that stuff.
In the pattern I described, Microsoft would just enable this feature for everyone and sit back and wait for employees to start begging their employers to expand their license because they got a taste of some great functionality.
That’s exactly what OpenAI does with ChatGPT Pro.
Then you have Copilot Chat, that's the free version. It indeed can't interact with company data unless you manually upload it.
If you have a tab selector "Work / Web" at the top, you do have the paid version. Select Work to make it able to find stuff in sharepoint, look at your calendar and emails etc.
If you don't have this selector you only have the free version.
> If the normal search system has access to the data why doesn’t copilot? Surely it would be trivial for Microsoft to run copilot in a way that keeps company data private so that they could actually turn it on and make it useful by default.
Upselling of course. Finding stuff in the gigantic trash heap that is Sharepoint is its main added value for me.
Also, RAG will cost a lot more compute.
The same M365 Copilot app is installed for both “free users” and the premium licenses that have access to all of their Graph tenant data. From the sounds of it, you only have the basic Copilot Chat (web grounded) that comes with the E3/E5 license. If you have the full license for M365 Copilot, you’ll see a work/web toggle at the top of your chat screen.
I assume my company isn’t paying specifically for copilot and what you’re saying is exactly true.
I’m just not sure why that makes any bit of sense as a product growth strategy.
Instead of trialing an amazing experience they’re just showing their customers that copilot sucks. But if you buy a license it’s awesome, we promise!
Why not actually give me something useful and then cut me off from access when I hit a usage cap? Then uncapping my usage is the upsell. Literally copy what OpenAI does.
"The problem is that it doesn’t actually do anything."
I disagree, and I think you'll see what I mean when you shift the frame of reference from "person working at a large company with Office 365 installed" to a couple hundred million average Joe's having access to it at home.
Webs search is now a horrifying wasteland, and people know it. Remember the conversations we all had just a few years ago: LLM will replace web search. That's the key point here - not "replace web search" for the subset of people who have office jobs, but web search for the vastly greater number of people who just have it at home on their computer.
The tech - the products - are good enough for your average person at home who wants a starting point and a structure to work through for something they know nothing about. I think that's actually one of the strengths of the tech as it exists for the winder audience: you don't really need ultra accurate, super precise, info and checklists and guides when you just want to know what to look into to do some decorative tiling on the top of an old table you bought; a way to make sense of and work through a type of pop media you have become interested in; to give you a starting point to work through some new problem you have encountered in day to day life.
That "80 percent vaguely accurate-ish" threshold that LLMs can broadly deliver for a novice is actually good enough for that vast majority of things people deal with that aren't really super-critical. Are you idly curious about some ways to think about how to replant and re-do your back yard greenery? Curious about how to make sense of all the competing numbers and criteria and features when looking to buy your first air-conditioner? Want to take a vague, repetitive, not very well put together response to something your neighbor starting holding forth on on Nextdoor and make it tighter and better expressed?
That little Copilot icon that comes default in Windows legitimately can help you there.
I was listening to an NPR story yesterday about how fashion designers are now using “AI” to spot fashion trends. Aside from the fact that it is now unclear whether a journalist means “software” when they say “AI”, a recurring thought I had was that maybe lots of people are now using AI to make inconsequential decisions. One of the predictions they hyped in the story was that “yellow will be hot”. Ok, sure… but if it’s not, it doesn’t really matter. I can see that AI “helps” in this scenario to reduce decision fatigue, but you might have been just as well off flipping a coin. Even if you buy yourself an expensive coin, it’s a lot cheaper!
> LLM will replace web search.
This kills the crab. I mean, kills the Internet. And it's not clear what happens to the existing search advertising business in this scenario.
Google already killed the internet as we knew it. There's a reason engines like yagi have their own spider for "the small web" because google basically ignores everything but the big commercial clickbaity crap.
I’m just conveying my experience. Whatever copilot is at my company’s Microsoft portal seems to do absolutely nothing and connect to nothing.
Even if this is my company not paying for the license, it seems like a pretty miserable way for Microsoft to try to tempt companies into buying one by delivering a completely useless “light” experience.
Everything you’re describing that’s wonderful about the Windows copilot button is the stuff I’m already doing on ChatGPT.com because that brand name came first.
I agree about this. It's very difficult to figure out. In fact my working theory is IT has banned it as much as they can (we get strange and difficult to interpret pronouncements about permission to use LLM).
Windows 11 in general pushes so much unwanted crap at us in ways that we can't control that it's reasonable to assume IT can't make the icons go away if they wanted to. And investing time building a workflow that disappears when IT has finally figured out how to nuke it (if that's the intent) isn't worth the risk.
For reference I work at a hospital so generally IT is extremely sensitive about potential breaches and leaks. In general the policy is we are allowed to use LLM. The organization is on Azure but I can't even find anyone in IT to tell me if we are even allowed to use Power platform (which is also in this weird state of letting you build things but they don't actually work). CoPilot is there ... ish. It's just not very powerful at all.
I totally understand their point of view considering the sensitivity of your data.
In power platform it's very easy to make an oopsie and expose huge volumes of data on some unprotected object storage or an external platform. We have a big review process on such developments, we certainly don't want users to cowboy that on their own.
It's basically the old "Thumbdrive left in a taxi" these days. With the huge difference that these things now lead to huge fines under GDPR etc.
Right. But it's there and we can start building things is my point
> That little copilot icon is sitting there in the task tray, clickable and usable.
I'm using Win 11 and I didn't even notice Copilot on my task bar until you mentioned it...
I think that is a wakeup call for me to finally flash my pc to linux
What's worse then copilot on task bar is the copilot key on keyboard. This key doesn't even have its own scan code, instead it send something like Win+Shift+F23.
Sounds not to bad for me, or have you bound anything to Super-Shift-F23?
I get what you - a HN user - mean. But for the broader planet of Windows users, it just becomes this thing you click to get some general info from on whatever general topic, question, or need comes to mind. I think it is rapidly becoming quit banal - a thing people think of as just being another part of using their computing device. No fanfare; no fireworks; no hand-wringing over rights or privacy or training data: just a thing you click on to ask questions to as part of their normal day.
Man, you have really bought into the LLM kool-aid. Your latest response feels as if it was spit-balled by ChatGPT.
You misunderstand me: I don't advocate - I'm just expressing what I think is likely for the billion people out there who now have access to these tools. This Hackernews community is a rather rarefied atmosphere. When we step out into the broader public, there is a whole different world going on. Whether we like it or not, that world is what we have to deal with.
Was this generated with an LLM?
If that's the case, it's a very complacent strategy compared to ChatGPT, which people yearn for and automatically turn to.
It also disregards a big chunk of people not using Windows, where the little Copilot icon is not built in.
ChatGPT also don't make a common mistake that Microsoft does, they don't squeeze on the compute too much. Microsoft does which makes it worse than ChatGPT even though it's literally the same engine. That keeps the GPT brand valuable.
That little copilot icon will sit there unused and unloved by people who will, much like with online ads, screen it out whilst they do what they actually want to do.
For some, but what about the others? The reddit mentality of "everyone hates LLM AI tools" is just a very loud segment. What about everyone else? What about a random Joe who asks it questions to find some path through figuring out why their computer is acting all wonky? What about the people who don't know the first thing about dealing with their water heater suddenly going south, and ask it some basic questions of how to approach that problem? That is, treating it as an OS-native improvement over a search engine.
The list goes on and on, and we have to take time into consideration: people may have default strong feelings against the tech in question now, but often that's just a default stance. Over time, people will dip their toes in, and make use of it to whatever degree makes sense for them. Don't mistake current vocal criticism for the end all, be all, stance that will last forever: people get used to stuff, use it a bit, or more than a bit, time passes, and the tech slowly melds itself into people's lives in some manner and in some form: MS will wait and observer, and evolve the product to suit that slow change over time.
> What about the people who don't know the first thing about dealing with their water heater suddenly going south, and ask it some basic questions of how to approach that problem?
Why would you trust an LLM's advice about an appliance that can hideously disfigure you when they regularly fuck up basic advice about computers?
Unfortunately the majority of people don't know what they don't know. I've seen plenty of cases where an LLM is grossly wrong, but those don't seem to be publicised as much as "LLM does $awesome_thing" in the mainstream media.
Because a new icon in the system tray means nothing to people. People screen out most of what is in the system tray.
If they want to find something via AI, they’ll use their web browser and go to ChatGPT or Gemini.
People ignore the system tray so much, because it is so full of garbage, that Microsoft had to provide a "Hide unused icons" option, and eventually made a control panel to manage which icons you see.
I mean I remember when browser toolbar spyware was rampant, people literally ignored half the screen when they managed to install 30 toolbars.
Ads make money because people click on them.
A very tiny number of people is all you need.
> That little copilot icon is sitting there in the task tray, clickable and usable.
A thing that gets in the way is not usable, is annoying.
Yes. We work a lot with Microsoft and they are very hardball on "adoption". They are always pushing us to send weekly "did you know..."? emails, yammer campaigns, seminars blablah. Very American also, this does not work in Europe where people are much less 'team players' and much more critical of advertising.
I personally thinks this just pisses people off. It certainly does me. I don't want to be told what to do, especially by some vendor. If I don't want to use it I don't want to be herded into using it. If they try anyway I will just push back and start hating on it just out of annoyance.
However this has always been Microsoft's adoption model. Very pushy. Unlike for example Apple's where they make things look, feel and work really good so people actually want to use it.
Hmm, my first thought was, 1.81% of 440M enterprise customers after two years sure sounds bad, but I had no reference frame to compare to. That's 8 million users, which is negligible compared to ChatGPT numbers, but it is also an enterprise tool, not a consumer product. So I thought it may help to recontextualize this number to the realm of adoption of enterprise tools, maybe even specifically Microsoft enterprise tools.
So I thought, maybe the Copilot GTM is probably following the same strategy as another enterprise tool that was pushed onto its existing customer base: Teams. So I went looking for data on the rate of Teams adoption.
Interestingly, the sources I found suggest that Teams also had 8 million users in 2019, two years after release!
https://www.demandsage.com/microsoft-teams-statistics/
https://www.businessofapps.com/data/microsoft-teams-statisti...
Note, this was before Teams got a massive boost in early 2020 during the Covid push to remote. And for all the hate MS Teams gets, we see where it is today.
So maybe this is just following the normal curve of adoption of Microsoft products. Maybe it's even doing better than Teams, considering Copilot is not free, as Teams was at launch.
Unsurprising. I tried it probably a year ago. I asked it what meetings were in my calendar for the day and it couldn’t even tell me that. To add insult to injury, they wanted an annual commitment with up front payment at the time.
Many other recent Microsoft products are below average. I'm thinking Teams, Outlook, Windows 11. Azure and Office are better.
I guess they spend their money on sales and licensing and not on developing good products now?
Sales and marketing is the only thing they've ever been good at. Really.
They make their money because they can go to the largest corporations & governments, talk to the CTO and tell them they have a product for everything that checks all their regulatory requirements, and all these products kind of sort of work together.
Who else can offer that?
For Microsoft, engineering is just about checking boxes on feature lists. Quality doesn't matter and engineering is a cost to be minimized. The people who make the purchasing decisions aren't the ones who have to use their stuff.
Yes, that's the thing. I see the same. Every time we had to move from a third party product to a MS one we always had to limit what we wanted to do because the MS solution was far less capable and if we asked about it we'd just get stupid answers like "why do you want to do that, we don't work that way".
But they push it through the CTO with really good initial pricing, then when you're hooked they screw you on the next contract negotiation.
The old saying is: "Microsoft is much better at talking to your boss than you are". Which still rings true IMO.
Even office isn't that good. It's basically been static in features for 25 years (until copilot came around). It took them a decade to surpass the 65535-row problem in excel. Microsoft get really lazy when they have a virtual monopoly. We saw the same with Internet Explorer. They just left it to whither away until it was too late.
And yeah Teams, I can't stand it. Outlook too, in particular the "new" one that doesn't do half the things the old one could do and requires all your email to be in MS cloud even if you have another provider.
> Microsoft get really lazy when they have a virtual monopoly.
That's not very specific to Microsoft really. You can say the same thing about Google, Apple or any other big corp that managed to capture a market where users don't really have real alternatives.
I don't agree, for example Google hasn't done to Chrome what Microsoft did to IE, just basically dropping all development. And Chrome probably has a bigger marketshare than IE ever did.
well also it's all tied together. no shuffling contracts for slack, email, etc.
I mean, it worked for Oracle right?
The crazy thing is how the premium M365 Copilot has only gotten worse as the free Copilot Chat has been rolled out. When the AI assistant from Microsoft can't create a calendar entry in Outlook, what are they even doing with this tech at MS?
You either get A) hallucinations of "I created the calendar entry now" (it didn't) or B) an .ics file you need to download, then import to Outlook manually.
More tests about this scenario in here: https://www.perspectives.plus/p/assistants-without-hands
The Meeting Management MCP server (aka Outlook calendar) can do that easily with pretty much no configuration required on the end user side.
A year is a long time in the tech industry, let alone LLM (or Copilot) history.
The difference is that Copilot will have access to your calendar _and_ associated resources (messages, meeting transcripts, etc.) and be able to relate them to files you have access to. It’s become quite useful in telling me which of the action points for a specific meeting are being handled by whom _now_ and where the resulting documents are even if the meeting was weeks ago.
8 million active users after almost 2 years from making the $30pupm license available. Less than 2% of Microsoft 365 paying customers choosing to pay extra for Copilot. It's not difficult to see why Microsoft themselves have stopped reporting on AI revenue, as well as not disclosing any official numbers on M365 Copilot sales. Luckily, a source leaked these figures for Ed Zitron to report in his newsletter.
I don't know, does it matter? What I mean is that the product only just, in the grand scheme of things, got rolled out. MS isn't under some instant pressure to make loads of cash off this new feature. Year by year, people will just become used to having a tool, part of their OS, that they can ask questions to to get some general information on whatever topic.
It's easy to pick an arbitrary date and point out the lack of profit. But why don't we pick a date a year from now? Or three years. Or five years.
I think over time users will come to assume that their computing device has, as part of the many tools on it, a tool that they can ask about stuff and get general answers.
Is it profitable right exactly now? No. But I suspect it will become completely commonplace in a few years, and not even worthy of note or comment for the average user.
MS isnt some cash-strapped startup that needs to post instant profits right this moment.
> MS isnt some cash-strapped startup that needs to post instant profits right this moment.
Hmm no, but they are clearly under pressure. Integrating Copilot Chat in office was a pretty big move, and the constant rebranding of everything also feels like they have zero vision and are losing control and scrambling to fix it.
I wonder if Copilot Chat in office will really convince users though. As it doesn't have access to Graph it will lose some of the most valuable 'intelligence'. And thus users will try it, get convinced it's pretty useless and be more hesitant to pay for a license in the future.
Part of the reason is Microsoft's borgy/corpo-confusing service levels.
I have a paid 365 account and couldn't determine from logging in or account info screens if I was on paid or just the freemium version with my 365 plan
In testing out what I did have access to with Copilot, it was incredibly bad compared to ChatGPT or Claude, so I decided not to pay for Copilot whenever I see an ad for it.
Funny how renaming standard Office apps to "Microsoft 365 Copilot" in mobile apps, web home page for office.com etc. did NOT make people realize they should buy an additional $30 plan called "Microsoft 365 Copilot" to actually get to use all features of Microsoft 365 Copilot.
If only MS could have asked Copilot whether their naming strategy makes sense.
And then they have differerent versions of copilot. The consumer one at copilot.microsoft.com, the office one at really hard to remember m365.cloud.microsoft/chat etc. It's really annoying.
And it used to be better. Copilot.microsoft.com would just redirect to the best version you had access to. But they dropped that for whatever reason.
The product naming is completely insane. In the span of a year my corporate portal changed name from portal dot office dot com, to Microsoft 365 dot com, and now its copilot something or other dot com.
The portal completely changed in design and the former portal functionality is hidden in a tiny search icon in a knock off chatgpt interface.
The average non-technical corporate user must be so confused.
The confusing nature of product naming is I think pretty typical of MS. It’s just like with personal accounts, where there is no Outlook or Hotmail domain for logging in, it’s still live dot com, and Windows Live hasn’t been a thing for at least 13 years.
Microsoft did this same thing with Teams and OneDrive back in the day. It took quite a while before they straightened that mess out.
Not to mention ton when they renamed Lync to Skype…
Sharepoint has been a complete mess ever since its inception. It lacked any data organisation features so unless you had really strict processes and oversight it turned into a shitheap within months. This is still really the case.
Even now it's mostly useful as a teams backend. But even that doesn't solve the shitheap problem, as any team you create in teams, or every yammer group automatically creates sharepoint sites which are not very useful as a sharepoint site and clutter up search results.
Users often don't understand that when they share something in teams there is an actual sharepoint behind it and sometimes people fiddle with permissions there and then other people don't understand why they're not allowed to do X or Y because the controls aren't fully exposed in the Teams interface. It's such a mess.
And .NET Core to .NET, not to be confused with .NET Framework
.NET was always a clusterfuck of naming, as in the early 2000s Microsoft was slapping the .NET label everywhere. .NET Passport (now known as Microsoft accounts), .NET My Services, Visual Studio .NET (which was the same as regular VS, just with support for building C# and VB.NET apps added), .NET Server 2003 (Windows Server 2003), et weary cetera.
In this respect, as in many others related to .NET, Microsoft was inspired by Java. In the late 90s, Java wasn't just a language, it was a VM, runtime environment, enterprise platform (Java EE, now Jakarta EE), smartcard technology, remoting protocol, operating system, desktop environment, floor wax, dessert topping...
So Microsoft did with .NET what they now do with Copilot, right? GitHub Copilot (AI-assisted code completion and chat), Microsoft 365 Copilot (the suite), Microsoft Copilot (the chat thingy), Windows Copilot (the chat thingy, but directly on Windows), Copilot for Azure (LLM doing RAG), Copilot for Dynamics 365 (I assume something similar)...
I’m pretty sure those figures aren’t global, since very few people at Microsoft have access to that kind of data outside their area of direct accountability.
I can't believe that asking copilot about a Windows issue gives me worse answers than asking chatGPT.
I wonder if the system prompt was written by a committee
1.8% is terrible adoption, especially in light of MSFT hijacking the right-CTRL key of every Windows licensed PC to rename it "CoPilot" key. I'm thrilled that Copilot is sucking wind because their over-aggressive pushiness about CoPilot in Windows and Office has been annoying and wasted my time having to figure out how to disable it but the REAL reason I despise CoPilot is that while stealing a key on my laptop's keyboard, MSFT literally broke the key permanently.
They changed how that key works at a low level so it cannot be cleanly remapped back to right-CTRL. This is because, unlike the CTRL, ALT, Shift and Windows keys, the now-CoPilot key no longer behaves like a modifier key. Now when you press the CoPilot key down it generates both key down and key up events - even when you keep it pressed down. You can work around this somewhat with clever key remapping in tools like AutoHotKey but it is literally impossible to fully restore that key back so it will behave like a true modifier key such as right-CTRL in all contexts. There are a limited number of true modifier keys built into a laptop. Stealing one of them to upsell a monetized service is shitty but intentionally preventing anyone from being able to restore it goes beyond shitty to just maliciously evil.
I agree. Making deals with laptop manufacturers to put their trademark on keyboards is an unhealthy-for-society abuse of their market power and should be treated as such in every jurisdiction that they trade in.
However, I've had no issue remapping it to the Context Menu key on the Chinese Lenovo Xiao Xin laptop I bought recently. It shows up in the PowerToys keyboard remapper tool as an F23 key. (Yes, there are more than 12 F key codes! I believe there is 24 in Windows.)
Other OSes will not even be able to say things like "Press your Copilot key to open FEDORA SEARCH" or whatever without being cease'n'desisted to high heaven.
Terrible.
> However, I've had no issue remapping it to the Context Menu key on the Chinese Lenovo Xiao Xin laptop I bought recently.
That's probably because you're not using it as a modifier so the OP's complaint of the automatic key-up event doesn't impact you (it's not necessary to have a keep-pressed state in this type of use).
Correct. Further down thread I wrote to another poster a summary of what's really happening now when you remap the CoPilot key - copied here...
The CoPilot key is really sending: Shift+Alt+Win+Ctrl+F23 which Windows now uses as the shortcut to run the CoPilot application. When you remap the CoPilot key to right-Ctrl only the F23 is being remapped to right-Ctrl. Due to the way Windows works and because MSFT is now sending F23 DOWN and then F23 UP when the CoPilot key has only been pressed Down but not yet released, those other modifiers remain pressed down when our remapped key is sent. I don't know if this was intentional on MSFT's part to break full remapping or if it's a bug. Either way, it's certainly non-standard and completely unnecessary. It would still work for calling the CoPilot app to wait for the CoPilot key to be released to send the F23 KEY UP event. That's the standard method and would allow full remapping of the key.
But instead, when you press CoPilot after remapping it to Right-Ctrl... the keys actually being sent are: Shift+Alt+Win+Right-Ctrl (there are also some other keypresses in there that are masked). If your use case doesn't care that Shift, Alt and Win are also pressed with Right-Ctrl then it'll seem fine - but it isn't. Your CoPilot key remapped to Right-Ctrl no longer works like it did before or like Left-Ctrl still works (sending no other modifiers). Unfortunately, a lot of shortcuts (including several common Windows desktop shortcuts) involve Ctrl in combination with other modifiers. Those shortcuts still work with Left-Ctrl but not CoPilot remapped to Right-Ctrl. And there's no way to fix it with remapping (whether AutoHotKey, PowerToys, Registry Key, etc). It might be possible to fix it with a service running below the level of Windows with full admin control which intercepts the generated keys before Windows ever sees them - but as far as I know, no one has succeeded in creating that.
> Shift+Alt+Win+Ctrl
What??? Who has introduced that, is there a reason, why they are using all the modifiers?
> is there a reason
Using more modifiers reduces the odds of conflicting with any other existing shortcut/hotkey - and since the intent is that the combination is only ever generated automatically with the CoPilot key - there's no issue with users having to remember or perform it. The problem isn't with them using all those modifiers, it's that they are doing it in a non-standard way that breaks remapping that key. They could have easily used all those modifiers in a way that doesn't break full remapping.
I did not know that - what an unpleasant choice.
I am not sure how terrible that number is. They probably have many users with only very limited activity or users that simply do not need AI for the workflows they are doing. Intuitively, I would expect the number of users that actually have some use for AI to be in single digits.
Remapping right control works fine for me on win11
It doesn't work like it did before and will fail in some contexts. For example, if you try to use it with other modifier keys (as I do), you cannot because they are ALL already pressed and those presses cannot be masked. A real modifier key can be combined with any or all of the other modifiers. There is now no way to use that key in combination with the other modifiers.
It seems to work fine in vscode setting custom keybinds at least.
I havent tried it anywhere else to be fair
Yes, remapping can work in narrower cases where right-CTRL is only used in combination with non-modifier keys.
CoPilot is really sending: Shift+Alt+Win+Ctrl+F23. When you remap the CoPilot key to right-Ctrl only the F23 is being remapped to right-Ctrl. Due to the way Windows works and because MSFT is now sending F23 DOWN and then F23 UP when the CoPilot key has only been pressed Down, the other modifiers remain pressed down when our remapped key is sent.
So, what's happening now when you press CoPilot after remapping it to Right-Ctrl... the keys actually being sent are: Shift+Alt+Win+Right-Ctrl (there are also some other keypresses in there that are masked). If your use case doesn't care that Shift, Alt and Win are also pressed with Right-Ctrl - then it'll seem fine - but it isn't. Your CoPilot key remapped to Right-Ctrl no longer works like it did before or like Left-Ctrl still works (no other modifiers). Unfortunately, a lot of shortcuts (including several common Windows desktop shortcuts) involve Ctrl in combination with other modifiers. Those shortcuts will still work with Left-Ctrl but not CoPilot remapped to Right-Ctrl. And there's no way to fix it with remapping (whether AutoHotKey, PowerToys, Registry Key, etc). It might be possible to fix it with a low-level service running below the level of Windows with full admin control which intercepts the generated keys before Windows ever sees them - but as far as I know, no one has succeeded in creating that.
That's so strange. Microsoft really makes me scratch my head sometimes.
I know power toys has that setting for low level hook but I think it's only for the windows key
> Microsoft really makes me scratch my head sometimes.
Indeed.
> I know power toys has that setting for low level hook but I think it's only for the windows key
As far as I've been able to determine, PowerToys remapping is the same application level mapping that can be done with AutoHotKey, other similar programs and manual registry keys.
Github Copilot is a pain too. Somehow they are capable of building potentially world changing LLM's, but can't build a working subscription billing system? I'm completely locked out of my Copilot access, that I paid for, with no response from support. Whole experience has somewhat soured me on relying on Microsoft for any critical part of my workflow if the response when things go wrong is complete silence.
Put the LLM on answering (or at least triaging) your support tickets faster.
That's strange, my experience with GitHub and GH Copilot have been great, unlike with other Microsoft products that have their billing integrated in the Microsoft account or MS365, or Entra or whatever they call the "company accounts" today.
Plus for 10$ a month (~8.50 EUR last month) I get more-than-enough-for-me access to Gemini, Claude and OpenAI models from my code editor. I wouldn't pay double for Claude Code or Cursor, so I feel like it's a good enough deal.
> Somehow they are capable of building potentially world changing LLM's
Copilot doesn't run on Microsoft's own LLMs, they're all third party. ChatGPT in office copilot, in github you can choose various other ones too.
Microsoft is really just a reseller there.
Personally I think Claude Code (with Opus in particular) is far better. Unfortunately they are very aware of this and thus it is eyewateringly expensive.
I never understood how 365 Copilot's chat llm was so much worse than the free offerings from OpenAI at the time.
Copilot is often wordy, wrong, and seems like it can't do a lot of the same things I can accomplish without logging in on ChatGPT
Instead of fixing critical bugs in Microsoft Dynamics 365 and rolling out actual useful features, Copilot (with limited business context) is being prioritised and poorly shoved into every aspect of the product.
In my limited experience using copilot at work, i think it has been a game changer for non-technical computer users. Many of my coworkers and i use it as the first attempt at research especially now that it is connected to SharePoint within my company and provides references to the documents it is pulling from. Saves hours for things like compiling international regulations or standards documentation. Additionally, it has super charged (maybe too strong of a word) document search within our environment... Sharepoint search is the worst.
The recent Copilot search that was rolled out to replace the traditional M365 search is actually pretty good. When you use that search UI specifically.
If only asking from Copilot in a chat, the results are less spectacular or reliable. That just shows how a chat UI isn't the best solution for everything. Yet taking the M365 Search terms as input for the LLM to translate into a query that doesn't require the user to know the exact keyword to search - yes, that's a clear step forward.
And Graph grounding remains one of the features behind the premium license. So, only ~2% of M365 users will currently see this benefit. Unless the pay-as-you-go pricing option has huge adoption that we don't have any leaked figures on yet.
Depends a lot on the context of a chat, since it will prioritize the app you’re in, etc.
Augmenting/replacing search seems to be one of the truly useful benefits of Copilot. Not sure if that's because it's good or because Teams/Sharepoint search is so bad.
In true Microsoft spirit, they completely tanked the name. This was launched just shortly after GitHub Copilot was the hottest trend. Everyone associates Copilot with GitHub and think it's a developer-product, not an office-product.
Changing the name of the entire Office suite into Copilot didn't make things better. Now you have no idea of knowing what you are really getting. Is Office Copilot? Is Copilot Office? Is Office AI? Is AI Copilot? Is time real or just a concept? Who knows? The strongest product in Microsoft's portfolio with an established name is now gone.
Give it a year and it will be renamed again to Microsoft Teams 365 Copilot Enterprise with Office subsystem for Windows Business. Just to cover all bases.
MSN, .NET, Windows Live, Live... Microsoft can't help themselves when it comes to going all-in on new branding.
Who wants an LLM in their office suite? Of course this is a failure.
This is a strange perspective. Indeed, no one is paying for Copilot, but no one is paying for AI in general. No one is making any profit on AI. This is not a Copilot problem, but a characteristic of the AI market in general. Only the ones who are selling shovels are making money (nVidia).
I had a subscription for a year and it was absolutely worthless. Has anyone found any single worthwhile use case?
Yes I regularly use it for things like “I remember communicating about subject X in 2023, please find all references to that topic since that date”, and it will search and summarize all Teams, Outlook and OneDrive datapoints that I have access to in the organization and neatly summarizing it and linking directly to said datapoints.
I also use it to play around with data and roughly graph it to understand trends without having to roll out a Jupyter notebook.
I find it quite useful.
My partner uses it. Her use cases are often things like "go to all these URLs and collect some information from each one" or "collect all instances of X from these documents"
Use it like chat got ... There is no useful integration with any ms office app. Excel copilot can't edit cells, ms teams copilot cannot look meaningfully through your messages... Never tried it for other office apps...
Yes. Worthwhile in terms of 'worth using it', not 'worth 30$ a month' though.
What I like:
- Meeting catch-up (I don't use meeting minutes much)
- Finding stuff in my messy uncategorised mailbox
- Finding stuff in sharepoint
Things that were a heavy disappointment:
- Everything in excel
- Rewriting text (it's always too noticeable that it's AI, it should be able to learn and match my style)
- Powerpoint presentation generation. It could not modify for a long time and even now it's just not useful
Tbh I only use it a few times per month where it's actually useful. For other things I've tried but it needs so much explaining to get it to do exactly what I want that it's more efficient to just do it myself.
I used M365 Copilot at work, and it's amazing at summarizing discussions in Teams meetings, and terrible at accessing cells in a spreadsheet from which it's called. I'm guessing that will improve with time and a few releases.
Copilot in Excel is getting some significant updates. It can write Python now that Excel has support for that. There’s also a new “COPILOT()” function in Excel that works pretty well.
Thanks for the tip. That function sounds pretty useful. I've been meaning to automate categorising manual survey responses in an excel file. And trying to do it through copilot studio (at my work we don't have access to a simple API option like ChatGPT offers). But if I could put a copilot prompt in every row that would be amazing.
> terrible at accessing cells in a spreadsheet from which it's called
Are there any companies solving LLM-based interactions with arbitrary messy spreadsheets?
Not the nice demo ones with a single consistent table and maybe some charts/PivotTables - the ones with 20 sheets each of which has 3-4 different data tables laid out side by side, which should be linked to each other but likely aren't?
I've long thought that one could build a probabilistic graph representation, annotated with text, of "this is the likely meaning of this area in the sheet, this is how it relates to other columns in other areas, formulas are inconsistent in this way," allow an LLM to ask questions of a tool that can traverse this graph until it determines an optimal plan, and allow it to output and iterate on e.g. xlwings code to execute that plan.
I'm frankly surprised Microsoft didn't create an entire "skunkworks" for this problem - is anyone else doing so?
I am doing exactly this. In stealth, launching soon
It costs something ridiculous like £30 per user per month. Imagine trying to justify that for 10,000 employees. It’s obscenely high for something so mediocre.
I suspect in this case, enterprise adoption of chat-based AI is downstream from consumer adoption. Like many things, trendsetting/taste-making happens on the consumer side and then comes to enterprise. ChatGPT was first to the consumer consciousness, while Claude made its mark as most tasteful... Copilot just didn't register.
I just want to know how I get rid of Copilot. It’s just annoying.
I bought it for 2 months. Was ok. Dumped it and pay for $30 chat gpt and $100 Claude. Covers pretty much nonstop usage between them.
Great, funny image for the piece.
Full license of copilot is the best user experience. Love it.
It's being shoved down our throats at work, which is really annoying. Beyond being able to search for documents, there is little utility beyond what a local <1B param LLM gives me.
Even the "productivity" suggestions in teams are ridiculous - one of them is "roast me by my calendar"...wtf is that? How is that related to productivity at all?
Think for many people the free component is simply enough.
My company’s security and legal are super paranoid and cannot be convinced to let us use it. So already it’s a dead deal
Realists, not paranoid. The threats are very real, and even very probable at this stage.
Since when has Microsoft had a hit product before version 3? It's simply not in their corporate DNA.
Besides, I just recently saw a review of the newest Copilot for Excel by a serious spreadsheet jockey, and he was truly blown away. Basically, it knows everything Excel can do, how to do it, and it now truly understands numbers as numbers and when to use formulas, etc. etc. Basically, it went from smart automation to full on AI.
If Microsoft doesn't fall for short term thinking, the product will eventually succeed.
Could have been leveraging the new Agen Mode in Excel preview released as a Labs feature. Built on Claude and not GPT-5, this is what the MS teams behind apps like Excel, PowerPoint etc. are now doubling down on.
Possibly the Office Agent can become a V2 of what M365 Copilot was originally sold as. It's not at all surprising that it takes a few years to figure out what LLMs can really be used for in Office tools. Whereas asking for businesses to fund all these experiments with a $30pupm license is not the best move to create early adoption and fans...
Microsoft copilot is literally just Cortana. Which they were forced to remove by users.
It’s garbage. My company has it, presumably by default. You ask it a simple question like “find the company holidays document” and it just can’t do it. It’s somehow not hooked up to the basic sharepoint document storage for the company.
Traditional search works fine.
Then you probably don't have the paid version. The free version ("Copilot 365 *chat*") does not have the ability to search in any company information even in M365.
If you see a work and web tab at the top, click work. Then it should be able to. If you don't see this, your company is not paying and it just won't work.
Implementation issue. Ours is hooked up and it’s really good at finding stuff on sharepoint, emails, chat etc
Sharepoint is automatically integrated. So that should always work on a paid M365 copilot.
But only for paid users. Free users can't have access (and also there is zero custom integration on the free version).
Microsoft's moat and cash cow is Active Directory.
If corps ever stop needing that, MS will be in trouble. Until then, few worries. They can play at following fads (not that I think AI is a fad) and not worry about execution.
Active Directory comes free with Windows Server, it was never a "cash cow", that would be the Office suite.
Microsoft is actively killing of Active Directory and they're replacing it with Entra ID, which is "just" OAuth and hence is relatively easily replaced by a competing product.
Windows Server 2025 included many AD changes that were non-cloud related. They're not actively killing it.
https://learn.microsoft.com/en-us/windows-server/get-started...
Most of those are scalability / security improvements that are only relevant to fortune 500 sized megacorps.
These are the kind of release notes you'd expect to see for a dying product on life support, used only by those few remaining customers that are too big and too invested in it to migrate away.
Also, the Server 2025 release is the first one to change anything in Active Directory, the 2022 and 2019 versions had essentially nothing in the release notes for AD.
A few performance tweaks and a long-overdue security catch-up or two in a decade is not a vibrant, living and breathing product. It's a shambling zombie.
That's a weird take but you're entitled to it, of course; normally we see products die with zero improvements for many years. I'm well aware of the rel history of AD, but thanks for the re-education.
Well to be fair, Entra is still backed by AD behind the scenes as the database. But yeah, I’m also not sure a directory server is still required to store domain objects; tree structures are pretty much solved in all kinds of databases and don’t warrant LDAP anymore anyway.
Yet they get big American companies to pay for this.
something, something, clown world
I am just a typical Office user in a small company environment. Copilot buttons started popping in various Office applications. When I asked Copilot in Outlook to summarise mails from one of my contacts it happily searched the Internet and said it found nothing. AAA experience.
The most pathetic aspect of Microsoft’s Copilot rollout is not Copilot’s poor performance & integration with SharePoint, but that Microsoft actually renamed Microsoft Office to Microsoft 365 Copilot on Windows.
You’re a company whose main product has a brand recognition that ranks right up there with Coca-Cola and Apple, and you rename it after a new product that is a pile of hot garbage. Absolutely mind-boggling.
Anyone who's actually used Copilot won't be surprised by this. It's garbage, even the free tiers of other LLMs are better in comparison.
It’s an annual commitment by default. Admins might enable it if it was monthly. However, you can’t risk one license locking you into an M365 contract. Most admins can’t and shouldn’t make that decision. Failure on Microsoft’s part for the annual requirement. Otherwise, most would toggle the license on and give it a try.
Yeah what we did for a while was just have a bunch of licenses and rotate them through the org, getting people's feedback after a month or so.
Of course all the people asking for these were already very interested in AI so they are not a very representative group. We did take that into account.
I agree. Copilot 365 has been really underwhelming to me. It was also released way too early, half the features in excel didn't even work. That will have left a bad impression on the earliest of adopters. Which are usually the biggest advocates.
But also, 30$ per month is a lot. For around 300 users you're already talking 100k per year! It's also many times the price that companies pay for their existing M365 licenses including office apps, hosting etc. So it means a completely rebudgeting of their entire IT operations. That kind of money leads to highly critical evaluations that reach the very top. They don't do much volume discount because it's all compute heavy.
Another thing I've been seeing is that users who get Copilot often don't use it at all. From the users buying it, I've heard that in our org only 5-10% actively use it. Others might tap it once a month or use only one or two specific features like the teams meeting summaries. Only the most serious AI bros use it for everything.
I'm in that group myself too. I use it for asking questions about meetings (like "i was distracted for a bit, what did I miss") and for finding stuff in SharePoint which always ends up being a huge mess.
I don't use the email rewriting because Copilot can't match my style which makes it very obvious that I've used it which automatically leads people to devalue the message ("he didn't even bother to write it"). Something the end of this article also alludes to when it refers to an all hands email from Nadella being made with copilot.
Also the pushiness of Microsoft is exhausting. Since that ignite 2022 it's's constant Copilot this Copilot that. For me that has the opposite effect. I just end up rolling my eyes when they rebrand entire product groups and keep rolling their product names. It makes it feel like they have no idea what they are doing and just scrambling to keep up.
The dumbing down is also a problem. When 'researcher' was released first it would spend half an hour on my queries making really great and lengthy well-researched results. It was really powerful and comprehensive. Now all I get is a couple of minutes. It's already lost its value. I'm sure they do this to tune the compute costs but it makes the value prop even worse.
I don't know where this is going but our company is not buying much of it for now. And tbh I can see why.
Wow, perhaps more people feel like me: avoidance of all Microsoft products, unless forced to use them. I just don't trust Microsoft, I mean at all, and the last thing I wanna do is use copilot because it's Microsoft.
Isn't vscode copilot the biggest market share still?
I doubt the bleed stops there. Hell, I stopped my outlook subscription and am getting an Apple next after they shoehorned that absolute garbage into my email.