A New AI Browser & Jobs Friday without Jobs Data 10/3/25

0:00Bring in show music, please. Hi, I'm CNBC producer Katie Kramer. Today on Squawk Pod. Jobs list Friday. With the government shut down and no numbers released, what are we to do? Austin Goolsbee, Chicago Fed president, joins us. It's the first Friday of the month. We're sitting here on the quietest jobs day. Something's wrong with the universe. Where does a data-dependent central bank go from here? What the market expects is not in the Fed's mandate. What should be in the decision of the Fed are the economic conditions on employment and inflation. And perplexity out with a new artificial intelligence browser, Comet.

0:44Founder and CEO Aravind Sreenivas explains the next era of AI assistance. Building something like Comet requires orchestration of several different models. One model that truly reasons, one model that pulls context, one model that synthesizes and processes them, another model for computer use. Those conversations plus Apple removes an app that tracks ICE agents. But what's the difference between that app and Waze, which tracks police? I don't want violence to anybody who can be identified in any of these things. You don't want to get a ticket either. It's Friday, October 3rd, 2025. Squawk Pod begins right now.

1:22Stand Becky by in three. Two, one, cue it please. Good morning, everybody. Welcome to Squawk Box right here on CNBC. We're live from the NASDAQ market site in Times Square. I'm Becky Quick along with Joe Kernan and Andrew Ross Sorkin. And here we go. It's a Jobs Friday without any jobs. I still say it is a Jobs Friday. It is Jobs Friday. It's Jobs Friday. It's Jobs Friday. There's just no report that is coming. So we're going to guess what we're going through this. Because I'm certainly not willing to give up Friday. No. Okay. No. It's like Taco Tuesdays, Jobs Fridays. We'll give those up either.

1:57And we are on day three now of the government shutdown. The Senate expected to vote again today on the same proposals that failed Tuesday and Wednesday. Senate Majority Leader John Thune saying that after today, weekend votes are unlikely. Democrats insisting any funding bill extend health subsidies due to expire at the end of this year. Republicans say the issue should be dealt with separately. Yesterday, President Trump announced that he would meet with Office of Management and Budget Director Russ Vaught about the talks over the agency cuts. And we're going to see where all of that ultimately shakes out.

2:30There's a lot of great things about 2025. Health care. I mean, you know, medical advances. I thought you were going to say health care insurance. No, that's not great. That's not great. Health care advances that you can't afford. It's great. Yeah, but it's not because of subsidies. It's because it's a disaster. Kimberly Strassel today urging Republicans, please, you're either going to have the Democrats are going to have to be to give in and be chastised for what they did. Or Republicans are going to own the calamity that is Obamacare. And that's a reason that that keeps costing more and more and more.

3:06It's not because we're not extending covid era super subsidies. It's because it doesn't work right. It doesn't work right in terms of holding costs down. And we should, Republicans should have had the wherewithal and the stomach to, years ago, to propose something different, because now it's probably too late. Now it's probably too late. Look, there's, we've all talked about the need, and by the way, President Trump talked about this, said he was going to do it the first time around, that he was going to fix health care. Let's try to fix health care. Let's do it. In a good way, though. You can't really, I don't know if you can fix Obamacare.

3:40That's the problem. There's so many things that are implied with it. that are bipartisan that they probably could do. If you find ways to reduce some of the fraud, if you say everybody's got to have a little skin in the game, you've got to do a$10 co-cate to come through with this. But the way it's designed, it almost induces, it almost induces bad things to happen. And there probably should be limitations. There probably should be caps on, I don't know if it's$120 ,000. That seems like a relatively low number to have a cap on it. But if you're at$400 ,000, you probably should not be getting any of these subsidies that come along.

4:09Right. Kimberly Strasson, continuing to shovel money into temporary COVID area bonuses to avoid a cliff is the equivalent, isn't even as good as putting a Band-Aid on a femoral artery bleed. It's actually making the cut bigger. It just makes things going to be worse down the road. Apple removing IceBlock and other apps from its app store that can be used to anonymously report sightings of U.S. immigration and customs enforcement agents. The move coming after pressure on the tech giant from Attorney General Pam Bondi. The FBI said last week that a gunman who attacked a Dallas ICE facility had recently searched apps tracking the presence of ICE agents.

4:52Now, in a statement to NBC News, Apple said, We created the App Store to be a safe and trusted place to discover apps based on information we've received from law enforcement about the safety risks associated with ICE block. We've removed it and similar apps from the App Store. Now, Ice Block had been downloaded more than one million times since it was introduced this year, according to data provided by NBC News, the app tracking firm App Figures. It's an interesting one because we also, I mean, I'm shocked it's been allowed all these years. You know, when you go and people take it for granted now, you go on.

5:28I know what you're going to say. You go on Google Maps or Waze or any of these things. And you know where the police are. And it shows you where the police are. Like Fuzzbuster, the original Fuzzbuster. There's all of these other apps that will show you. You can listen in now if you want to, the police radio scanners and things. I mean, there's lots of apps and access. I think in this case, though, they're worried about violence against the ICE agents. With Fuzzbuster, with Waze, you're basically saying, okay, I'm going to slow down because I know there's a police officer here. But I've thought the same thing with Waze or people - Oh, no, I appreciate that.

5:58All I'm suggesting is until one, knock on wood, or the plastic here, until one day that somebody uses it for a different reason. Right, right. That's true. Just even the idea of using it as a fuzz buster, you know, you're invading enforcement. To commit a crime? What do you mean? To invade your privacy? What do you mean to use it for a different reason? Well, if you know where the police are. I'm suggesting if there is... To commit a crime or something? Or to... I'm saying if there... Let's say there's a terrible, violent criminal who decides they want to kill the police. And they see that the police...

6:31I don't even want to talk about it on television. I'm just saying the question about all of these apps that we live with today that allow you to find all sorts of types of people. And that's the thing. And this is the thing. We've either decided that we should have an app universe or a technology universe where we can track lots of different types of people as long as they're public. I was going to say by the way, they're tracking us. And they're tracking us. As long as it's public, everybody can track everybody. or you really should decide that you really shouldn't do this. And this is a bad policy across the board.

7:06The idea that we are specifically look, I don't want any any violence to to folks in the ice, you know, people working for ice, but I don't want violence to anybody who could be identified in any of these things. You don't want to get a ticket either. Right. Sure. But I think that in the tradeoffs, I get it. Hey, there's nothing that's private. you know phone numbers addresses and you know that's what i'm saying so i'm actually surprised maybe i shouldn't be surprised that apple's taking it down probably i imagine given whatever government pressure you know from pam pam bondi but the idea that and that's another interesting idea the idea that one group of police should be um you know guys yeah uh shouldn't be a group of law enforcement one group of law enforcement shouldn't be um you know identified on one of these services, but everybody else should.

7:57There's a lot of interesting things. Some people carry cameras, some people don't carry cameras. There's lots of things going on. California Governor Gavin Newsom reacting to the Trump administration's 10-point plan for universities that want preferential access to federal funds. Newsom threatening to pull billions of dollars in state aid to any California university that signs on to the proposal. The Trump administration plan includes demands to cap international undergrad enrollments, ban the use of race or sex in hiring and admissions, and a five-year tuition freeze. Newsom says his state will not bankroll schools that sell out their students, professors, researchers, and surrender academic freedom.

8:38The University of Southern California was among the nine schools receiving letters from the White House, and this is going to create a real battle between the state and the federal government insofar as these universities rely, in large part, subsidies from both yeah because I was just trying to look up who my guess would be more from the state but it's more from the state so the truth is that he has more leverage to some degree but if you look at the budgets of these universities all of which would argue that they need more money not less or it creates a real real conundrum funding comes from the state of California for the University of California but that's see yeah yeah for University of California.

9:16Oh, USC.

9:20Cheese will be next. Coming up, perplexity. The$20 billion artificial intelligence startup is out with a new browser. Founder and CEO Aravind Sreenivas calls it more of an AI personal assistant than anything else. Most of the AIs are working on the foreground. You ask it something and it's responding synchronously to you on that chat window. So the future is AIs that are just doing work for you, even as you sleep without you even asking for it. SquawkPod will be right back. The future of AI.

9:55This is SquawkPod. Up in Andrew, Q. You are watching SquawkBox on a Friday morning. Normally Jobs Friday. We don't have the jobs numbers for you, but we do have a lot of other news for you and some amazing guests. I'm Andrew Ross-Sorkin, along with Joe Kernan and Becky Quick. This is an exciting one. Perplexity rolling out its AI browser. Comet to all users this week and joining us right now in an exclusive interview is Perplexity's founder and CEO, Arvind Trinvig. It's great to see you. Congratulations on doing this. I'm sort of fascinated by what this really means. For those who don't know what Comet is, who haven't used it yet.

10:31I, by the way, have been playing with it a little bit now, trying to integrate it even into my calendar. We can discuss that in a second. What is happening here? So Comet's not just a browser. It's meant to be a true personal assistant. And when you hire an assistant or anyone to join your team to help you do your tasks, one of the things you just take it for granted is their ability to use the internet, their ability to use search, web browsing, and all the tools that exist out there on the open web. That's not something you can take for granted with an AI because the model is obviously giving it reasoning capabilities, but AI is clearly going to only deliver value as an end-to-end orchestrated system that knows how to use the browser, how to upload files, how to download them, how to pull context from relevant tools and data that's only relevant to you and nobody else.

11:23So that's where AI is clearly headed next. It's not just about answering questions or being a secretive chatbot that tells you good stuff you want to hear. It's truly about delivering value and you being able to delegate tasks to it. So let's talk about some of those tasks, because obviously the biggest bottleneck, I think, in all of our lives at the moment, at least in mine, is dealing with both email and my calendar. And frankly, I wish this could also do with text messages and Slack messages and WhatsApp and the various different messaging systems that are taking place. But coordinating all of that, you are effectively now setting it up so you can connect into these services and schedule stuff via AI.

12:03have emails written for you. How good do you think that that technology is right now? And what are the risks at the same time? I think the technology is getting really good. In addition to Comet, we have a product called the Email Assistant. The main paradigm shift there is it's going to be working in the background. Most of the AIs are working in the foreground. You ask it something and it's responding synchronously to you on that chat window. The future is AIs that are just doing work for you, even as you sleep without you even asking for it. It's auto-drafting your responses to emails that were sent while you were not even active on your device.

12:43It's scheduling your meetings on your behalf and moving things around even without you asking for it. And it's prioritizing your emails that are sent to you. That's the future. We're moving more towards asynchronous multitask abilities. Right. And that's why we're building something like comment and background assistance there, too. In terms of challenges, it's mainly the ability to pull context reliably, which is, you know, the thing that's solved with all the progress and like frontier reasoning models. And security-wise, we are very clear that like none of the data is violating the SOC2 compliance.

13:18So as a company, we are SOC2 Type 2 compliant and we just try to ensure that all the data is securely stored and logged. Well, so that's what I was going to ask in terms of where the email ultimately gets stored and logged, where the calendar gets stored and logged, because for it to work, I would think the best, you would actually want to ingest all of the email. This is why I've always thought that someone like Google should have an advantage just about over everybody, because they have the email sitting on their servers right now to the extent that they want to allow their AIs to read them.

13:51How should we think about that? You don't need all the emails to be stored in an index. In fact, the magic of all these tools is they can pull whatever context is necessary on demand and ignore everything else. For example, if you're asking, prepare me for my day, I just need to pull the calendar events for that particular day and the emails that have been unread on top of your inbox for the last day or two, but not your entire inbox history. And so that's just like a tool call to all these different tools. It's not necessarily like downloading them and indexing them. But don't I want it? Look, here's the thing.

14:31And this is the complicated part. I would imagine, ultimately, if you're going to rearrange my calendar for me on my behalf, you would want to know who was important to me, not based on the last 24 or 48 hours of emails, but maybe on the calendar and being able to look back even two or three years or maybe be able to look into emails from seven years ago to get the context on what's going on with this person. How is this person important in my life or not? Am I going to prioritize this person over the meeting that's currently scheduled right now? I mean, that's always been the hardest part about this dream that we all have that AI will one day schedule us for us.

15:10Yeah. I mean, that's essentially what Comet tries to solve because it's having a higher bandwidth access to understanding you through not just your emails, but everything else you do on the computer. You're browsing history. You're basically onboarding Comet with your browsing history, but all the history stays locally on your device, on your client. It doesn't actually get sent to the servers. But any time you ask Comet to do something, it tries to understand you by pulling your browser history. So if something is important to you, it is usually revealed in other things you do as well, not just stuff that's on your emails.

15:44You know, one of the things that's so interesting about the way Perplexity works is you are leveraging so many different models. So which models are you using for this component part of the email, calendar, Comet program at any one time? Or is it multiple models? It's, as always, multiple models. In fact, building something like Comet requires orchestration of several different models. One model that truly reasons, one model that pulls context, one model that synthesizes and processes them. Another model for computer use capabilities to actually go and control websites or tools. And we see the future as something where people don't...

16:26I mean, we've been saying this even in the era of just AIs answering questions, where no one even cares what models are being used for what purposes, as long as this end-to-end system does what the user really wants. I think it's even more true in the agentic world, where the workflows are even more complicated. As the frontier models get better and better, you know, each time OpenAI releases a new model, sometimes the prompting shifts in terms of how you get the answer, what kind of answers you can get, what you need to do to sort of get things to work the way they're supposed to. I don't know how quickly you think each of these frontier models is going to get upgraded and you then have to implement them into your own service, but what kind of challenge does that unto itself pose?

17:11We definitely had a hard time adopting GPT-5, for example. We actually expected it to be a massive jump in capabilities, but for some reason, all the prompts that we had written for 4.1 never worked well for 5, and we spent several weeks trying to write a better prompt for 5, but we gave up, and we just continued to use 4.1 for a majority of our workloads. They have since then improved some of their models. GPT-5 thinking is actually a pretty decent model, So definitely there are going to be challenges in adopting new models. For example, Anthropic tries to make sure there's good backward compatibility with the prompts they've written for their previous generation of models.

17:49There's also the challenge of writing one prompt that can work across different model companies. For example, we've seen prompts that we write for Claude and GPT's work consistently well across both, but it doesn't actually work when we use models like Gemini. I'm curious just to take a 30 ,000 foot step back for just a second. There's a lot of questions right now in AI about whether we're in this sort of gold rush or whether we're in some kind of sugar rush. I think a lot of people even looked at the investment that NVIDIA made in open AI, both as a good investment on one end and another. other people said, well, maybe this is a little bit like vendor financing in the late 90s, where a lot of these companies ultimately don't have the cash to actually buy all the chips and do all the things that they've committed to do.

18:37What do you think? Well, it's all about whether you can truly deliver value. If something like Comet can help you get your work done, like you said, you're trying to use it for emails. If instead of hiring one more person on your team, you could just use Comet to supplement all the work that you're doing. And that's worth like$10 ,000 a year. And if you scale it with a number of people who are like you, that's clearly adding value to the GDP. It's letting you do more work at the same unit of time. And I think the human value, conservatively estimated human digital knowledge work value contributions to GDP is around$25 trillion.

19:17dollars. So we just need to do at least 20 % more productivity for that. And that would automatically take care of all the capex that NVIDIA wants OpenAI to spend, which is$4 trillion for 10 gigawatts, right? So we just need to do$5 trillion, like bare minimum, like table stakes. And I think we can do that. That's extraordinary. And finally, a question about copyright, because there's a brouhaha in Hollywood right now about what OpenAI has announced in terms of an opt-out kind of program. There are questions about how publishers are going to join your Comet Pro subscription. I want to understand how that works.

19:53How do you think long term folks who have copyrights are going to be paid or not? Yeah, so that's, I'm pretty excited about Comet Plus. It's a program that we have put a lot of thought into as something like Apple News Plus that works in a world where both humans and AIs are going to consume content. As a company, we're not really interested in regurgitating existing content like we're not training models on the tokens but we want that content to surface in our answers we want that content to be available to our agents that are going and doing deep research on behalf of the user so not every bot is a faceless crawler bot like some bots are actually just doing work on behalf of the human just like how you would have delegated an intern to do something for you uh and that's why there needs to be one subscription plan This is like how you buy one subscription plan for your company that your team also uses.

20:42It's similar to that where you and your personal AI can use the same thing. And we have Washington Post. We have CNN. We have Conde Nast. Several fortunes. All of these joined. And we hope to expand the program to more people. Okay. Arvind, love to see you hopefully in person at some point. Comet's fascinating. I do worry about at some point we have to have a longer conversation about whether it can go off and do things without you knowing it, which is its own question. But we'll continue this conversation soon. Thanks. Thank you so much for having me, Andrew. Thank you. Next on Squawk Pod, Austin Goolsbee, head of the Chicago Federal Reserve.

21:19What a jobs list Friday means for the central bank. Employment data is half of what they take into account when deciding interest rate policy. What you just described sounds like you're driving in the fog in the car and not really clear about what's ahead. I do think that if you got dust in the air, if you got fog, a lot of times you might slow down.

21:44You're listening to Squawk Pod from CNBC. Here's Becky Quick. All right, it's 830. No jobs report today, but we have something better. We want to bring in our Steve Leisman and Chicago Fed President Austin Goolsbee. Steve. I don't know about me being better, but Austin certainly has the potential here. Austin, thanks for joining us. Austin Gillespie, Chicago Fed President. Austin, it's 833-38-39, and we don't have the jobs report. I want to know how confident you feel right now that you know what's going on in the U.S. economy. Well, I will say two things. One, it's the first Friday of the month.

22:24We're sitting here on the quietest jobs day. Something's wrong with the universe. We didn't get the final countdown walk-on music. And so the Chicago Fed, not because we knew there was going to be a government shutdown, we didn't, but because we've been trying to get better real-time measures of what's happening in the job market, we've recently rolled out the Chicago Fed labor market indicators. And one of those indicators, we turned to 11 different data sources. We take the official data plus high quality private sector sources. And one of the numbers that we keep is a real time estimate of what will be the unemployment rate.

23:07So barring getting an actual unemployment rate, we released yesterday that the unemployment rate would have been announced to be 4.3 percent. That is unchanged and stay where it is. Austin, I looked at that data and your methodology there, and it looks a little bit like you're California sober on that in the sense that you have not really forsworn the official data. A lot of it's based on the official data. So it seems like we're kind of cool in knowing what today might have been. But we don't know about, let's say, next week when we don't get jobless claims, although we are getting some state level stuff.

23:47Goldman Sachs came out with a report this morning estimating that in the absence of two states, they think it would have been 224 ,000. But isn't it right as time goes by, we lose contact with what we think is the reality? As time goes by, it will get worse. There's no question the best jobs data in the world come from the Bureau of Labor Statistics. We need that data. We want that data. We can use what we have, which are private sector sources. we can look at the state unemployment claims and plug those into the formula over time exactly as you say steve it's going to get farther and farther from the truth even the private sector sources rely on the bls data to kind of normalize and make their make their samples representative but we fight with the army we have at moments like this where it's critically important that we're figuring out whether the economy is in a moment of transition.

24:46This is what we have. And thus far, it still continues to point to a pretty stable labor market, in my view. So does that keep you on the trajectory you were on? I'm trying to think, and I don't love this metaphor because I hate, I don't even watch anything about planes and problems. But you know, you're at 36 ,000 feet, you're cruising at 562 miles an hour with a nice, with some kind of tailwind, whatever you want to call it, and one of your screens goes out. Do you keep on the same trajectory, which right now the market believes is to cut interest rates? Well, look, you know that Paul Volcker used to tell me our job that the Fed is to act and the market's job is to react and let's not get the order mixed up.