Why Roomba Died + Tech Predictions for 2026 + A Hard Forkin’ Xmas Song

Casey, what’s going on?

Well, Kevin, I want to get ahead of something before you and all of our listeners read this in the tabloids. I did have to get rescued off a ride at Disneyland—

Come on.

—this past weekend. Yeah.

Really?

Yeah. We were over at Tiana’s Bayou Adventure, the recently rebranded Splash Mountain, and we were just about to go up the big hill and come down for the big splash when the ride broke down. After 20 or 30 minutes, some very nice people came and walked us out of the Bayou Adventure through the backstage area. I won’t say that it was scary, but there is something embarrassing about having to be pulled off of a log flume.

Wait, you’re not kidding? This is an actual thing that happened to you?

This is a real thing that happened to me, and this is part of the story of technology: You can suspend your disbelief and think you’re floating through a magical bayou adventure, but in reality, Kevin, the tech could break at any second, and then you’re going to have to get pulled off by a man with a hook in his arm.

Yeah, it sounds like instead of going on the Bayou Adventure, you almost went on the Die You Adventure.

Yes.

That’s—

That is exactly right.

I’m Kevin Roose, a tech columnist at The New York Times.

I’m Casey Newton from Platformer.

And this is Hard Fork.

This week, “Buy Robot?” Roomba co-creator Colin Angle joins us to talk about why the iconic vacuum company went bankrupt. Then, our 2026 tech predictions. And finally, we’re going caroling. “A Hard Forking Christmas” makes its annual return.

Gloria.

Get my earplugs.

1. Roomba Enters Bankruptcy

Well, Kevin, as we kick off the show this week, a legendary American company is entering an exciting new chapter. Unfortunately for them, it’s Chapter 11.

That’s the bankruptcy chapter?

That’s correct. On Sunday, iRobot, maker of the Roomba, filed for Chapter 11 bankruptcy after finding itself outcompeted by a bunch of Chinese fast-following companies.

Yes. So this is one of the longest-running robotics companies in the world. iRobot was started in 1990. The first Roomba came out several years after that, and it has been synonymous with the whole category of household robot vacuums. But as you mentioned, they have been running into some issues in recent years.

Yeah, but you could say they’ve been bumping into walls.

Yes. And in 2022, they had their biggest obstacle, their biggest wall, which is that it looked like iRobot was going to get acquired by Amazon. Amazon made an offer to buy the company for $1.7 billion, but antitrust regulators in the U.S. and the E.U. stepped in and said, “We don’t like this, and we’re going to attempt to block it.” That made the deal fall apart, and in 2024, they officially walked away. Now the company is bankrupt.

They kind of zoomed away, you know?

Yeah, kind of turned around—

Mm-hmm.

—and scooted away. We should say, I think this is a fascinating story, not just about a company that made an iconic product and then fell on hard times, but about the complicated balancing acts and trade-offs involved in regulating the American tech sector.

Yeah. And one of those trade-offs, Kevin, is that now iRobot will be owned by a Chinese company named Piscea Robotics, which is its largest creditor and a competing design manufacturer. There are now those who say, “Hey, this is actually a much worse outcome for us,” because there is no longer an American company making a very popular household robot. Instead, this is one more category of manufacturing that we seem to be ceding to China. So all of that gives us a lot of good reason to talk to the founder and former CEO of iRobot, Colin Angle.

Yeah. Colin co-founded iRobot in 1990 after working at MIT’s AI laboratory and eventually went on to serve as CEO of the company for 33 years. He stepped down in 2024 after the attempted Amazon acquisition fell through, and today he is making the case that iRobot was a casualty—or roadkill, as he put it—the direct result of misguided antitrust enforcement by the U.S. government and a tragedy for American innovation.

Yes, but as you will hear, he is also willing to admit that he made some mistakes along the way, and that gives us a rare opportunity, Kevin. Usually when CEOs come on this show, they want to make the case that they are one of history’s greatest geniuses and that they are about to change the world for the better for all of us. It is much rarer for a CEO to come on and say, “Hey, look, here are some of the hardest calls I had to make, and here are some of the ones I got wrong.”

Yeah. So in order to hear his angle, let’s bring in Colin Angle. Colin Angle, welcome to Hard Fork.

Great to be here.

2. iRobot Loses Its Edge

So we’ve invited you on today to talk about iRobot, the company that you started and ran for many years, the maker of the iconic Roomba—

—a robot vacuum. To start, I’d like to kick things off with just the most basic question possible: Why did iRobot go bankrupt?

You know, it’s a complicated question. Simple to ask, hard to answer. We led the world. We invented the category of consumer robotics with the launch of the Roomba in 2002, and for 17 or 18 years, we really grew the category. Then, as it got more competitive and the world changed, we had to change with it. Ultimately, bankruptcy is something that happens when you need some external help to take the next step in the chapter of development.

Liam McCabe at The Wirecutter, who has reviewed house-cleaning robots for a long time, argued recently that iRobot hit its peak in 2018 with the release of the Roomba i7 Plus, which could clean specific rooms on command and automatically dump the contents of its dustbin into its own dock. Shortly after that, Chinese competitors began to pull ahead, including Roborock, which I think became iRobot’s biggest Chinese competitor and entered the U.S. market in 2018. What were the core problems, in your view, that started to erode that market leadership?

It was certainly the advent of this new type of competitor: the Chinese fast follower, who had access to the Chinese marketplace, which iRobot effectively did not for various pragmatic and political reasons. That gave the competition a protected market to cut their teeth on. I also think that the marketplace was not a level playing field, as I described.

What do you mean by not a level playing field? Do you mean the ability to sell in China specifically?

Correct.

Or is there something else?

Yeah. For a small period of time, iRobot was the leading manufacturer of vacuuming robots in China. Then it stopped because China decided that this was a market of interest and that they were going to ensure that Chinese companies were advantaged to succeed there. The Chinese government provided direct subsidies to our competitors, and we had to overcome that as well. In an emerging marketplace, it’s a cage match, and it certainly got hard. It got increasingly competitive.

Companies need to adapt. Ultimately, we were open to other strategies to enable us to continue to innovate, which is why, when Amazon came and said, “Hey, we think we have a vision that is compatible with your vision,” it made a lot of sense, and we were excited.

We want to talk about the acquisition, but I have one more product question first.

Sure.

I want to say, it’s really hard to invent a whole category and to be a market leader for 18 years. As you say, competition is brutal, and I do appreciate you sitting through our questions about some of the difficult decisions that you had to make along the way, some of which didn’t turn out right. This is just something that happens to CEOs.

But I think there’s one question that comes up over and over again when you read the Roomba obits. It is that Roomba relied on cameras to get around and not LiDAR, this sort of next-generation technology. It seems like the Chinese competitors really embraced LiDAR, and consumers preferred that over time. So while you were the leader, you guys didn’t embrace LiDAR, and I’m curious what role you think that played in all this.

It’s—I’m entertained by the question because it’s wrong. It was wrong then, and it’s wrong now. Does your autonomous-driving vehicle have LiDAR on it? No. Why?

Right.

Because it is a dead-end technology that gives you no situational awareness. And so the—

Wait, but Waymos have LiDAR on them.

This is sort of the classic Waymo-Tesla—

Divide.

Sort of the divide—

Okay.

Sort of shrunk down to the household robot.

Well, sure. Then call it the Roborock-iRobot divide. Under my leadership, I was playing the Tesla card—which said, “I will put every penny of COGS in the Roomba against better visual understanding”—because I believed at the time, and this was very publicly stated, that Roomba with a camera could understand what it was doing, could understand whether it was doing a good job or a bad job, and could play a role as the hub of the smart home in a way that a robot driven by LiDAR could not.

The maps that you make with a camera are more rugged and robust to moving furniture. We explicitly decided not to put LiDAR on Roomba because it took away from the COGS that I could put against CPU and other technologies. This was not a, “Why did you fail to innovate?” This was a strategy.

Right. I, I guess I'm not contending that most consumers care whether their robot vacuum uses LIDAR or cameras. I don't think most people know. I don't think most people care. I certainly had bought Roombas, uh, you know, for my first apartments and homes before I knew what what technology was running on them. I just care, does this thing clean my floors or not?

Right.

Um, and I think— You know, over time, there started to feel like a gap between Roombas and some of these Chinese competitors with other features, things that were unrelated to navigation. Things like, does this thing have a wet mop as well as a vacuum in it? Can there sort of be these other features that can maybe make this thing a little more useful or require a little less oversight? So are there other features that the Chinese competitors and other competitors introduced into their robots that you wish the Roomba had adopted earlier?

Honestly, we got wet mopping wrong. We built our first mopping robot 2 years after Roomba, and we felt like creating an independent mopping solution was the right strategy. It gave the best customer benefit; it did the best job.

As in having a separate robot that mops the floors from the one that vacuums the floors.

Yeah, we called it Scooba. We built it, and it worked great, but the customer didn’t go for it. It was never going to work as an independent product, but it was going to work as an incremental feature, and we reacted late. We took some damage and had to do some rugged hustling to catch up.

I deserve arrows for, for the, for mopping. I believe that the LiDAR people are still wrong, and I just wish we had moved faster and had a longer runway before that, because that led to a great place.

3. The Amazon Deal Collapses

I feel like we missed the middle of the story, which is the attempted sale to Amazon. You mentioned it earlier, but I want to revisit that. In August 2022, Amazon offers to buy iRobot for $1.7 billion. It ultimately falls apart over antitrust concerns.

I think some people look at that and think, “Well, that seems fine because Amazon’s already plenty big, and I don’t want it to own every vertical of e-commerce in existence.” I imagine that you may feel quite differently. Make your case that this was a bad outcome for America, that iRobot was not able to become a part of Amazon.

Well, that’s pretty easy. We got to a point where, for very good and understandable reasons, the market pressures and the competition led to an economic model where iRobot was finding itself pressured to innovate.

iRobot was a proxy for the United States because we were the only public company doing consumer robotics. If you were a fan of having consumer robotics be an industry where America could claim leadership, you needed iRobot to continue to be able to invest like the big dogs and create innovations that were going to give consumers the choice they were asking for.

The challenge was that the acquisition seemed to become all about Amazon and not at all about consumer preference, leadership in the consumer robot industry, and doing what the European Commission and the FTC are charged with doing, which is defending the worst abuses of monopoly.

Mm.

At the time, iRobot’s market share in Europe was 12% and declining. The market leader in Europe was a company that had been in the marketplace for 3 years. That’s kind of the definition of a vibrant, dynamic marketplace. How do you even create an argument that says there are monopolistic forces, antitrust forces, at play here?

What was Roomba’s market share in the U.S. at the time? I’m just curious.

It was hanging around 50%.

Okay, so not an overwhelming monopoly, but a dominant player in the market.

No, and the market share, unfortunately for us at the time, was going down while there were multiple healthy competitors that were seeing their market share go up. Again, it’s very difficult to say, “Okay, this is a monopolistic position.”

In your view, was the failure of the antitrust regulators that they misjudged what kind of acquisition this was—that this was more of a lifeline or something strategic to let the company keep operating from a place of weakness, rather than Amazon trying to gobble up a company on the rise that was going to enable it to extend its dominance into another area?

It would be impossible for me to believe that anything was missed. iRobot spent tremendous amounts of its own resources, and Amazon spent many, many times what iRobot spent over the course of 18 months doing our best to make incredibly clear the facts and circumstances regarding the acquisition.

There was nothing unsaid, and there was no, “Gee, golly, I wish we’d asked that question,” but we did. So why did it get blocked? It wasn’t ignorance.

One concern that came up a lot at the time was that iRobot famously used these cameras, and there were worries that Amazon was just going to have detailed views inside the homes of many, many Americans. This could be used for targeted advertising or just other uses that people might find creepy.

Of course, now we’re in a world where iRobot is going to be owned by a Chinese company, and so all those same images are going to be in China. What was the conversation around that at the time with those regulators, and did you float the possibility that, if it wasn’t Amazon that had access to this data, it might be something that you liked even less?

I think that we continue to have, to my knowledge, incredibly rigorous privacy and security. In fact, images never leave the Roomba. We defend it in depth, meaning that all of that use of cameras was done on the edge, on the robots.

No image ever left the robot unless you gave explicit permission for the exact image that we showed you we were going to share—all opt-in. iRobot adheres to the highest level of cybersecurity. We were the first consumer product company ever to meet the requirements of the gold standard in data security.

So if it wasn’t the privacy concerns that you think led the FTC to block this acquisition, and if you don’t think it was ignorance on their part, what was it? Why do you think the government stepped in to prevent this merger from happening?

I have no logical answer to that question. I do not believe there is any economic rationale that would say that the FTC or the European Commission had any evidence that the consumer was going to be negatively harmed by the acquisition of iRobot by Amazon. I believe that passionately. The consumer only lost.

If that’s not the reason, why did the Federal Trade Commission, with Commissioner Lina Khan outspoken in her opposition to big tech, choose to take a position contrary to the formal charge of the FTC to protect the interest of the consumer? Answer it with your own gut.

What is your own gut?

I believe iRobot had nothing to do with the blocking of the deal with the FTC.

That it was essentially just that Lina Khan didn’t want Amazon to get bigger, and that was the end of the conversation.

I believe iRobot was—this is Colin’s personal view—

Yeah.

We were roadkill.

Yeah. Yeah. Well, as we’re running out of time here, let’s talk about what you think the implications of this are for future efforts to get robotics companies off the ground in the United States. We’ve had some robotics companies on the show over the past several months. What does your experience tell you about their likely path forward? What concerns do you have about the future of American manufacturing?

The tragedy of the blocking of the transaction is that we did it to ourselves. The net result, which I have argued was done with eyes wide open, was putting the consumer robot industry in a box, gift-wrapping it, and handing it to someone else.

Markets are difficult to create. They’re fragile. They need to be nurtured. They need to be cared for. There are many, many people who are working incredibly hard to take any cent of an exciting new market away from the inventor.

We, as a nation, need to decide what we want to have happen. Here we're sitting around with physical AI—humanoids. It's at an exciting point. Billions of dollars have been invested in leadership and in creating American companies that are doing tremendously exciting research, and the ramifications are worth trillions of dollars of value creation in the future, if what has been written and invested in is to be believed. So, are we going to protect and catalyze American companies to continue to lead? We need to reestablish confidence that the FTC's role is to prevent the abuses while catalyzing American competitiveness.

Hmm.

I have real concern that this may not be the commonly held view of the role of the FTC.

Hmm.

If nothing else, let the tragedy of Amazon's attempted acquisition of iRobot serve as a lesson as we think about an industry which honestly could be 1,000 times larger than robots vacuuming. Now, let's not screw it up. Let's treat this as the golden goose that could help the U.S. economy drive economic leadership for decades to come, because it is in our hands.

Hmm. Hmm. Well, Colin Angle, thanks so much for joining us.

Great to be here, and thank you for the conversation.

When we come back, I predict that I crushed Kevin at last year's tech predictions. Let's talk about it.

4. 2025 Predictions Get Graded

Well, Casey, we're coming down to the end of 2025, and as is our custom on this show, it is time to stare into our crystal balls and look at our predictions for the coming year.

Let's look at those balls.

But before we talk about our predictions, we should make our disclosures. The New York Times Company is suing OpenAI and Microsoft over copyright violations related to the training of large language models.

And my boyfriend works at Anthropic.

So, Casey, remind me what your predictions were last year.

Well, Kevin, as you know, each year we make 3 predictions at different levels of confidence. Each year we say something that we are highly confident is going to happen, something that we have medium confidence will happen, and something that we have low confidence will happen.

Right.

Let's dive into them.

Yeah.

Last year I said, with high confidence, that I thought the AI culture war was going to begin in 2025.

Yes, and you really nailed this one.

Thank you. I had basically predicted that we were going to see congressional hearings about the way that ChatGPT responds to certain questions, and outrage about the way that LLMs respond to politically sensitive topics. If you go back through the past 12 months, you can see that in March, Representative Jim Jordan subpoenaed 16 tech companies in an effort to discover whether the Biden administration had pressured them to, quote, “Censor lawful speech in their AI products.” Then, in July, Missouri's attorney general announced a new pressure campaign against many of those same companies by complaining that, when it came to President Trump, chatbot makers weren't being nice enough to him and were ranking him low on various lists that the attorney general thought he should be ranked highly on.

This was exactly the sort of thing that I thought was going to happen. Honestly, it seemed obvious to me, and it did in fact happen.

Yes. And we also saw the Woke AI executive order in July, which put some shiny government gloss on this requirement that chatbots remain ideologically neutral if they're going to be sold to the government. So, yeah, I think this was a good prediction, and you made it with high confidence, and it came true.

Yeah.

Look at you.

Look at me. What was your high-confidence prediction?

My high-confidence prediction for 2025 was that a newly released crypto meme coin would briefly reach a market cap of $100 billion before crashing. The closest we came was the Trump meme coin, which peaked in value in mid-January at a market cap of around $15 billion and then crashed. As of this recording, its market cap is just over $1 billion. So it did not reach the high point that I thought maybe it would, but I think, in spirit, I'm going to chalk this one up to a win.

Yeah, I don't think it's a win. You got 15% of the way to it being true and failed. That's an F, my friend.

All right.

Let's talk about our medium-confidence predictions. Mine was that in 2025, Waymo would go mainstream, and I think I got this one right. I thought maybe there might be an SNL sketch about it. That did not happen, but Waymo is now about to start operating in New York City, which I do think will be the gateway to it becoming more mainstream.

There was also a question about Waymo on Jeopardy, and it just announced that it was going to expand into a lot more cities, including Miami and Washington, D.C. It currently operates in the Bay Area, L.A., Phoenix, Austin, and Atlanta. Next year it's going to Nashville and London. So I do really think that Waymo went mainstream this year, Kevin.

Wait, if I get a fail for my prediction about a crypto meme coin, you do not get a passing grade for predicting that Waymo would go mainstream.

How? Why not?

There was no SNL sketch. That was your resolution criteria. I would say, for the most part, Waymo is still mostly a California phenomenon. Sure, it's operating in these other cities, but it's been slower to roll out than expected across the country. There's been some resistance to it in some of the East Coast cities where it has announced plans to launch.

So, yes, I agree it has been a big year for Waymo, but I still think there is more to come as far as the public's embrace and understanding of the driverless car.

Of course there is more to come, but I think Waymo has become a household word in 2025.

Okay, I'll let you have this one because I am a fair and generous co-host.

Thank you. What was your medium-confidence prediction?

My medium-confidence prediction for 2025 was that Apple would acquire Snap.

Hmm.

And, Casey, I did not get this one right.

You know what? You have my sympathy here, because I thought that was a pretty good prediction. Snap has been languishing for well over a year now. Snapchat is a very widely used product, but the business they built around it has just never been very good, and I think there was at least some thought that Apple might pick them up and do something with its hardware, which I think they could have really turbocharged.

But they had other plans, which was to do nothing with Siri.

Yes.

Yeah.

Yes, they were busy not building AI features. So this one will have to wait, if it happens at all, until next year.

That brings us to our final predictions for last year, our low-confidence predictions. My low-confidence prediction was that X, the former Twitter, would be merged into xAI.

And, Casey, did that happen?

Believe it or not, it did. It happened in March in an all-stock deal that valued the social media site at $33 billion and xAI at $80 billion. Both of those numbers seem ludicrous to me, but what can you do? It's a private company.

So, if you listened to our predictions episode last year and then went on to Polymarket or one of these other prediction market sites and placed a large bet on the Casey Newton low-confidence prediction, you made some money.

I mean, I think arguably I did go 3 for 3, and I'm told that possibly our producers have made a sound effect.

Wow, look at that.

Thank you. That was very satisfying. This is a double-edged sword because sometimes, if you go 3 for 3, it means that you weren't going very far out on a limb. But I don't know. I don't know if a lot of other people were betting that X and xAI were going to merge by March of this year, so, yeah.

Yeah. Well, I give you a 2.5 out of 3.0, which is still pretty good.

Fair enough. All right, what was your low-confidence prediction?

My low-confidence prediction for 2025 was that at some point during the year, OpenAI would officially declare that they had achieved AGI. And, Casey, that did not happen either.

No, it didn’t.

That was another miss for me, and I’m told that our producers have also made a special sound effect for me. Okay.

Sometimes when you swing for the fences, you do strike out.

Yes. Yes, this was my low-confidence prediction for a reason. I did not actually believe it was highly probable that OpenAI would declare AGI. I did think that it was possible because, among other reasons, they had this sort of contract with Microsoft that, in my view, incentivized them to jump the gun or declare AGI early in order to get out from under their commitment to share their IP with Microsoft.

As it turned out, they found another way to get around that by just completely restructuring the company and redoing some of their language in that Microsoft deal. So they don’t have quite the incentive to declare AGI early that they might have a year ago, but they did not do it by just getting onstage and having Sam Altman declare that the machine god had arrived.

They didn’t. Now, interestingly, Kevin, just within the past few days, Dean Ball, a recent Hard Fork guest and former AI adviser to the Trump administration, said that he believed that Claude Opus 4.5 met OpenAI’s definition of AGI—

Yes.

—which I also think is a bit out on a limb there. But it does speak to the fact that, if nothing else, maybe we made some progress toward AGI this year.

Yeah. I mean, I think if I’ve changed my view on AGI at all in the past year, it’s that it will not be a single point in time. There will not be one day when everyone agrees, “Oh, it’s here.” I think it will be a steady progression of things that feel like AGI for some people and not for others.

And that makes me sad because I was planning to take AGI Day off. I thought, “If we get there, I’m going to the beach.” But now I guess I can’t do that anymore.

Nope. All right, so those were our predictions for 2025. Casey, what is your high-confidence prediction for 2026?

Oh, man. Well, you know, Kevin, after going three for three this year, I’ve never felt more pressure on me than I do right now on the Hard Fork podcast. But I think I have one thing that I’m going to feel confident about with high confidence in 2026, and it’s this: I think that 16-plus becomes the new norm for social media accounts worldwide.

Hmm.

Recently on the show, we talked about the fact that Australia had just said that you must be 16 to have an account on 10 different apps, including YouTube, Instagram, TikTok, Snapchat, and more. I think you’re just going to see more and more countries do this. Malaysia and Norway are looking into it, as we discussed last week, and I would bet that by the end of 2026, we’ve seen at least 5 other democracies introduce similar rules.

This thing is just sweeping the nation, and I think it makes a lot of sense. We’re going to get a really interesting experiment into what happens once you kick kids off these big mainstream apps.

I like that. What would the resolution criteria for this be? Would it be 5 countries—

Yeah.

—10 countries?

Let’s say 5 more countries enact under-16 bans of roughly the same shape as the Australia ban.

Wow, okay. So that will go into our predictions for next year. We are, as always, going to put these up on Manifold, the prediction market site where you can bet play money on whether they’ll happen or not—basically just have some fun figuring out whether you agree with us or not.

And kids, this prediction market is not in Roblox yet, but stay tuned.

Yes. Okay, I like that prediction. I agree that it is likelier to happen than not.

Kevin, give me a high-confidence prediction for 2026.

My high-confidence prediction for 2026 is that an AI company will solve one of the Millennium Prize Problems.

Hmm. Tell us about the Millennium Prize Problems.

The Millennium Prize Problems are a set of difficult and important open problems in mathematics. This so-called Millennium Prize was started in the year 2000 by the Clay Mathematics Institute, and they basically said, “Here are 7 of the hardest problems that we don’t have solutions for. If you solve one of them, you get a $1 million prize.”

Hey, can you give me a little flavor of this? Because I did get a B in high school trigonometry.

Yeah, so you’re probably not a candidate for solving one of the—

How dare you?

—the Millennium Prize Problems.

I have ChatGPT access.

But these are things like the Riemann hypothesis, if you’ve heard of that.

Mm-hmm.

The Navier-Stokes equations that describe the motion of fluids.

Mm-hmm.

The Yang-Mills and Mas-

Wait, I can describe the motion of fluids. Dramatic, beautiful, inspiring.

The Birch and Swinnerton-Dyer conjecture.

Mm. Have you solved that one?

No.

I actually don’t even have a conjecture about what it’s about, so I’m really at square one there.

So basically, these are problems that are not just hard computationally. Some of them require entire branches of mathematics to be developed in order to even start solving them. And these have been seen for years as the ultimate prizes for mathematicians and others in the hard sciences.

AI companies in particular have been watching these problems very closely to figure out, “Hey, could we throw one of our models at one of these and come out with a solution?” They don’t really care about the $1 million prize. A million dollars, you know, buys you snacks in the microkitchen. But—

It gets you one meeting with their top AI researcher.

Right. Right. But it’s more the principle of the thing, the idea of cracking open some of the hardest unsolved problems in mathematics, that is very appealing to them. There were some reports earlier this year that DeepMind was working on one of these problems, in particular the Navier-Stokes equations. And so my prediction is that in 2026, one of these problems will be officially proven and solved.

All right, there it is. I don’t know how to think about the likelihood that that comes true, but I see no reason why it couldn’t come true.

Yeah. Well, I think if one of the AI companies doesn’t do it, you should commit to doing it.

You know what? I have a lot of time open in the back half of next year, so I might take a crack at it. All right, let’s talk about our medium-confidence predictions—

Yes.

—for next year.

What is yours?

And here is mine. I believe that in 2026, we are going to see a group of Democrats win big in the midterms by leaning into anti-AI sentiment.

Ooh, I like this one.

Yeah. So we are already seeing some stirrings of anti-AI feeling coalescing into political action. In recent days, Senator Bernie Sanders called for a moratorium on the construction of new data centers to try to give us more time to figure out how to regulate AI. To me, that was a bellwether announcement.

We’ve also seen a lot of NIMBY behavior in counties where data centers are being constructed. People are saying, “I don’t want you to build this here. I’m worried about the environmental impact, and also I don’t want you to take my job, so don’t build this thing here.” And if we’ve learned one thing from American politics in the past 20 years, it’s that NIMBY movements are extraordinarily powerful.

Yes.

You want to get somebody to the polls? Make them think that their vote will result in something not being built.

Right. It’s—

And it will work. You know what I mean?

Exactly.

And so you have a bunch of Democrats who are still casting around, looking for a message, and I think you have the president giving them a gift because he is still very much in league with the Andreessen Horowitz wing of the Republican Party, which is all gas, no brakes when it comes to AI development. And Trump is basically giving them free rein to say, “You write the policies that you want,” including backing a moratorium on all state-level AI legislation.

Yeah, it does kind of feel like AI is polarizing in a very predictable way. Republicans think it’s good. Democrats think it’s bad. I think that’s why this is a reasonable prediction. I don’t know if it’s going to happen in the 2026 midterms or in the 2028 presidential election, but I agree that this will become a very salient issue for voters in one of those 2 elections.

Yeah, and I mean, you say that Republicans think AI is good, but I would say Republicans do not necessarily think “data center in my backyard” is good, and they do not necessarily think “AI takes my job” is good, right? So that is the actual political opportunity here, and I just think you’re going to see some Democrats seize it.

And by the way, I also think you’re going to see some Republicans seize it. And if you want a secondary prediction here, it’s that you do see some sort of schism next year within the Republican Party over this very issue.

What would the resolution criteria be for this? What would convince you that, in 2026, a group of Democrats had won or found political purchase with this argument?

Let's say you see 3 Democrats flip seats next year who espouse anti-AI sentiment—

Okay.

—in their campaigns.

I like that. My medium-confidence prediction for 2026 is that a primarily AI-generated work will be nominated for a major entertainment award. What do I mean by this? Basically, the big 4: the Emmys, the Grammys, the Oscars, and the Tonys.

I think this is bananas. Why do you think this is true?

So here's my case for this. Number 1, we have talked on the show this year about how AI slop among many people is very popular, right? In the past few months, at least 6 AI or AI-assisted artists have debuted on the Billboard charts. We also know that AI video is starting to get quite good.

Filmmakers like Darren Aronofsky have been experimenting with AI to do storytelling and visual effects in their films. We now have these professional-grade tools for Hollywood studios and filmmakers who are increasingly interested in using them. Some filmmakers obviously are not, but they are not going to be the ones who prove this prediction true.

We've also seen some willingness on the part of the groups that decide these awards to allow AI to get its foot in the door. In fact, earlier this year, the Academy of Motion Picture Arts and Sciences, which votes on the Oscar winners, decided that the use of AI in films would neither help nor harm the chances of achieving an Oscar nomination. Basically, they said it's about the quality of the work.

As long as a human is involved in the film's creation somewhere in the process, it can still be considered for an Oscar. I've also heard rumors that Demis Hassabis at DeepMind is very interested in having them win one of these awards, too.

Yeah, I bet he is. That doesn't mean it's going to happen. I'm also very interested in winning Best Picture, Kevin.

I think you've got a chance at a Tony this year.

Very nice of you to say. Look, this is not going to happen in 2026, particularly in movies, and that is just because among the voter base for these awards, there is such justifiable anxiety about jobs being lost to this technology that they are not going to want to clap it on the back at the end of the year and say, “Huzzah, what a great film you've made.”

So I think film and TV are not going to see it happen. What I will say is that when this does eventually happen—and I do think that eventually something made largely via AI probably wins one of these major awards—it's going to happen at the Grammys.

Because I think people there are already using so many software tools to make songs, and it's not that big of a leap to say, “I'm going to have something generate 100 hooks for me, and I'm going to pick the best one and make a song out of it.” So I think it will eventually happen with a Grammy, but it's not going to happen in 2026.

You might be right, but just so I'm clear, I did not predict that an AI-generated work would win one of these awards, just that it would be nominated.

I see.

That's the resolution criteria.

And it is an honor just to be nominated.

Okay, Casey, what is your low-confidence prediction for 2026? Take a flyer.

My low-confidence prediction is that OpenAI is going to retire Sora—

Mm.

—the text-to-video generator. Sora is an app we talked about a lot on this show this year. It was OpenAI's move into social networking and generated a lot of attention, in part because of the novelty of being able to make a video out of text with your own image or the image of certain celebrities or OpenAI CEO Sam Altman.

But it also generated a lot of controversy over, in particular, the fact that, at least when it launched, it was very easy to make videos out of copyrighted IP. This has caused a bunch of legal headaches for OpenAI. Most recently, they had to strike a deal with Disney and give them a bunch of stock just so that they could get access to some Disney characters.

But I think by the end of 2026, they throw in the towel on this, and here's why. We know that there is a sense that they might be just beginning to fall behind the frontier labs when it comes to the quality of their models. That has huge business implications for them.

This is a company that is running on a razor's edge of existence, where it needs to raise so much money so that it can continue to lose so much money so that it can have a chance at inventing the model that will eventually turn a profit. And I think by the end of 2026, Sora just looks like a big distraction.

During this code-red moment, Sam Altman said, “We are going to place the focus on ChatGPT.” I still think that OpenAI will see value in having video generation. They're just going to say, “Hey, ChatGPT is our champion here. That is the app that is already on everyone's phone. We are going to pivot around building social features into ChatGPT,” rather than have this app out there that is already starting to decline in number of downloads.

So that is my low-confidence prediction for 2026.

And I think this one is a little crazy because I agree that it's possible that OpenAI will retire the Sora app. I think I would have agreed that it was possible they would retire the project altogether until this Disney deal from the other day, which made me confident that this is something they feel like they can actually make real money from.

The terms of the deal—we still don't quite know what they were—but it is plausible to me that major entertainment studios are going to look to Sora as a way to cut their own costs. It's going to become maybe a more professional tool rather than a consumer social media product.

But I think the video-generation technology is good enough and useful enough to professional filmmakers that it will continue to exist.

I mean, I don't know. I think that over the next year we're going to see a bunch of really weird and bad Disney IP slop. It's going to be pregnant Mickey kissing Sonic the Hedgehog or whatever. And Disney might have second thoughts about this, and I wonder how many other companies are going to be lining up to sign a similar deal.

Mm. All right, my low-confidence prediction for 2026 is that Apple will replace Tim Cook after his retirement with an outside CEO. Now, Casey, I'm—you're shaking your head at me. I presume you do not think this is likely. I do not think it is likely either, which is why it is my low-confidence prediction, but I do think it would be interesting.

Sure. Yes, I agree it would be interesting if Apple got a new CEO. I don't think it's going to happen next year, but why do you think it might?

So I think Tim Cook is very likely to retire next year. The Financial Times recently reported that Apple was stepping up its succession-planning efforts as it prepared for him to step down. They also reported that there was a front-runner for Cook's replacement: Apple's senior vice president of hardware engineering, John Ternus.

And I think most people who are careful Apple watchers expect them to go with someone safe, someone inside the company, someone with a long history of running a major division, maybe someone from the hardware side of the business, since that's where most of Apple's mojo has been coming from recently.

But I think this is a company that is slowly and belatedly coming to the realization that it needs to shake something up. It is a stagnant, dormant company that is coasting on its laurels from the tenure of Steve Jobs and the early years of Tim Cook.

I think that there is an outside chance—I would put it at maybe 20% or so—that the Apple board gets Tim Cook's retirement notice and decides, “You know what? We have to go with a dark horse here.”

And who do you think that dark horse will be?

Okay, I have 3 ideas.

Okay.

Again, not saying these are likely—

Don't say Katie Perry.

Not Katie Perry.

Okay.

Number 1: Jony Ive.

Okay.

Jony Ive, of course, was Apple's longtime design chief and worked hand in hand with Steve Jobs. If Apple were trying to convince the public and investors that it still had that special Steve Jobs magic in its DNA, that it was returning to its roots as a design-led company, what better symbolic gesture could it make than bringing Jony Ive back from OpenAI, where his company was recently acquired to work on a hardware product, and convincing him to come back and take up the flag and lead Apple into a bold new era that is really a bold old era?

Mm-hmm. That's not going to happen. So who's number 2?

Number 2, also probably not going to happen: Brian Chesky—

Brian Chesky.

—CEO of Airbnb.

Hmm.

Now, Brian Chesky has been running Airbnb. That's his baby. I don't actually expect him to leave, but I think this is the rare job that could actually convince him to leave Airbnb and do something new. He is known to be a huge Apple fanboy and a huge Steve Jobs fanboy.

He has studied the careers of top Apple executives closely. And I think if the Apple board were looking for someone from the outside who's still very familiar with Apple's culture, its processes, and the way it makes money, it could do worse than Brian Chesky.

Well, I absolutely think he would do that job, and for this reason: Who wants to run Airbnb? Oh, my gosh. Come on, no thanks.

Yes.

Yeah.

Okay, I have a third dark horse candidate. This one would be a real humdinger. And that is Mira Murati.

Mira Murati.

The former OpenAI executive, now CEO of Thinking Machines Lab, a very buzzy AI startup, but not one that has spent a lot of time releasing products this year. I think there is a very remote chance—maybe I would put this at sub-1%—that Apple, in its attempt to reclaim some authority in the AI market, would acquire Thinking Machines Lab, bring Mira and her team into Apple, and maybe, as a result of that deal, say, “You know what? You’re pretty good at this AI stuff. You should run the company.”

Hmm, interesting. I have trouble seeing that happen unless a virus wiped out some significant portion of the population, and they thought, “Who is still around?” And they’re like, “What about Thinking Machines? Are they still up and running over there?” So that’s how I could see this happening.

Okay.

Yeah.

Well, that’s why it’s my low-confidence prediction. But if this happens, I will be able to say, “I told you so.”

Well, Kevin, as always, I appreciate you going way, way out on a limb with these predictions. It’s making me wish that I had perhaps gone further out on a limb. But I do think when the dust settles at the end of next year, I’m going to have gotten more of mine right than you will have on yours.

Well, that’s why we do these predictions at confidence intervals.

Hmm.

If you’re batting 1,000, you’re doing it wrong.

That’s true. But unfortunately, I am still batting 1,000. Anyway, what else is on the show?

Okay, well, that’s our predictions. And, Casey, you wanted to make one more set of predictions this year. Talk about that.

That’s right. So here’s the thing. I imagine listeners may get to the end of this segment and say, “Okay, guys, thanks for these predictions. Here’s what I really want to know: What is going to be the state of the AI bubble at the end of 2026? Is it still going to be foamy and frothy, like it feels right now, or is there going to be a major change in the landscape?” Kevin, what say you to that?

Yeah, so this separate category of AI bubble prediction, I think we are going to see some economic turmoil as a result of AI in 2026, but I do not think we are going to see the kind of calamitous bubble pop or market collapse that some people have predicted. I think Nvidia, Microsoft, Google, OpenAI, and Anthropic—these tentpole AI companies—will mostly be fine. There may be some startups that raised money at high valuations that will have to either do a down round or be acquired at a discount, sell themselves off for parts, things like that.

I’m skeptical of some of these so-called neoclouds, the companies like CoreWeave that exist to rent GPUs and other AI infrastructure to the big suppliers. I think they might struggle in 2026. But I would be very surprised if we had a single huge dot-com-style collapse in 2026, and so I don’t want to make a prediction to that effect.

But I do think that one possible thing that could happen next year is that there are companies that are so-called SaaS companies—software-as-a-service companies—that start to really struggle. Why are they struggling? These are companies like Salesforce and Zenefits, companies that sell software to businesses on a premium subscription plan. I think some of those companies’ customers may start building software themselves.

Mm-hmm.

Because now, with these AI coding tools, they can spin up a custom CRM, a custom benefits manager, or custom payroll software that suits their needs exactly, and they can do it at a fraction of the cost of buying a subscription from one of these big enterprise companies. So that is my AI bubble prediction: Some of the companies that claim they are AI companies will start to struggle, but it will not be because this technology has not panned out or is not useful. It will actually be because the technology is quite good.

I think that is a very sensible prediction. My views largely align with yours. I think at the end of 2026, things will still feel quite bubbly, and for this reason: The frontier AI labs are going to be making tens of billions of dollars more in revenue by the end of next year than they are this year, and that is going to prop up large swaths of this economy, right?

There is absolutely no sign that businesses are going to stop buying services from these frontier labs, or that consumers are getting turned off by them, and that is the thing that is powering this whole bubble. Now, as you say, on the margins, some companies may not be able to figure it out, and I wouldn’t be surprised if we saw some big flame-outs next year, maybe some of the companies that you mentioned. But I do think that at the end of 2026, the engine is still going to be very, very hot, and it’ll be interesting to see that happen.

I’ll make one more quick AI bubble prediction, which is that 2026 is the last year to buy a house in San Francisco. I think this coming year we will see the IPOs of at least 1 major AI company. Anthropic is reportedly planning an IPO for next year. We may also see OpenAI go public. SpaceX may go public. Stripe may go public. There are these other highly valued private companies that may go public.

When that happens, their employees get rich, and they go buy houses. So if you are looking for real estate in San Francisco, the right time to buy was yesterday. Well, that’s enough predictions. Let’s do a production. When we come back, our annual Christmas carol.

The people demand it.

What’s the worst that could happen?

Well, we could sing.

5. The Christmas Song Returns

Well, Casey, the holidays are upon us, and as always at this time of year, we have a special gift for our listeners.

We do, and it’s the sort of gift that I think maybe this is the year that Kevin won’t want to give. Maybe he’ll come up with a different gift. But in fact, it is the same gift that we will now regift to you in 2025.

You hate this part of the episode, but I love it because it is our annual chance to break out our mellifluous singing voices and show our audience the Christmas spirit.

Yes, and this is my annual chance to remind you that Kevin is an accomplished singer and longtime choir member, while I have the singing voice of a broken harmonica.

Yes. I think you’re a very good singer, for what it’s worth.

Thank you.

And it is also our annual chance to wear our ugliest holiday sweaters. If you are not watching this on YouTube, you are missing out, because you are wearing a sweater with Santa as a golfer?

That’s right. As a huge golfer, I thought, what better chance to bust out the sweater that I spent $50 on yesterday because it was the only sweater that could arrive in the studio on time?

Wow, I’m so impressed.

Whereas Kevin is wearing a beautiful sweater that says “Tree Rex” on it and has a T. rex with a star on top of his head. It also has many flashing, real Christmas lights affixed to the sweater. So you’ve got a really high-tech production going on over there, Roose.

I do, yes, and I’m being warned that this may cause seizures for some of our YouTube audience. So I’m going to turn off the electric blinking lights here and just wear the Tree Rex sweater. But yes, this was a big hit at the kids’ holiday party last weekend.

The last thing we need is the seizure-cancellation scandal to take us out into 2026.

Okay, Casey. If you will indulge me in our annual caroling tradition, it is time for this year’s rendition of A Hard Forkin’ Christmas.

And we should say, if you heard it last year, it's not the same song. Things have changed. See if you can spot the differences.

Your lyrics.

Thank you, sir.

Do you want to do some scales to warm up?

No. Let's just croak it out.

Okay. Who's going to be the Elphaba, and who's going to be the Glinda?

You're definitely Elphaba.

Okay. This is not a Wicked thing. This is a Christmas classic. Iconic. Families will be singing this around the hearth this year.

Call that Hearthfork.

Oh, boy. So, Casey, let's close our laptops for the year and wish our listeners happy holidays with “A Hard Forkin' Christmas.”

Let's do it.

On a Hard Forkin' Christmas, my true love gave to me: a bot trained on all our IP.

On a Hard Forkin' Christmas, my true love gave to me: 2 robot pants and a bot trained on all our IP.

On a Hard Forkin' Christmas, my true love gave to me: 3 code reds, 2 robot pants, and a bot trained on all our IP.

On a Hard Forkin' Christmas, my true love gave to me: 4 Humane Pins, 3 code reds, 2 robot pants, and a bot trained on all our IP.

On a Hard Forkin' Christmas, my true love gave to me: 5 Bum Bum creams.

Delivered from a drone.

4 Humane Pins, 3 code reds, 2 robot pants, and a bot trained on all our IP.

On a Hard Forkin' Christmas, my true love gave to me: 6 Roblox scandals, 5 Bum Bum creams, 4 Humane Pins, 3 code reds, 2 robot pants, and a bot trained on all our IP.

On a Hard Forkin' Christmas, my true love gave to me: 7 White House meme coins, 6 Roblox scandals, 5 Bum Bum creams, 4 Humane Pins, 3 code reds, 2 robot pants, and a bot trained on all our IP.

On a Hard Forkin' Christmas, my true love gave to me: 8 Mecha Hitlers, 7 White House meme coins, 6 Roblox scandals, 5 Bum Bum creams, 4 Humane Pins, 3 code reds, 2 robot pants, and a bot trained on all our IP.

On a Hard Forkin' Christmas, my true love gave to me: 9 Meta reorgs, 8 Mecha Hitlers, 7 White House meme coins, 6 Roblox scandals, 5 Bum Bum creams, 4 Humane Pins, 3 code reds, 2 robot pants, and a bot trained on all our IP.

On a Hard Forkin' Christmas, my true love gave to me: 10 Signal war chats, 9 Meta reorgs, 8 Mecha Hitlers, 7 White House meme coins, 6 Roblox scandals, 5 Bum Bum creams, 4 Humane Pins, 3 code reds, 2 robot pants, and a bot trained on all our IP.

On a Hard Forkin' Christmas, my true love gave to me: 11 Chimpanzinis, 10 Signal war chats, 9 Meta reorgs, 8 Mecha Hitlers, 7 White House meme coins, 6 Roblox scandals, 5 Bum Bum creams, 4 Humane Pins, 3 code reds, 2 robot pants, and a bot trained on all our IP.

On a Hard Forkin' Christmas, my true love gave to me: 12 AI bubbles, 11 Chimpanzinis, 10 Signal war chats, 9 Meta reorgs, 8 Mecha Hitlers, 7 White House meme coins, 6 Roblox scandals, 5 Bum Bum creams.

Bum, bum, bum.

4 Humane Pins, 3 code reds, 2 robot pants, and a bot trained on all our IP.

That's intellectual property. Can Ezra Klein do that? I don't think so. Merry Christmas, you filthy animals.

Merry Christmas. See you next year.