Microsoft’s OpenClaw-inspired Copilot Features & Slowed AI Spending, Atlassian & Figma’s Lead in R&D

0:13Welcome, everyone, to The Information's TITV. My name is Akash Pasricha. It is Monday, April 13th. Today on the show, the Information Publish exclusive reporting on how Microsoft is tackling the open claw era head on. We'll talk about how the company is responding to the popular agent tool with our Microsoft reporter. We'll then talk about the state of the SaaSpocalypse with our Applied AI reporter who wrote a column about software R &D, and we'll also bring on a GP from Emergence Capital for his view on that topic. And we will wrap the show with our AI infrastructure columnist. She has a piece out today talking about how Microsoft might be playing catch-up on the issue of power for data centers.

0:53It's going to be a fun show, so let's get right on into it. Microsoft has been paying close attention to OpenClaw, and it has some ideas of its own on how it can compete. My colleague Aaron Holmes had some exclusive reporting on what changes the company is making to that effect. I want to bring on Aaron to tell us more all about his reporting. Aaron, welcome to the show. It's great to have you here. Happy to be here. How has Microsoft reacted to the OpenClaw craze? So essentially, you know, Microsoft wants to make its co-pilot tools more useful to people when they're doing white collar work, as well as just, you know, in general.

1:31And when OpenClaw, which to recap is this open source agent that was originally powered by Anthropics models, but can also use OpenAI's models. When that came out earlier this year, you know, I immediately heard that executives, including Satya Nadella and his deputies, were testing out the tool almost immediately and starting to kick around ideas about, you know, how Copilot could function more like OpenClaw. Specifically, that would mean being more autonomous and being able to carry out long-term tasks and work in the background at the same time that a human is using the same Microsoft Office documents.

2:11And as of yesterday, Microsoft confirmed to us that they are working on OpenClaw-esque features in Copilot. And I think the real challenge that they are grappling with is just how to make sure that those features are safe for enterprises and that, you know, an autonomous AI agent is not going to go too far or go rogue, as we've heard that they are sometimes want to do. Do we have any sense for how far along these efforts are when some kind of a product might be actually released? So they're very nascent and I'm told that, you know, depending on how safe Microsoft thinks they can make these tools, that the product plans could change or they could even be scrapped.

2:51But I've also heard that some folks at Microsoft are optimistic that these features might be ready to preview in time for Microsoft's developer conference, which is called Build, and that's in June. So we might get, you know, a first look at some of these features in the next couple of months, depending on how the development goes. And so I just want to confirm, you talked about these longer running agents running in the background. I mean, the way that I understood OpenClaw was, again, you could sort of use it to create your own assistant or fleet of assistants. I mean, what Microsoft is building, it's sort of closer to that category of being able to build multiple different agents that do different things for you.

3:31Yeah. And so the idea is, I think, like currently the way Copilot functions is mostly that you have to prompt it to do something and then it will do it. You know, for example, you can ask it to summarize an email or summarize your meetings for the day and it'll do that. And I think the vision here is to let customers create always on agents. For example, a personal assistant agent that is automatically scanning your inbox and will send you a digest at the beginning of the day with what you have coming up and what your priorities should be. Another goal is to have these agents be capable of basically working in the background without taking over your computer.

4:11So, for example, like you might direct the agent to reorganize an Excel spreadsheet while you're still working in a different tab of the spreadsheet, which is sort of an evolution of what Copilot is able to do today. So you and I have talked previously about how Microsoft has been leaning more on Anthropics models to help power many of its Office 365 AI tools in the background. How much are they leaning on Anthropics for this, given that I know Claude has been pretty central to the OpenClaw story? Yeah. So, I mean, it's true that OpenClaw was originally first made possible by Claude Code. The creator of OpenClaw originally called it ClaudeBot and then changed it for legal reasons.

5:00That person actually was then hired by OpenAI and is now more focused on promoting OpenAI models to power OpenClaw. But it's true that Microsoft has been increasingly using Anthropic models. They released some features last month called Copilot Cowork, which is basically like a reinterpretation of the Claude Cowork tools that have become extremely popular. At the same time, we've seen that Microsoft thinks it can basically capitalize on using both Anthropic and OpenAI models. For example, some features in Copilot will use an OpenAI model to compile a research document and then an Anthropic model to review and fact check that document.

5:43So at this point, they're very much interested in combining the models and basically doing whatever it takes to make Copilot work as well as any of the other state-of-the-art AI tools out there. So in other words, we shouldn't assume that they're only leaning on Anthropic to make this OpenClaw tool work. It's going to go to whatever model they need, and they have access to all these different models, essentially. That's right. And I mean, they get OpenAI's models for free, essentially, because they were one of the biggest early backers of OpenAI. But what's interesting is they do have to pay Anthropic to use its models in their products.

6:17And I learned earlier this year that Microsoft is paying Anthropic handsomely. You know, they were paying, I think, in the middle of last year, several hundred millions of dollars on an annualized basis just to get Claude models to power some of their products. And that spending has only risen since then. Last question for you on this topic. Who is running this OpenClaw-esque effort inside Microsoft? And what does that tell us about how much of a priority this is for the company? Yeah, so it's interesting because Satya Nadella actually did a pretty big reorganization of his top executives last month.

6:57And as part of that, he actually combined the consumer and enterprise organizations working on Copilot. So now it's all unified, the engineering efforts. And he elevated four executives to essentially all oversee Copilot in different ways and report directly to him. So one of those executives, Charles Lamanna, who has previously overseen business applications at Azure, is now the one overseeing this OpenClaw-esque effort. And he has appointed Omar Shaheen, who was a longtime office executive, to lead that team of roughly a dozen people within his org. I want to pivot very quickly, Aaron, to a report that CNBC published.

7:43they talked about a memo that OpenAI published over the weekend, I believe. And that memo talked a little bit about how much demand they're seeing for their joint product with AWS and then how much demand they're seeing for their Microsoft-affiliated products. Just walk us through exactly what the memo said, and then we'll talk about your reaction to it. Yeah, so it was a really interesting story. So yeah, OpenAI has this exclusivity deal with Microsoft, where Microsoft is the only cloud provider that's allowed to host and resell OpenAI's models to cloud customers. But when Amazon invested$50 billion into OpenAI earlier this year, they inked a new deal and agreed to create a joint product on AWS.

8:30And this product still hasn't come out, so we don't really know how it's going to work. But, you know, Amazon has implied that it's not the exact same as reselling OpenAI's models, but that it will give Amazon cloud customers access to build AI agents powered by OpenAI's models. And I'm also told that senior executives at Microsoft are concerned that that, you know, might run legally afoul of their contract. So apparently, according to CNBC, this fascinating report is that OpenAI sent a memo to staff essentially saying there's huge demand for this Amazon product and that maybe the Microsoft deal has limited OpenAI's ability to grow its revenue as quickly as it wants to, which is also interesting given that we recently learned that Anthropic has been rapidly growing its revenue and possibly surpassed OpenAI's revenue run rate.

9:20So if I just connect the dots here. So one part of the story, in my view, is AWS sort of coming closer into the fold here on the AI story, because a couple of months ago, we were talking about AWS seemingly had fallen behind in terms of the narrative on AI. this memo maybe suggests it, of course, has this giant$50 billion investment or up$50 billion investment in OpenAI. That seems to be working pretty well, it seems, in terms of getting it in front of the right people. And then on the other side, I mean, what was the anthropic implication? What was the last piece you said there? Yeah, I mean, basically, I think reading that memo that OpenAI sent to staff where they said that Microsoft's exclusivity has limited their growth.

10:08To me, you know, I think you could read that as possibly an explanation for why Anthropoc has seen faster growth among enterprises in recent months. And OpenAI is essentially saying like, look, we haven't been allowed to be on the biggest cloud provider until now. And maybe once that product debuts, we're going to see this explosion in revenue growth reading between the lines. Or another interpretation, I don't know, I'm asking, maybe they're just shifting the blame a bit? Like, you know, maybe there wasn't as much demand to begin with. I mean, I don't know. We'll never know. But the memo said what it said, I guess.

10:43So Aaron, I want to thank you for coming on. That is Aaron Holmes, our Microsoft reporter here at The Information. The SaaSpocalypse has continued as enterprise software firms try to convince investors they are protected against the AI threat. My colleague Laura Bratton has some new data that shows which companies are taking the biggest swings to protect their turf. I want to bring her on to talk all about that. Laura, welcome back to the show. It's great to have you here. Good to be here. Thanks, Akash. So you had this column out over the weekend looking at the financial profiles of different enterprise software companies, and you focused on a few data points in particular.

11:24Tell us about what you focused on. Yeah, so we focused on how companies are spending on research and development as a share of revenue versus sales and marketing as a share of revenue. And what we found was that Atlassian and Figma, in particular, spend a lot more on research and development as a share of revenue than some of their other software peers. And it's notable that a lot of the research and development figures include stock-based compensation. And, you know, Figma went public last year, so, you know, that raised its research and development figures a little bit. but I think it's notable still that these companies are spending so much more on research and development.

12:06So why did you focus on R &D? And R &D at software companies is what? Like it's just software engineers making new products, essentially? I mean, I don't anticipate it's people working in laboratories here. So I guess the reason we decided to look into this is because we were trying to get some measure of how software companies are investing in building new products because new AI agents and AI tools are in many ways threatening their competitive modes and how strong their traditional products are on the market. And so this was really a measure of how much they're spending relative to revenue on building new AI tools.

12:47Okay, so which company stood out to you then if you were looking at R &D? So Atlassian and Figma were really the biggest examples. Atlassian spends over half of its revenue on research and development. And Figma spends over 90 % of revenue on research and development. 90 %? What's typical for the software sector? So the software sector, it's typical to spend, I mean, in 2025 at least, in that calendar year. It was more like 18%. Yeah. Huh. And, I mean, I'm curious. Did you ask the companies about, you know, why they were spending so much or what did they tell you? Yeah. So, I mean, for Atlassian in particular, it's really vocal about how its R &D strategy is unusual and how it spends so much of its revenue on R &D, or I should say so much on R &D as a share of revenue.

13:42um and you know that's included in its risk factors and sec filings where elassian kind of talks about like you know we spend more on research and development than other software companies and this is a risk because if our products aren't competitive that could affect our you know financials um but then you know other software companies service now and salesforce in particular um their their revenues are much larger and they're much bigger companies so So just because R &D as a share of revenue hasn't necessarily increased a ton, that's also because its revenues are growing. But I still think that this is an important metric to look at because as we think about how companies are investing in AI and as we talk about, say, hyperscalers investing to build AI data centers, this is a helpful metric to look at software companies and think about how they're prioritizing growth.

14:29And some growth investors that I talked to said that, you know, they're looking at this and they wish that software companies, you know, spent more on R &D. But then others argue that R &D actually isn't that important. And it's how well you execute on whatever products you introduce. Now, if we put this story into the context of the SaaSpocalypse that we're seeing play out and all the questions surrounding enterprise software companies, I wonder if you just look at these two examples with Atlassian and Figma, do you see that R &D spending as an offensive move or a defensive move right now? Yeah, so I think if you were a growth investor, you would think of this as an offensive move.

15:11But then again, it's kind of like what came first, the chicken or the egg? Yeah. So, you know, I think it really depends on your perspective. Like in some ways, taking an offensive move is playing defense right now, right? And so, yeah, I think you could really argue either way in this scenario. But, you know, from an investor standpoint, I think the argument can be made that this is an offensive move as companies try to create new tools that are more competitive in a changing market. And let me ask you this. I mean, as you looked at this data, what kind of questions did you think about moving ahead?

15:52I mean, are you anticipating that more software companies will increase their proportion of revenue that's allocated to R &D spending over time? Did the analysts you talked to, did they say that it will need to come up to remain relevant? Or is this sort of just a case of a few outlier companies? Yeah, so that's a really interesting question and it really varies by the Wall Street analysts that you talked to about whether or not they think R &D spending should increase or not. In the SEC filings of the larger companies that I looked at, like Salesforce and ServiceNow, they said that they expect, you know, R &D spending as a share of revenue to remain stable as they increase their R &D spending to create, you know, things like AI agents or AI applications relative to revenue.

16:36and they expect sales and marketing spending as a share of revenue to actually come down because the whole argument can be made that as they use AI agents in their sales and marketing and research and development efforts, that those costs should actually come down because AI could make their workflows more efficient and the time to market for these products shorter. So, you know, some analysts think that R &D spending really shouldn't go up because they should be using AI to make these processes more efficient. Great. Well, Laura, I want to thank you for coming on. That is Laura Bratton, our Applied AI reporter here at The Information.

17:17To talk more about the SaaSpocalypse, I want to bring on someone who has studied the sector very closely for many years. Jake Saper is a general partner at Emergence Capital. Jake, welcome to TITV. It's great to have you here. Great to be here, Akash. Thanks. So we just finished up talking about a column that my colleague wrote about the proportion of revenue that has been allocated to R &D spending at some of the top software companies. And, you know, the piece got some interesting comments on it. One comment that someone left on the article was they said doubling down on R &D spend makes no sense when that's precisely the vector you're getting disrupted at.

17:55So that's kind of an interesting point. Where do you land in terms of how important R &D spending is and how you think that's going to go over time? I think R &D spend is critical, but I think the most important thing to be thinking about is what is being spent on. More engineers building more features on an old model is like adding horsepower to a horse. The real question is what is under the R &D spending? What are they spending it on? Most of these companies are spending to defend the old tool-based regime, which is like building more features, shipping more product. which is effectively continuing to sell a fishing rod.

18:31But AI is making the fishing rod less valuable. Customers don't want better tools. They want the outcome the tool was supposed to help them achieve. And so the better question is, which companies are using R &D to shift from selling tools to delivering outcomes? Okay. How do you measure that? I mean, I'm thinking about the financial statements. I mean, all we have is a line item. It's pretty wild. So what I'm suggesting here is these companies should think less about trying to perpetuate the product that they have been selling and think about delivering a different form of value. And so when you think about the financial statements, it's going to be hugely disruptive.

19:11But I think the companies that don't actually take this leap are dead in the water. Ultimately, if you are selling a tool to a buyer who is going away, it doesn't matter how much R &D you spend on making the tool better, you're not going to have a business. So what I'm encouraging, and this is a wild idea, but what I'm encouraging a lot of SaaS companies to think about is changing what they sell, not selling software, but selling a guaranteed outcome. Okay. So sounds to me like you are very much on the side of AI is going to kill the SaaS sector altogether. No, I wouldn't say that because I think a lot of SaaS companies are actually going to make this shift, right?

19:50A lot of companies are going to make this transition. And I also think for many companies, particularly those in like highly regulated, complex environments, this is going to be a very long transition. But what I do think is that every SaaS company that survives needs to think outcomes first. I think historically you've thought about how do I spend R &D to make the tool better, to make my users happier? I think the only SaaS companies that will survive are going to change their metric to what is the thing we are selling that is delivering an outcome and how do I quantify that outcome? so to take an extreme example of this to build on the article that you guys published what if Figma instead of selling tools to designers and people that did design sold design like what if they sold um a finished designed product right and they stood behind the outcome so instead of hiring Figma to help you do the design yourself what if you went to Figma and said, I want this design done.

20:45And Figma, using a combination of AI and the world's best designers, delivered that outcome. But I mean, that sounds more like a services business than a software business. Yeah, that's exactly what I'm saying. So then what happens to the margins there then? Yeah. So I think that if you are able to deliver that primarily with AI and with a human wrapper on top, you should still be able to achieve reasonable margins. It is likely that the margins of the future are not going to be the 80 % margins of the past. But I do think, and I'm seeing this now because we have a full investment practice around what we're calling AI native services businesses.

21:18These are services businesses that are powered primarily by AI. And I'm seeing these companies achieve 50 and 60 % gross margins consistently if they actually have an AI tool that's delivering most of the value. So, I mean, this is kind of interesting because I'm trying to sort of understand this and wrestle with it because on one hand, we say AI is going to lower your cost base and has the opportunity to expand your margins what you're saying is yeah you know you might expand your margins if you're selling the same thing but if that thing becomes obsolete and you have to change it then then basically i mean the pricing practice will become different overall so everything will change everything is going to change and it all starts with the builders of these of these companies shifting their mentality from maximizing margins and trying to maximize price per seed or whatever the legacy metric was to how do I deliver a value and stand behind it.

22:13So here's a very wild idea, and I don't know if this is going to happen, but in a future where let's just say that code will become commoditized, right? And there's lots of arguments as to whether or not this will take place, but let's just assume for a second that code creation will become commoditized. What is the value of a technology vendor at that point? The value of a technology vendor in a world where code is commoditized is effectively as an insurance provider. They are ensuring that an outcome gets done and their brand reputation and finances stand behind it. So again, to use the Figma example, imagine a world where you hire Figma not to be a design tool provider, but instead to be your designer.

22:54And Figma stands behind the output of what they provide. And if it's not good, there's some sort of financial remuneration. We're seeing this now with a bunch of companies that we've invested in. We've invested in an AI native fund administrator, an AI native insurance broker. All these companies are selling an outcome and they have the guarantees behind that outcome, but they're using AI with a human wrapper on top to do it. So, okay, so that's a pretty fascinating view of how the sector would play out. I mean, because I know VCs like to take both sides. I mean, what would the opposite reality be?

23:35I mean, what would be the arguments against what you're describing happening be? Well, I think when you think about AI native services, the most obvious argument against it is the argument you made before, which is that it'll never scale from a margins perspective, that this will just be, you know, services 2.0, and it'll be 20 % gross margins, and it won't be venture backable. I just, I'm not seeing that play out in practice. Like ultimately it is true that AI can do 80 % of the job that these services used to do. And so there is a 20 % of the job that remains where you need a really highly skilled human rapper to check the output and then guarantee it's right.

24:12But I'm seeing it work. Like I'm invested in a company called Hanover Park, which is an AI fund admin. Right, yeah, we've had them on the show. Yeah, you've met Chris. And like I'm on the board, I can tell you the margins are very good. okay let me ask you about another uh scenario that i've been thinking about and i i believe if i'm not mistaken this was jensen who who brought this to my attention i don't think he was the one that that originated the idea but the you know he he was asked about the sas apocalypse um and you know he made the point that well people are so worried about about uh companies not needing as many licenses uh in reality you know maybe because they're hiring less or what have you.

24:54In reality, you might have this team of agents, and each agent might actually need a license. And so the license model that everyone's worried about going away actually could maybe expand a little bit in the short term. Do you think there's any validity to that? I think it's possible, but I think the more interesting businesses will not be those that sell software to agents, but those that use agents to deliver an outcome. By definition, you can actually capture more margin if you're providing more of the value in the stack. So instead of selling agent software to agents that would do fund admin, why wouldn't you just do all of the fund admin?

25:29Right. Right. So I'm not saying that's not a viable business model. I'm just saying I think it's going to be less interesting ultimately. To your earlier question on why this won't work, I think the transition from building software companies to building AI native services businesses is going to be a very foggy and messy one. the lessons we learned in the SaaS era on how to price, how to build product, how to go to market, all those things are very different in the AI Native Services era. So what I and we at Emergence are trying to spend our time on is to document this transition. And we've written this playbook that literally is a guide for founders on how to make this transition we have on our website.

26:05It is going to be a messy transition. Lots of folks are going to do it poorly. There'll be lots of what I call the Mirage product market fit, where you have these services, AI native services businesses take off really quickly with great revenue growth, great customer retention, but they don't actually have product market fit because they haven't actually built AI that's delivering most of the product. Ultimately, it'll still be mostly humans to your earlier point. I think that's the biggest near-term risk with this business model is it's just going to be a bunch of humans that are masquerading as AI.

26:34I think some of the companies, just as a tidbit from our playbook, are actually pausing sales periodically so that they can ensure that the delivery of their service is really high quality, which is something you would never do in a software era. So the way you build these businesses is different. And unless these founders learn these lessons, I think that you'll have a lot of flame outs. Great. Well, Jake, I want to thank you for coming on. That is Jake Saper, General Partner at Emergence Capital here on TITV. Investing in not just data center capacity, but also power has become quite the balancing act as companies try not to fall behind, but also stay prudent in terms of spending.

27:14Microsoft knows that dynamic all too well, but in the information's latest AI infrastructure column, Ann Davis Vaughn argues the company might be playing catch-up. I want to bring her on to walk us through her reporting. Ann, welcome back to the show. It's great to have you here. I am, Cash. It's great to be here. So let's talk about Microsoft's power investment strategy. Walk us a little bit through how that has changed over the past few years. Well, so I think it's important to start with the fact that Microsoft has been very early on the AI power needs story because it provided the training capacity for ChatGPT, which kind of started a race for all the hyperscalers as of about November 2022.