Episode Transcript
[00:00:00] Okay. I have a bold story to tell you about the world of AI and the economics and business models of AI and why Microsoft could turn out to be the leader in this whole space.
[00:00:15] And it's a little bit of a history lesson, but I think if you bear with me, you might agree with me.
[00:00:22] So back in the fall of 2022 when ChatGPT was launched, we all became enamored with this stuff.
[00:00:29] And Microsoft did a deal with OpenAI. And the original business model was that ChatGPT or OpenAI would be the engine in the Microsoft Copilot. And that was obviously really good for OpenAI and really good for Microsoft. What happened for the last three years is the Microsoft engineering teams went a little haywire and started building copilots in all the tools.
[00:00:56] There was a standard Microsoft Copilot, then there was a Microsoft 365 copilot. Then there was a Copilot in Excel, there was a Copilot in Word. And from what I understand, these engineering teams built, there was a copilot in Dynamics, a bunch of them. And Satya sort of let this all grow because Microsoft is a very innovative company. The engineering teams there do tons and tons of stuff. When you go visit them, you see lots and lots of things they've built internally that don't turn into products. And a lot of these got launched and the Microsoft Copilot studio was built. Agent 365, which is an agent management system, was built. They built out the Microsoft fabric, which is an infrastructure to connect to heterogeneous data systems. They built graph connectors so you could bring data into the Copilot through the Microsoft Graph process. And then they launched Work iq, which is an open architecture for data and context, expanded beyond Microsoft 365 data. And that was sort of all becoming more and more confusing until recently when they've announced the introduction of Anthropic Claude as one of the models. So I sat down with Seth Patton, who's the head of Microsoft Copilot product Marketing. And I've known Seth for several years. We talked a lot during the days of Microsoft Viva, when that was launched. And he walked me through the big picture. And at the end of the meeting I took a step back and I thought, wow, these guys are going to walk away with this market and it isn't that different to what happened 30 years ago, or maybe it was 20 years ago when Windows was launched. But I don't want to spend too much time on that. Let's just talk about this. So where we are with these tools is Anthropic, OpenAI, Gemini from Google. There's a bunch of stuff going on in Meta and there's others are slowly becoming use case dependent AI models that we all want to use. No one model does everything. They've been trained in different ways, they've been optimized in different ways, they're designed for different use cases. Both OpenAI and Anthropic are trying to go public in the next year, maybe this year. Their revenues are similar in the 20s of billions. In fact, it now looks like Anthropic might be bigger than OpenAI because they've been more focused in their products. And by the way, the products sit on top of the model. So it isn't the model that you're really buying, you're really buying the experience on top of the model. And so when OpenAI built Sora, that was a product built on the OpenAI engine and they discontinued Sora because they found that the amount of compute capacity needed was so high that they weren't going to be able to monetize it. On the other hand, Anthropic made other decisions to optimize around code generation.
[00:03:53] And you can see when you use Anthropic versus OpenAI, you get a totally different experience because the anthropic model is just better at building stuff and visualizing it. And there's going to be other models because the technology and the algorithms and the science of the large language model is in the public domain now. So kids in college can learn about this, they can get their hands on open source stuff and they can build it. So I don't think this is going to be like the relational database industry where it all became centralized around one or two companies. There's going to be a bunch of these out there and there are going to be models that work in small devices and there's going to be models that work in industrial machines. And there's Grok, which runs in the Tesla car or perhaps the Tesla robot, if there is such a thing.
[00:04:45] So the market isn't one model or which model you buy. The market is the application platform that you build on top of that uses the models that you want. Well, either on purpose or inadvertently, Microsoft is landing in that direction. Because if you look at the newest version of the Copilot, which we have, and Seth went through this for me, it can transparently take your problem and send it to more than one model and not only pick the one you want to use, but show you which one is better. Which means it could have an intelligent router that could take all of the queries you want to do and send it to the model that's the most useful or the most powerful. And by the way, there's a big cost issue on this. All of these AI companies are starting to jack up their prices because they want to make some money and go public. So you don't want to spend a bunch of money running tokens on Gemini and find out it gives you the wrong answer, and then you got to go and spend the same money again on Anthropic and find out it didn't give you the right answer and then find another model to give you the answer. You're just wasting money. It's like having five electric companies and they provide different sources of electricity. It isn't actually like electricity because this is not fungible. These things do different things and they act in different ways. And they give you different answers too, because they have different data sets underneath them. In the model. Companies, the, what are called the labs, spend a lot of money not just on infrastructure, but on data labeling. So they have in some sense proprietary data labeling that's not the same between the different models. So if you use it for research or science or domain expertise, different models have different levels of expertise. And that's probably going to continue unless they all combine for some reason. But I don't think that's going to happen either. So in some sense the value is moving up from the back end to the front end. Because if you're a big bank and I, and we talk to big banks all the time, they don't want to, you know, pick one of these platforms and find out it's kind of good for this, but not very good for that. And then they have to have a license to all of them. They would like a front end system that optimizes the experience for all of the use cases and allows the either the engineer or the user to pick which model or the system itself it needs. And of course they want a development tool that sits on top of that. They want a semantic layer or a context layer that connects that to all the backend systems they have and pulls the business rules out.
[00:07:22] And then they want integration with the desktop, which happens to be 70 to 75% Microsoft Word, Excel, PowerPoint, my schedule, my messages that I'm sending in teams, the recordings of my meetings. And so even though Cowork is fantastic, Cowork doesn't really integrate with Microsoft by itself, although you can get Cowork in the Microsoft Copilot now. So in some ways Microsoft wins again.
[00:07:52] Thinking back about the days of Apple versus IBM versus Microsoft, for those of you that like history because they have such a massive install base and such a massive customer base and such a massive value proposition in the desktop. And my new laptop that I just bought from Lenovo has the Copilot embedded. So I don't have to install it, I don't have to buy it, I don't have to license it, I just click on it and all of a sudden it works. So you argue. And for somebody my age who's watched this happen before that, the early stages of this, with anthropic and OpenAI and Google Cloud and all them selling lots and lots of capacity to other providers, including Amazon, that's not going to go away. But the corporate end user market and the corporate development market could move towards Microsoft. Now my experience with Microsoft over the years is they really do build a lot of software and they launch a lot of things. And if you're not a Microsoft watcher or a big Microsoft shop, it's very confusing which of the Microsoft products is going to be around for a long time and which is going to go away at any given moment. So it's confusing to buyers sometimes which part of the Microsoft suite is the most important. But if you take a step back, which is what I'm doing, and you look at what they're doing with the user experience by the user, the new user experience of the Copilot is very elegant and you can see where all the built in applications go. You can see where the Galileo Copilot application goes. By the way, we're going to launch the Galileo solution for Copilot at our conference in June. Most likely. So it's coming along and it's going to blow your mind, but we'll talk about that later. So they really have built a nice elegant user experience. You need to sort of play with it and we can show it to you if you want to see it.
[00:09:47] And they have this entire library of tools underneath it now. One of the things that I think is also very significant about where Microsoft is going is this confusing thing called Work iq. Work iq, which I don't think they marketed it very well in the beginning, is the unlocking of the data interfaces into the Copilot or into the AI layer. By the way, people aren't calling it AI Layer, they're calling it the AI Surface. Now the Surface, the user tools are now called Surfaces, just so you understand that word. So the AI Surface can access all sorts of data systems. Not just access the data, but build a context layer on top of it. Now, I just did a podcast on Context Layers and wrote an article on Substack about it. But the context layer basically says if I'm going to connect to some, you know, sales data system or a bunch of customer data, or an Excel file with a whole bunch of Usage data in it, or a financial system in QuickBooks or a financial system in Dynamics, connecting to that database is sort of a big step, but it's not the end user. I need to know what that data means and how it's organized and what is the metadata and the context around it. That's called the context layer. So what you can do with Work IQ is they're formally opening up the API to the context layer. Companies can now build much more sophisticated data integration. This is what we're doing, by the way, with Galileo. Much more sophisticated data integrations into the copilot. So the copilot doesn't just pull a number out of some spreadsheet or some financial system or some CRM system. It will know as much as you want to tell it about that system. Which is really the reason we want AI to replace a lot of these legacy systems. All the legacy systems we have in hr, recruiting systems, learning systems, hcm, payroll, all that have very important business rules and business data and business artifacts, skills models, career models, regulations, and stuff that we've built into these other platforms. And if we can't pull that business rule or security and privacy rule data into the AI, we can't easily, we can't rebuild it all from scratch. Then we can't leverage our legacy systems into the new world. Well, Work IQ is Microsoft's answer to that, and they haven't really positioned it very well yet, but it's coming. That's where it's going. One of the other announcements that came out this week I talked about the other day yesterday, I think, was what is called the context engine from ServiceNow, which is their attempt to do it. Gloat has announced a product that does this, and I think there will be others, because this is sort of the lifeblood of IT technology companies, and I wouldn't be surprised if there's quite a few.
[00:12:44] And once Microsoft sees momentum from software companies and ERPs and application vendors into Work IQ, it will become more and more obvious what's going on. One of the other way that I look at Work IQ is through the digital twin solution that we have. So we have a solution here. From a company called Wisdom Labs. Vivint is the product name that indexes into AI all of our emails, meetings and documents in the Microsoft graph. You know, you would sort of expect this to come out of the Microsoft product set, but they don't have it yet. But they will. And what you can do with our digital twin is I could, I can go to Barbara, one of our salespeople or Bill or our accounting folks or whatever and I can ask their twin, what is the status of this contract?
[00:13:36] What are the last three conversations you've had with them? Can you send me the last 10 emails we've talked to them about? Or I could ask, because I'm at the top of the pyramid, I could ask the whole company, show me all of the emails we've sent to Company X in the last 60 days so I can get caught up really fast on what's been going on. That's a huge use case for customer service, sales, marketing, all sorts of internal operations, nearing knowledge management applications that is enabled through this work IQ interface that will all go into the copilot. So what Microsoft's going to start doing, and I'm sure their salespeople are doing this now, is they're going to basically go to companies and say, look, oh, you like OpenAI great, you like Claude for code generation, that's great. You know, I don't know what they're going to do about Gemini, maybe they'll have a truce with Google. But just use the Copilot. You can access all that stuff and you can access all the content in your Microsoft system and you can access your calendar and you can access your emails and you can can access these other applications and Galileo and all that stuff. It's all there. And you'll see in the interface, you just click a button and you can get to it. Or you can just look at it globally and ask a question and let the copilot decide which of these engines to use. And the interesting thing they're doing on that, as you can sort of read about from them, is the newer versions of the Copilot interface negotiate the different backend systems and then show you a comparison of which one's better. So you could do a test before you spend a bunch of money on one of these backend systems to decide which of the directions you want to go with your AI problem or your AI application. Now from a financial standpoint, I did a little bit of back in the envelope math here. Microsoft has claimed in the public announcements that they have about 15 million licensed users of the Copilot today, I'm sure that number is going up. If you multiply 15 million by $30 a month, it's five and a half billion dollars. And the 2025 revenue from Anthropic is about $8 billion. OpenAI claims to be 20 billion, but a lot of that revenue is trading AI credits to software vendors that sell end user products. So this $5 billion of Microsoft end user revenue is huge because not only is that number going to grow because there's 450 million or more users of Microsoft tools. So, you know, they barely scratched the surface of that. But there's a pricing model out there that's $99 a month, not $30 a month. So it could end up being 15, 20, 25 billion within a year, and they'd be the Same size as OpenAI on end user revenue. And then the question is, who has more market power? Microsoft with the end user licenses, or OpenAI or Anthropic with the compute and intelligence license. And what happens in at least the business world, and I don't know about the consumer world so much, is that the company with the contract with the corporate buyer tends to be the one that has the most power. If you're Anthropic and you go to Walmart or you know, pick a bank or whoever it is, and you say, we want to do a big deal and we're going to sell you a whole bunch of credits, yeah, they're going to negotiate and yeah, they're going to do a deal. But if Microsoft has a contract to provide the same service through a better Surface, maybe at a reduced rate per credit because they're buying it in volume and reselling it, you know, that deal might go to Microsoft. Now, I'm not saying that's going to happen, but it could happen. And if you fast forward a few years and Microsoft's not going away, there's. There's probably a hundred times as many Microsoft salespeople and accounts than there are Anthropic or OpenAI. Those companies are pretty small. I can see Microsoft just wiping the other guys out and they become labs again. Now they're not going to go away. They're certainly not going to take this sleeping.
[00:17:41] They're aggressive companies. They're going after applications.
[00:17:45] They go after the new tools vendors, they sell compute to advertising vendors. Like one of the biggest consumers of AI tokens is Meta, because Meta's AI is not that far along yet. So they still buy a lot of AI services from the other guys. And you know, you look at other companies like Meta, they have to buy AI from somebody too and you know, eventually I suppose Oracle will be a provider of this stuff too. So I just think the Copilot, which came to market very early and it's been a little bit hard to pin down what IT is and a lot of customers I talked to licensed it and are not sure what they're doing with it yet because they haven't built applications on top of it does have the potential to be maybe the predominant front end to corporate AI. Now somewhere around 20 or 30% of the corporate world uses Google. Those are mostly smaller companies, they're not the same size companies as Microsoft. So the Google world, those of you that like Gmail, I hate it, I can't figure out how to use it, are, you know, not, not going to use the Copilot, obviously and maybe that's fine. We don't know what Apple's going to do with the, with the phone and you know, most of us that use Microsoft have iPhones, so there's a player there. But I have a hard time seeing anybody else possibly reaching this level of Surface activity with AI, at least in the near term unless something big happens. So I'm going to write an article about this and talk more about it. It's not super relevant to those of you working on HCM HR HR 2030 stuff, but it is for your IT counterparts because your IT counterparts are probably buying licenses from all of these vendors and experimenting with them. And the Copilot is a little bit less mature in some respects, but it's getting better. Like, you know, for example, if you use Excel a lot and you stick an Excel spreadsheet into Claude, it does an amazing job of analyzing data. I mean really unbelievable. But in the new version of the Copilot, where the Copilot resides right next to Excel, you can use the Copilot to analyze and manipulate Excel while Excel is running in parallel. So they're getting pretty good, they're coming along. I mean, I think the PowerPoint generation stuff is, still has a ways to go and there's lots of things yet to be improved. But now that they have a single Surface and a single user experience strategy, I think it's going to get better and better. We don't exclusively use Microsoft, we use a lot of other things here too. And you know, there are issues and Microsoft's not perfect as is anything else, but they're very good at seeing these big trends. Their go to market strategies are extremely powerful. They have great client relationships.
[00:20:36] Many, many Microsoft trained professionals know how this stuff works, so it's easier for them to adopt it. And, you know, maybe the other big vendors still continue to grow and they become technology underpinnings under underneath a lot of Microsoft shops. And some of them will have end user applications that'll be successful, too. If you look at Sana, which is a, you know, now owned by Workday, the Sana experience is exceptionally good, too. And there will be customers that will use SANA for some users and COP for other users. So, you know, it's not like everybody's going to use one tool, but I don't think you can count Microsoft out. And so I just wanted to reflect on this a little bit for you guys to think about. I'll write an article on this so you can see more about what's going on. Thanks a lot. See you guys later.
