Microsoft is walking back Windows 11's AI overload
windowscentral.com204 points by jsheard a day ago
204 points by jsheard a day ago
> It appears this moment of pushback has resonated with internal teams: According to people familiar with Microsoft’s plans, the company is now reevaluating its AI strategy on Windows 11 and plans changes to streamline or even remove certain AI features where they don’t make sense.
Obviously this is a complete failure of governance. The very first thing they should have considered was whether or not these features made sense in the ways that they were being added. There should not be any necessary work to "rollback" features that do not make sense, because they should have not built them in the first place.
Even if we accept at face value that AI has made generation of code significantly cheaper, that doesn't justify the existence of worthless code. Taste comes from knowing what not to build.
Right now Windows is an unstable mess, filled with things that shouldn't have been built. The question Microsoft should ask themselves is why they built them in the first place, and how they will prevent this from happening again.
> The question Microsoft should ask themselves is why they built them in the first place
It seems like everyone except MS themselves knows why: they got tunnel vision from Azure and AI, and completely forgot about what actually made them successful.
Hell they even burnt down one of the most famous brands in the world, MS Office, for zero reason other than to try and whitewash their Copilot name. The marketing guys who made that decision urgently need to find another line of work, because literally a Labrador licking his balls all day would have resulted in a better outcome.
The PMs are completely asleep at the wheel, when they aren't actively self-sabotaging.
> The PMs are completely asleep at the wheel
or, everyone has career aspirations for which they need to demonstrate impact, relevance and in shipping products. Since the current hype is AI, making and being part of the AI hype means career advancement (at the time).
If they want AI hype they should be building up .NET to be completely versatile for AI, not just ONNX, but the full pipeline. Make your strengths a key indicator that Windows is the place for AI, stop using up 50% of my RAM for no reason, I need it for real work. Till then Linux has been my new permanent home for about 5 years now or so.
Have you lost sight of how much AI is being shoved down .NET tooling?
See AI components for Blazor, Aspire AI dashboards, Aspire CLI with AI, Powershell AI, aspire.dev web site proudly written with AI, .NET Upgrade tool is now AI driven,....?
None of those sound like the tooling I'm talking about. I'm thinking of libraries like ML.NET, training and inference, compared to Python its nowhere near, a lot of .NET projects wind up calling out to Python itself. I don't see why Microsoft couldn't do more in this area, if they're truly betting on AI they're betting on it the wrong way.
What pure C# inference tooling is out there? I know they have a solid ONNX engine, but not everything runs on ONNX.
I say this as both a Python and .NET developer mind you, but if Microsoft actually built up .NET more seriously to power AI infrastructure, I could see it making a big difference for them. Look at how many game engines use C# as opposed to literally any other programming language. C# could have been a #2 language for AI by now.
You have the great Windows ML experience. :)
Guess why Microsoft hired Guido and other Python devs, who gets the whole Python experience on VSCode, or introduced Python as better option to Excel, in detriment of .NET addins.
People forget that nowadays .NET is only yet another language on DevDiv, check the developer blogs for all languages.
That was F# failure as well, trying to cater to data science for its relevance, while other Microsoft departments double down on Python.
I'm never touching Windows again to be fair. They'd have to decouple it from their marketing departments sins. I see way more AI libraries in Rust that are as capable as Python libraries than I see for .NET for example. The diffusers library has a Rust equivalent, is there a true .NET equivalent?
Come to Windows and you will see.
Jokes aside, not really, however people have to accept to be polyglot, there isn't one language to solve all problems.
Regardless of whatever is in Rust, all AI key frameworks are in C++ and Python.
NVidia, Intel, AMD, Khronos aren't going to start publishing tools in .NET, Rust, Zig, or whatever is our liking.
So anything outside those stacks will always be a second class experience in IDE tooling, debugging, and libraries.
Well if its done in a dumb-as-a-fuck hostile style that whole world complaints for years, such effort and PM is utter failure and their CV should be tarnished with this for next 2 decades. And its up to us as a IT community to make it happen.
They harmed massively their own company, and failed at the most core reason why they were hired - add long term value to the company.
Its a bit the equivalent of architect building huge bridge that then falls, no souls harmed. Such person would have issue finding any other work. Lets do the same, name and shame shouldnt be that hard.
I saw a presentation awhile back which included the slide (roughly):
"Give a PM a numerical goal, and they will burn the company down to hit it."
As someone who has worked in big tech and seen decision-making in action, I 100% believe it. This is how incentives are structured.
The mandate/goal went pretty far up the chain, too. Windows got moved from being under Azure to under "CoreAI" in the org structure. Incentive structures usually reflect org structure. In this case the fingers can point pretty far up on why incentives shifted the way that they did.
That's a dramatic but fitting characterization of https://en.wikipedia.org/wiki/Goodhart%27s_law
Their shareholders did not want them to add long term value to the company.
Their shareholders wanted AI.
Thinking hard about how CoPilot fits into the MS ecosystem (Power BI, SharePoint, Dynamics, Office entrenchment, etc), and how their consultant mills work, I’m convinced there’s a meaningful space for unnecessary, unpopular, or suboptimal LLM solutions that can still be wildly profitable for MS.
Like, just because the outcome sucks and the solutions are user-hostile, let’s not assume the decision makers are dummies. I see profit motives as the likely delta between their decisions and our userland expectations.
Let me run the MS LLM department and I could easily explain to the board why we’re about to see a big upsurge Azure, office 365 integrated, and MCP-based solution spending… hint: it’s because the machine god will tell the consultants AND the customer those solutions are what’s SmartGood. We’ll sell ‘em a box that tells ‘em what to buy (lul, subscribe to!), the profitability part kinda writes itself.
You shouldn’t name and shame for following corporate policy. Your suggestion is ridiculous. If the decision has come down from the product leadership you are expected to follow it.
Knowing who the windows product leadership is should be easy. Find them on linked in. But even they may not be responsible if the direction came from the ceo or the cto. We know who those are.
Quit calling for naming and shaming of individuals just trying to make a living.
There's no way MS employees at all levels don't know. It only doesn't know organizationally. It's just the boring old incentive alignment problem.
There needs to be more squeaky wheels than anticipated at all times in IT to justify investments in software thereby your compensations and promotions. One easy way to achieve that is to keep throwing in shiny new things with more moving parts so to keep something on fire to keep spotlights on. Webdevs achieve this by wrapping wrappers, Google by pulling plugs randomly off the wall, and various parts of Microsoft for the past few quarters had done so by introducing new GUI toolkits and adding moar AI to Windows.
> The marketing guys who made that decision urgently need to find another line of work, because literally a Labrador licking his balls all day would have resulted in a better outcome.
Marketing Driven Development is terrible. If the CEO of Microsoft keeps pulling off these terrible moves time and time again, I would suggest he has overstayed his welcome, bring in fresh blood. Windows should be an OS not an ad platform. If Office doesn't want to be replaced and remain profitable maybe its time to trim your marketing department, clearly they are overstaffed if they can affect the entire OS itself.
I refuse to use Windows. I only use Mac and Linux now, unless an employer gives me a Windows device, that's the only exception, but given the choice I'll ask for Mac or Linux any day.
> The PMs are completely asleep at the wheel, when they aren't actively self-sabotaging.
You’ve never worked at MS, have you? PMs aren’t asleep at the wheel. They are doing their job because their performance reviews are tied to adding these top-down goals into the team’s roadmap. That’s the horrible part.
> burnt down one of the most famous brands in the world, MS Office, for zero reason other than to try and whitewash their Copilot name
Mac user and Office subscriber here. The wild thing is this soured me on the Copilot brand so broadly that I’ve recommended folks weighing it strongly avoid committing to it as their AI strategy. (None of them did.)
That infamous agentic OS tweet pretty much sums up the incentives and response to criticism at Redmond.
My work did although they did avoid paying for it (just using the free offerings).
It actually serves a purpose to them we can say every employee "has AI" so we're an "AI first" company at fancy press events.
Meanwhile nothing actually changed because the free features are even worse than the paid ones.
It's basically greenwashing all over again. AI washing :)
> The marketing guys who made that decision urgently need to find another line of work, because literally a Labrador licking his
They already made money.
They know what works to make money by convincing CEO VP PM devs. I do hope they jump to the next company (please meta or apple) and do their duties.
> because literally a Labrador licking his balls all day would have resulted in a better outcome.
Where I come from we prefer monkeys throwing darts.
I don't know about Windows, but it will take a lot more enshittification than that to burn down the Office brand. Excel alone carries it to dominance.
The Office brand is literally gone, they renamed it to "Microsoft Copilot 365 app". Check https://office.com
I'm shocked they didn't stash "defender" in there somehow. I used to joke that one name they'd rebrand the start menu as "defender for application launching" and rebrand the power button as "defender for powering on."
Microsoft's brands are historical markers. There's an era when a new Microsoft product is .NET, and an era when it's Azure, and one where it's 365 etc. If you have a new Doodad, if you say "Microsoft Doodad" the other divisions hate you because that's not their thing. Brand it "Hot Brand Name Unrelated Word" and now you're part of the family even though you have no product purpose and your customers will forever be confused.
"Azure Active Directory" wasn't Active Directory, and who'd have guessed a year ago that "365 Co-pilot" would mean the Office applications in 2026. Yes really.
> Azure Active Directory
At one point in time, before AzureAD got renamed to Entra ID (or is it just Entra now?) they had:
Active Directory Domain Services, Azure Active Directory Domain Services, Azure Active Directory. All three different products.
Copilot is such a dumb brand name. At least to me, it confers that I need to be a pilot and that it requires training to be one.
I just want to be productive, not fly a plane.
Also, the Copilot is “waiting in the wings” to take over your job.
And when you ask it to do something useful the answer usually is “sorry I can’t do that”.
That is insane. Microsoft Office is probably one of the most recognizable brand names ever. Reminds me of the time when they called everything .NET.
Literally nobody on the planet is worse at naming things than Microsoft.
Apple's not great, but Microsoft is worse.
Nothing about Apple's naming schemes seems immediately rage-inducing. Sure, their stuff is bland, and I think it's stupid how people refer to doing things "on iPhone" instead of "on an iPhone", but otherwise Apple's products are mostly descriptive. Garage Band has to do with music, Pages is a word processor, iCloud is a cloud storage thing, etc.
But even the Labrador licking his own balls that someone else mentioned would be better than Microsoft at naming things. I'm surprised they haven't changed Windows to Microsoft Azure Copilot Platform .NET 365 yet.
Should I get an M4 Max MacBook Pro, or an M4 Pro Macbook Pro? Or a Mac Pro? Or skip the computer altogether and get an iPhone Pro Max?
I mean, c'mon. They are deliberately trying to be confusing.