Why Oracle is Cutting Thousands, Claude Code Leak, Lessons from Google Deepmind CEO Demis Hassabis

0:13Anita Ramaswamy:Welcome everyone to The Information's TI TV. My name is Akash Pasricha. It is Wednesday, April 1st, April Fool's Day, also known as my birthday. That is no joke, folks. Welcome to the birthday show. We got a fun one for you today. A lot of news to get to. First up, Oracle is laying off thousands of employees, according to CNBC. This comes as the company's AI spend has been eating into its cash reserves. We'll discuss the news with our financial analysis columnist. Next up, the information published exclusive reporting around Amazon's AI ads effort in its own Rufus chatbot. Our Amazon reporter, Catherine Perloff, will join the show to share what she knows.

0:56Anita Ramaswamy:Anthropics data leak is in the spotlight. The company mistakenly revealed details about its next big Claude mythos model. The information covered it in our AI Agenda newsletter. We're going to bring on one of our reporters to help us break it all down. We'll then pivot to our special coverage of the information's 2026 NextGPs list. We've got a partner from Radical Ventures coming on, and we will wrap the show with a great conversation the information's editor-in-chief, Jessica Lesson, had with Sebastian Malaby, author of the new book, The Infinity Machine, which digs into one of the most visionary AI minds, Demis Hassabis, CEO of DeepMind.

1:35Anita Ramaswamy:It's going to be a fun show, so let's get right on into it. Oracle will lay off thousands of the company's workers, CNBC reported. It would be the latest company to cut staff as the entire sector tries to save up and try to better afford its mass investments into AI. I want to bring on Anita Ramaswamy, our financial analysis columnist, for her view on these cuts. Anita, welcome to the show. It's great to have you here. Thanks, Akash. So what did you make of these cuts? Tell me. So, you know, I think that this was something that the market and investors thought was a long time coming. For a while, if you look at the revenue per employee, which is a metric that we've looked at across a lot of different tech companies, Oracle has been operating a lot less efficiently than its peers.

2:20Their revenue per employee was around$350 ,000 for the most recent fiscal year. And if you compare that to companies like Microsoft, like Palantir, like ServiceNow, it's a lot lower. And so I think this was sort of a long time coming given Oracle's recent push to get into the cloud computing business via data centers that they're building out for AI, and that transition from going from a high margin software business to a very costly and expensive business to run.

2:45Anita Ramaswamy:So$350 ,000 per employee. They obviously have all the fixed costs in there as well after paying them. What do the company's margins look like right now? So the margins have been going down, Akash. And we saw them having a gross margin of around 70 % for however many years. And in the last year or so, each quarter, we've seen that tick down a little bit, a little bit. And in the latest quarter, the gross margin was around 66%. And this is, like I said, a result of the transition from a really high margin software business that doesn't have a lot of costs involved and into a lower margin cloud computing business, which has the infrastructure costs, the hardware, the data centers that you have to build.

3:27And that is a very expensive endeavor. And so we've seen that really impact Oracle. I mean, I think the other thing about their financial position right now is they are looking to cut everywhere that they possibly can. Oracle has issued tens of billions of dollars in debt to sort of finance this AI data center build out. And I was looking at some analysis about what exactly these layoffs are going to do for them. And it looks like analysts from TD Cowan are saying that the job cuts will free up around$8 to$10 billion in cash flow, which is really crucial for Oracle right now.

3:56Anita Ramaswamy:And to service those interest payments that they'll have to pay for all that debt as well. Yeah, absolutely. Now, I'm curious, just taking a step back from Oracle, we've seen other companies cut. Obviously, the block layoffs were what made a lot of news. There is this fear that tech companies across the board might use this window to cut and just blame AI efficiencies. I wonder if you've thought at all about, you know, which companies might be more likely to cut. You know, is it the hyperscalers? Is it the enterprise software companies? I mean, how are you thinking about that? Yeah, it's a good question.

4:36I guess let's talk about both. So we can start with the big tech bucket, the hyperscalers, if you will. We have already seen these companies do a lot of job cutting. We saw cuts from the likes of Amazon. They laid off 14 ,000 people or so last year. They announced another 16K this year. we saw Meta let go of 10 % of its Reality Labs unit, which happened earlier this year, so 2026 also. And they announced another round of job cuts even more recently than that. Microsoft has, you know, they didn't grow as fast during the pandemic in terms of their headcount growth as some of the other large tech companies did, but they have also frozen hiring at this point in time.

5:13So I think maybe some of the pain in big tech is behind us and it's possible that they will continue to cut jobs and we'll see how expensive the AI buildout ends up being. But I think a lot of that pain is already in the rear view mirror. Whereas when it comes to midsize software companies like the Salesforce, the Oracles, the SaaS sellers of the world, I think there might still be room to cut headcount. We saw Atlassian as the company that comes to mind, they cut 10 % of its workforce recently. And even Oracle, they teased this layoff. It wasn't like it came entirely out of left field. Their co-CEO said earlier this month that AI was helping their smaller engineering teams to deliver more complete solutions.

5:56Now, I don't know if that's the entire picture. I don't know if it's just really about AI making their employees more efficient or if it's a combination of that plus trying to cut costs wherever they can to finance their AI build-out. But I think that in general, across the board, these software companies tend to be a lot less efficient in terms of their revenue per employee. So that suggests to me we might see further cuts there. I mean, when I was looking at Salesforce, their revenue per employee, kind of similarly to Oracle's, was around$500K. Whereas if you look at Microsoft, they were making around$1 million or somewhere in that ballpark per employee based on last fiscal year.

6:31Right.

6:32Anita Ramaswamy:And I mean, we also we had Jackson Ader on the show last week talking about stock based compensation, stock based compensation and, you know, how that in some cases has gotten too high. And I imagine cutting would help you sort of help make things more efficient in that respect as well. You know, the last question I want to ask you, Anita, is this is just an open question. I don't know that we have an answer to this, but I wonder what all these people are going to do. I mean, you have all this big tech talent that is getting let go from these companies. Obviously, the other tech companies are not going to be hiring as quickly, and we know that AI is affecting that too.

7:10Anita Ramaswamy:I mean, I'm sort of wondering if what, these people are going to try to found their own startups? You know, maybe this is another wave of early stage founders that are coming on. Do you talk to people about that at all? Yeah, I mean, I hope that there will be new startup formations from this. I think that would be a really positive dynamic for the entire ecosystem. I think, you know, we've seen some of these companies, like even with Block, there's a Block Mafia Slack channel and all the ex-employees are kind of chatting in there and talking about their new ideas. I imagine that there will be some dynamic of that, especially considering this is tens of thousands of tech employees who are looking for new opportunities.

7:48Another interesting thing about Block was that there were a couple just on the fringe, on the margin, employees who they actually rehired after the layoff, either folks whose managers advocated for them or you know other sort of niche situations and so it's not a widespread phenomenon but i wonder if we might see some element of these larger tech companies saying oh we actually cut too deeply and we need some of these people in order to further our core business operations so very possible we might see that happen but only time will tell yeah it very much

8:18Anita Ramaswamy:in that case it sort of feels very much like a restructuring it's like hey you know this is coming across as a 10%, 20 % cut, whatever. Really, what we're saying is we just needed to put the people in different divisions, and so maybe we rehire them into different groups, reprioritize. Certainly something interesting to watch. Anita, I want to thank you for coming on. That is Anita Ramaswamy, our financial analysis columnist, here at The Information. OpenAI has generated a ton of buzz for its advertising play inside its chatbot, but Amazon is also quietly trying its own approach to ai ads with its rufus chatbot this is a story that few people have been paying attention to except for marketers who in some cases say that amazon's chatbot is actually better for them our amazon reporter katherine perloff published an in-depth story on that dynamic today and i want to bring her on to talk all about it katherine welcome to the show it's great to have you here hey akash and uh happy birthday thank you i hope that doesn't become a thing on the show from here Um, okay.

9:22Anita Ramaswamy:Uh, Rufus is the Amazon chatbot. I, I, I've seen it, you know, I'm on my app. There's a little tab, I think in some cases, talk to Rufus. Your story suggests that there is, uh, advertisements happening in the chatbot. I mean, help us understand this. Yeah. So, um, Amazon introduced Rufus as a shopping chatbot two years ago. Um, and that was to sort of help customers, you know, better find the right thing they wanted to buy. Um, So starting in November, they tested this kind of format called sponsored prompts. And the way these work is they are sort of suggestions, suggested queries that come up as like sponsored.

10:01So like, you know, why is X organic cereal, you know, how much fiber does this cereal have and, you know, might have the brand name in it. And the suggested prompts come up in search results when you're shopping and you're searching on the main Amazon search bar and then also on the product detail pages. So the ads kind of come up. The ads sort of are a question that if a user clicks on, they can start chatting with it. And it's sort of the ad is prompting a user to sort of start asking questions about a product.

10:38Anita Ramaswamy:So the idea here is it's not like when I go to the Rufus chatbot and I'm searching for things, it's not like an advertisement comes up the way that OpenAI is putting ads into ChatGPT. It's actually more if I'm on a page for a product, I might then see the roof of chap out on the side. And the sponsored piece is the marketer or the vendor, I guess, has paid to have a question there saying, tell me the health benefits of XYZ, stuff like that. Yeah, exactly. So, I mean, right now, the sponsored prompt product they're sort of, you know, pushing and have just made generally available is not actually an ad that comes up if you just type anything into Rufus.

11:22It is something that I kind of try to think about it as like it's not when the user starts the conversation, the ad starts the conversation. And I think it's kind of creative that Amazon did it this way because the problem with chatbot ads is that if you say what is the best cereal, it's possible that General Mills pays for that. But then Kellogg's is the answer. And that would be kind of an awkward situation. So I think that by this sort of ad format where they sort of situate it within the context of a product, like, you know, say on the product's like page, you kind of don't get into that issue where you're sort of recommend the chatbot might recommend a competitor because you're only talking about, you know, one particular product.

12:07Anita Ramaswamy:It's kind of a funny product, I feel. I don't know. It's like planting leading questions for people to ask. I don't know. Are people liking it? I mean, do marketers feel good about this? Tell me. Yeah, I know. I think it's just hard to really conceive of ads and chatbots, and that's part of the issue with all of this. But are marketers liking it? Well, so far, the results haven't been gangbusters, but it is still early. The beta started in November, and they just released it more widely last month. But yeah, it's not generating a ton of clicks. The ads aren't generating a ton of sales, but they are cheaper than regular Amazon ads, which is always good.

12:45And I think one of the biggest things that advertisers like about them is advertisers are sort of desperate to understand how people are engaging with chatbots and how user behavior is going to shift in chatbots. And so these kind of sponsor prompts, Amazon comes up with, and they draw from a lot of data. And marketers are kind of using these sponsored prompts almost like tea leaves to help them understand, you know, what are user questions in chatbots and how, what kind of questions might their potential customers be asking about their products. And like one example of how this could play out is an ad agency exec I talked to was able to use the sponsored prompts to sort of persuade one of his clients to put more details on their Amazon page that maybe before they were a little reluctant to share.

13:36And in this scenario, it was like, why did one version of a product cost more than another? Because they could see that the sponsored prompt questions were asking about that.

13:44Anita Ramaswamy:So the marketers like it better than the OpenAI advertising strategy. Yeah, I think so far because – I mean they just get a lot more information. OpenAI hasn't shared that much information about the reporting. You get kind of standard reporting with Amazon ads. And yeah, I think it's a lot more – there's like more details and it's a bit more thought out right now. So marketers can sort of intuit more from the experience. But I think neither are really like driving a ton of sales. So that has to be said. Right. I mean, the broader question I have here, Catherine, is Amazon obviously has this huge ad tech business.

14:28Anita Ramaswamy:It's a legacy player in that arena. It has a lot of the back end infrastructure, I imagine, getting marketers onboarded, helping them figure out what to put dollars against, reporting metrics back to them, knowing which metrics are even worth tracking. do you think OpenAI stands a chance here? And maybe it's not even OpenAI, maybe it's other chatbots. I mean, we know that Amazon's chatbots haven't really gotten the traction that some of the other labs have gotten, but I don't know, maybe they can make more money from it. Yeah, you know, I guess that is the big, the thing is about the users. So like, you know, there's two big things that advertisers care about.

15:07Like A, are people looking at their ads? And B, is the ad tech good enough that it works and it drives products. And Amazon says that more than 300 million people or customers are using Rufus or used Rufus last year. About a third, I guess, of the weekly active users that have opened AI, but still not.

15:26Anita Ramaswamy:Yeah, I mean, maybe like, am I included in that? Because I definitely opened it, but I don't know that I'm using it actively, really. Yeah, I don't know how they're counting. No, I'm just saying 300, you know, the number is sort of like I opened it once. I'm sure I'm in there somewhere. Yeah. No, exactly. So I think that that is really like what time will tell is like if Rufus is sort of a compelling enough consumer experience that advertisers feel like they need to be there. I think that ChatGBT is certainly more of a household name than Rufus. So I think that really OpenAI does have going for them.

16:00The ad tech, though, and sort of the trust of advertisers, you know, Amazon is an incumbent player definitely has on its side. Great.

16:09Anita Ramaswamy:Well, Catherine, I want to thank you for coming on. That is Catherine Perloff, our Amazon reporter, here at The Information. Anthropic has been dealing with a data leak this week that has revealed new features the company is planning for its next big model, Claude Mythos. The information covered the story in our AI Agenda newsletter written by our reporters Rocket True, Stephanie Palazzolo, and Aaron Holmes. I want to bring on Aaron to help us make sense of what is happening. Aaron, welcome back to the show. It's great to have you here. Happy to be here. So help us understand, what was this leak?

16:41So essentially, this was a mistake that Anthropic made. They were not, you know, breached by a hacker or anything like that. But they essentially, you know, accidentally uploaded a file to a public repository that's just meant to help developers understand how to use their product. And this file exposed some of the source code of Cloud Code, which is their popular AI coding agent. and essentially also gave away some of the upcoming features that they had planned, as well as some of the inner workings of how Cloud Code actually works. So do we know who did it? Like this was an employee that made a mistake or what?

17:20I mean, Anthropic said in a statement to us that this was a human error. So it was a mistake that an employee made, but we don't know exactly who or kind of where in the process that happened. Right. Right.

17:33Anita Ramaswamy:And I mean, you said in your newsletter today with Rocket and Stephanie that this actually is not the first mistake, I guess, that Anthropic has made around revealing details about the new model. I mean, there was a blog post that was published accidentally. Yeah. So a couple of weeks ago, or actually like one week ago, they accidentally exposed a blog post that also had some hints about their upcoming model called Mythos. So this is basically the the second kind of embarrassing leak for Anthropic in less than a week's time, both of which, you know, shined a light on some unreleased products they hadn't announced yet.

18:10Anita Ramaswamy:Okay, so let's put the human error aside here, which is funny to think about when you think about humans making errors and also agents making errors. I mean, what have these errors ultimately told us about Mythos? What are the features that they're planning to release? Yeah, so I mean, for one, it sounds like Mythos is potentially an even bigger and more powerful model than Opus, which is the current largest version of Claude. And we also know that they're planning a few features that would potentially, you know, let Mythos be an even more effective AI coding developer. One is this feature called Kyros, which would essentially let Claude just keep working around the clock in the background and send updates to the user that's programming it.

18:56But there's also new features that would allow people to collaborate with other humans and their Claude agents on shared coding projects. And there's also some more fun features, like apparently these animated sort of Tamagotchi-like pet icons that would sit on your desktop and represent Claude as an embodied animated character.

19:22Anita Ramaswamy:What about, so Kairos, tell me a little bit more about that. So this is like an always-on type of agent that runs in the background and can complete tasks without you even having to control it? Yeah, and essentially, you know, this is basically like a tool that would be working around the clock and potentially be more proactive to, you know, according to the description that got leaks to take initiative to explore, act, make progress without waiting for instructions, which I think is essentially the next step in, you know, these AI coding tools getting even more autonomous than they are now. So aside from us learning about the different features that Anthropic is working on, I mean, are there any bigger issues that Anthropic is facing with this leak?

20:08Anita Ramaswamy:Was there any deeper information about their models or about their software that could give competitors an advantage or even expose them to hacks? Yeah, so on the security front, it's probably not a huge issue. I mean, this doesn't really let anyone steal either the weights of their models or hack into their systems. But from an IP perspective, it's not great for Anthropic. This code revealed some of the proprietary techniques that they use to get CloudCode to work as well as possible. These tools are called the harness, but it essentially is the set of instructions that CloudCode uses to work through specific problems.

20:49And on top of that, it also is giving everyone insight into sort of the next round of features that Anthropic wants to publish, which could give competitors an advance notice on trying to catch up with those tools.

21:04Anita Ramaswamy:Okay. And last question for you. I mean, the cybersecurity community, how are they reacting to this? you know i think that the biggest reaction is just that this is a bit of a black guy for anthropic given that you know they have worked hard to build this brand of being very safety focused and um you know they they've always tried to sort of emphasize that they want uh to make sure claude is aligned and and not leaking into the wrong hands and um i think that one you know issue that comes up here is just that uh we know that anthropic has said publicly that a lot of its own code is written by Claude.

21:41And people have questioned whether this was an example of using Claude code to Vibe code their own applications that led to this accidental leak. They said it was a human error, but if a human didn't check Claude's work well enough, that could potentially lead to an issue like this. And that's a bit of a black eye for them.

22:00Anita Ramaswamy:Great. Well, Aaron, I want to thank you for coming on. That is Aaron Holmes, our reporter, here at The Information. This week, the information is featuring investors we included in our next GPs list. These are the people that our venture capital reporter, Julia Hornstein, has dubbed most likely to lead top VC firms across Silicon Valley in the years to come. Today, I'm sitting down with Molly Welch, a partner at Radical Ventures. She has contributed to her firm's effort to invest in companies like World Labs, renowned AI scientist Fei-Fei Li's AI startup focused on the 3D world. I want to bring on Molly to talk to us about her view on the AI sector.

22:38Anita Ramaswamy:Molly, welcome to the show. It's great to have you here. Hi, Kosh. Thank you for having me. I'm thrilled to be here. Well, I'm excited to talk about your focus in the AI sector. I want to start with the news of the week, though. So we've had these data leaks at Anthropic. I mean, what do you make of that? Yeah, I think the first thing I would say, very bad, no good week in security. A number of incidents that have been attracting attention. and it's yikes, right? There are a couple different things going on. I think there's some human error at play, right? But then there's also some of these very real software supply chain issues that are not necessarily, you know, a rogue AI agent has caused these things, but they probably are exacerbated by AI, right?