[a_bonobo]

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.

[ksec]

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.

[wkat4242]

> 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'.

[detaro]

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.

[spwa4]

What about the secondary effects of AI? I mean, now large-scale data gathering on your company simply by recording glasses + having AI transcribe the whole thing is possible. Hell, with modern cell phones you could do it live.

And the first product that lets a cell phone control keyboard and mouse sending camera to ChatGPT, having ChatGPT do all the work is not far off either.

Presumably both of these would violate your ... policies something awful. I should say will violate all your policies, because we both know this is going to happen.

You have no way of preventing both of these from occurring with IT policy.

[wkat4242]

Well to be fair we are already handing all our data to Microsoft on a silver platter since we moved everything to M365. So it's not like they didn't have it all before.

[Cruncharoo]

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.

[teruakohatu]

> 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.

[wkat4242]

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).

[seec]

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.

[ares623]

Boring doesn’t bring in the billions and billions in stock though

[PeterHolzwarth]

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.

[wkat4242]

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.

[a_bonobo]

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.

[CuriouslyC]

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.

[Incipient]

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.

[CuriouslyC]

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.