What products and services will benefit the most from recent developments in AI technology? We’re keeping close tabs on the question, and we’re curious if startups will be able to best lever AI improvements or if cloud platforms are the best positioned. Or maybe it’ll be the companies building generative AI models themselves?
I wonder if we’re missing a key type of software in the discussion: the operating system.
The Exchange explores startups, markets and money.
Read it every morning on TechCrunch+ or get The Exchange newsletter every Saturday.
Last week, Microsoft announced a slew of new products and features, including Microsoft Copilot for Windows 11. Calling it an “everyday AI companion” that will live in Windows and other Microsoft products, the company says the product is an extension of the company’s previous efforts to bring Bing’s early AI tools to Windows. It’s also the first update to an operating system that I have been excited about in ages.
Ask yourself how much your computing experience has really changed from Windows 10 to Windows 11, or from the last version of macOS you used to whatever you are on now. (It turns out that I am running Ventura 13.5.2 at the moment. Who knew?!) The same question works for iOS and other operating systems I touch day to day. I bet that your answer is similar to my own: not much. Unless you’re using some variant of Linux, of course.
That’s because operating systems have become docile computing layers that mostly serve as a base to run other applications off. I don’t care much about what OS I use, because I use Chrome across all of them. I don’t really need to know much about the operating system that is running Chrome for me at that particular moment; I just need to use web apps, thank you very much.
This contented stagnation is not a bad thing, mind. The fact that iOS still shows you an app grid when you fire it up is because it’s a simple and intuitive way to show a user their arsenal of applications. Microsoft is similarly invested in the Start Menu, which has worked pretty well since I was trying to keep my parents off the phone so that I could stay online as a child. And the various flavors of Android have their own takes on the app drawer.
For the apps we use today, our mobile and desktop operating systems feel solid, sorted, and, compared to prior periods, pretty secure. And they’re mostly free if you pay up for new hardware. It’s a winning formula for consumers who are getting a lot of bang for their implicit buck.
Similar to how it seems that modern smartphones have been “figured out” and are now only being improved incrementally, I’ve felt for some time that major OS companies have been working to polish and hone, rather than reinvent. Given that consumers seem pretty sated, it’s been fine so far.
New AI tech could upend the status quo, and I think that Microsoft adding AI at the OS level could be a big step in that direction.
The voice assistant trap
Back when Microsoft began to move away from Cortana, I was bummed. Apple’s Siri and Google’s own suite of voice assistants had blind spots, so a potential competitor falling behind meant that there would be less pressure on rival products to improve. Given my chronic iOS usage and penchant for sending a slew of slightly garbled Siri-transcribed text messages, I have long hoped for the ability to get more things done by talking to my devices.
The question before us now is just how good Microsoft Copilot will be, and what it will be like to use it every day.
I am considering working for a bit from my personal gaming PC just to get deep into the Windows update. I cannot recall the last time I wanted to do that. Not that working on Windows is very different from doing so on Mac OS; it’s just that I tend to segregate my work and personal lives so that I can better enjoy the latter when I finish each day.
No matter: The possibility of a generative AI–powered digital assistant that can execute tasks for me at the OS level is incredibly exciting.
One problem with using AI within a particular app or service is that its intelligence is often limited to the data the application in question has from me or my company, and the remit of the program itself. Your CRM may have some really cool AI capabilities, but that’s not going to help you outside that silo. When we bring AI to the operating system layer, it has the potential to wrap around many, if not all, other apps and services, creating the possibility for something more powerful and broad.
Only an intelligent OS can string together complex tasks across apps from different companies — nothing else will have the permissions needed, I think. And now that we are moving toward that reality — I expect all OS providers to follow similar paths to Microsoft’s — we’re getting away from AI-as-point-solution and moving toward AI-for-all-day-use.
And since we are seeing generative AI tech improve rapidly, we may be entering an era in which operating systems feel and operate differently over time, and their updates could bring more than just security patches. Cool, right?
Yes, but there’s some downside to consider as well.
How do we pay for it?
Remember Windows Vista? Do you recall which version you used? Was it Windows Vista Starter, Home Premium, Home Basic or Ultimate? Did your company use Vista Business or Vista Enterprise? I presume that I had one of the Home versions, but I’m honestly just shooting blind.
All those different versions of Vista gave you different features and were sold at different price points. You could have spent quite a lot of money on Vista — up to $399.99 for Ultimate, apparently. Can you imagine paying that sort of dosh today for an OS update? I doubt it, unless you’re running a company, of course.
Since most of us get an OS when we buy a new computer, we don’t think about its retail value much. What’s more, we don’t think about how good a deal the baked-in cost really is. My latest gaming rig came with Windows 11, for example, and it’s been updating itself merrily ever since. I never think about its current version, the need to upgrade or the cost.
At least two of those things are going to change in the AI-OS era. The current version of your OS will matter more, as I anticipate that AI inside the OS layer will improve rapidly, meaning that I will want to upgrade more frequently. And I think that’s going to cost me.
It’s not cheap to run generative AI models, and while those costs will decline with time, if I am going to interact dozens of times per day with Windows Copilot, as Microsoft seems to expect, the world is going to generate billions of generative AI calls every day. Perhaps some of that compute work will be pushed to the edge, but I doubt all of it will be. That would incur a massive central computing cost associated with OS-level, generative AI digital assistants, and that cost will need to be paid for somehow.
Perhaps this is why Microsoft is working to get more ads into Windows over time; it is going to need a recurring revenue stream from Windows users to pay for their AI costs.
The company could also limit Microsoft Copilot to Microsoft 365 subscribers, but that would mean most folks won’t be able to use it. (There were 67 million Microsoft 365 consumer subscribers, per the company’s latest earnings report; there are around 1.4 billion monthly users of Windows 10 and 11, in contrast.) No, Copilot will need to be incredibly cheap if it’s going to be used by most people. This could necessitate even more territory in an OS for ads, which I know I will hate with a passion.
Still, my willingness to pay for digital services is greater than the average person’s, so I don’t want to let my personal biases cloud our view of the market. How companies answer this cost question could divvy up the operating system market in an interesting and competitive way.
So even the bad news isn’t that bad. And the good news is that while AI is interesting when used inside apps, we’re seeing the first steps toward bringing something truly new and powerful to the OS layer. About time; I can’t wait until I can just talk to my computers to get more done, faster.