AI in 2026: hype, whiplash, and… home robots?

Note: Transcripts are automatically generated from episode audio, and are not fully corrected for spelling, grammar, and formatting.

RANA EL KALIOUBY: The thing I’m most excited about for 2026 is: How AI can actually help us build deeper human connections and more meaningful human experiences? And the way this happens is through AI that can really help you organize your relationships and your network and surface connections that you need. Do you know the Devil Wears Prada? It was Anne Hathaway, and she was with Miranda, Meryl Streep at this gala. And this guy and his partner were moving towards Miranda and she was like, “Oh my God, it was this guy.” And Anne whispers in Miranda’s ears, she’s like, “He’s the ambassador and his new wife.” And I was like, “That’s exactly what I need. I need an AI version of Anne Hathaway.”

BOB SAFIAN: That’s Rana el Kaliouby, AI scientist, entrepreneur, investor, and host of the podcast, Pioneers of AI. In today’s episode, I talked with Rana about her predictions for AI in the year ahead, how the technology will evolve, its impact on our work and our everyday lives. We also dig into some big headlines surrounding AI from whether there’s an AI bubble to the battle between OpenAI and Google. Rana cuts through the clutter, illuminating key challenges, as well as harnessing joy and excitement for what’s to come. So let’s get to it. I’m Bob Safian, and this is Rapid Response.

[THEME MUSIC]

I’m Bob Safian. I’m here with Dr. Rana el Kaliouby, AI scientist, entrepreneur, investor at Blue Tulip Ventures, and host of the terrific podcast, Pioneers of AI. Rana, great to talk with you.

EL KALIOUBY: Yeah, great to see you again, Bob.

Copy LinkAI innovation’s shift from academia to industry

SAFIAN: You are yourself a pioneer of AI as a researcher, as a founder, an expert in emotion AI, the relationship we have with technology. I’m curious because 2025 has felt like this tipping point in the mainstreaming of AI. For someone like you who’s sort of been doing this for so long, is it odd that AI is suddenly the center of every conversation or is it kind of – “finally”?

EL KALIOUBY: I think there’s a bit of finally for sure, but it’s kind of like there’s definitely some of us who are OGs in this space and have been around for a while, and then there’s all these newcomers sometimes without a lot of basis. So there’s definitely mixed emotions.

But the thing that strikes me the most, I just got back from Fortune Brainstorm AI conference. I co-chair the conference, I’ve been doing this for the last five years. And I had this onstage interview with three amazing, both kind of scientists and entrepreneurs who have straddled both the world of academia and industry in AI. And it struck me that just over the last decade, the power shift has moved from all the breakthroughs happening in academia to happening in industry. That was a really kind of thought provoking conversation. Have we stopped innovating in AI in the academic circles?  And it’s all interesting.

SAFIAN: Well, there’s so much money going into the industry side of it, right? The resources are there.

EL KALIOUBY: Right. And a lot of the building of, especially, the AI infrastructure is very compute intensive, and it’s quite expensive. I think what struck me from this conversation was that yes, a lot of the LLMs we use today are industry generated and they’re kind of shipped and published through industry, but to create real breakthroughs or the next breakthroughs in AI, we really need strong academic industry partnerships, and we need to reimagine what those relationships look like. So I thought that was really telling. I don’t think the real scientific breakthroughs will come from industry alone.

SAFIAN: That’s where your background was, academic research. I guess there’s a different way of approaching it when your priority isn’t how you’re going to monetize it.

EL KALIOUBY: Absolutely. You need to be able to think about these long-term big moonshots without it having to map to a business model or a commercialization plan in the short term.

SAFIAN: So Rana, I’m really eager to ask you about what we should expect for AI in 2026, if you’re game to talk about that. I’d also love to start maybe by asking you some updates about things that happened in 2025.

EL KALIOUBY: Let’s do it.

Copy LinkThe multi-trillion dollar question: Are we in an AI bubble?

SAFIAN: All right. So let’s start with the biggest headline really, which is whether there’s an AI bubble. You’ve asked guests and pioneers of AI about this. Is today’s AI obsession, has it become kind of a craze or are valuations out of whack? Where are we sitting?

EL KALIOUBY: Yeah. I think there’s definitely some craziness going on. There’s crazy valuations. There’s a bunch of companies that are raising seed rounds without a product, without customers at these crazy valuations. But I also think at the same time, it’s still very early days. And as you know, I invest in early stage AI start-ups so I’ve had to think a lot about this. The way I see it is that AI is basically creating this massive structural shift in how value is created. And we’re moving from this idea of a software as a service to service as a software. And to just simplify it, instead of selling you a tool that will help you do your job, I don’t know, 10% or 20% faster, AI can just get the job done. So instead of, I don’t know, selling a law firm a tool that helps its lawyers become more efficient, you just sell an AI lawyer.

And we’re seeing this shift in business model, which actually means that we’re expanding the TAM for AI. So we’re not just disrupting the software industry, we’re actually disrupting the services industry and the labor market. It’s a multi-trillion dollar opportunity, which is very exciting. And I think a lot of people don’t realize that. And that’s why I don’t think it’s a true, true bubble.

SAFIAN: I mean, I hear from the guests on my show this sort of back and forth between like, “Oh my gosh, I got to get a piece of this.” And at the same time, this sort of anxiety that if I spend on this, I’m not really sure what my ROI is going to be yet. And they’re sort of caught between like, “I don’t want to fall behind, but I also don’t want to waste money.”

EL KALIOUBY: Yeah. I think there will be, unfortunately, money wasted. Again, whether you are a company that’s leaning into AI or an investor that’s investing in AI, I think there will be some collateral damage here. But I also think that’s what happens when you’re exploring a new innovation anyway, right? You’re going to take a risk and sometimes these risks will pan out, sometimes they won’t. I try to channel my MIT MediaLab kind of days and the framework there is even if you “fail”, because we never called anything a failure, you learn. And I think it’s important to lean in and learn.

Copy LinkInside the competition between OpenAI and Alphabet

SAFIAN: Another headline I wanted to ask you about, the horse race between AI chatbots. We’ve seen Google’s Gemini come on strong against OpenAI’s ChatGPT and Sam Altman at OpenAI calling it a code red.

EL KALIOUBY: I actually think these chatbots are very quickly becoming commoditized. I think they’re being trained on very similar data sets, the algorithms are kind of the same. There might be differences at the fringes. So for example, the new Gemini/Nano Banana Pro is really good at image generation in a way that ChatGPT isn’t quite yet. Although I just read today that they just released a new version, so I’ve got to try that. But there was this really interesting study that my son shared with me, and it’s called the Artificial Hive Mind, where this group of researchers studied how similar or different the different models are. And by and large, they’re kind of the same.

SAFIAN: I mean, I’ve seen this headline this week about someone using AI for dynamic pricing, right? Instacart, I guess is testing it, that the price of everything may shift in the future, like airline tickets. Do you look at this as good, bad, inevitable, sort of this is just one of the AI implications that we’re going to have?

EL KALIOUBY: I have mixed emotions about that one, because first of all, I don’t actually think you needed the new version of AI to do dynamic pricing. Dynamic pricing has been around algorithmically for a long time. Something feels really cringey about dynamic pricing for just our groceries. Something doesn’t sound quite right about that. But if you think about it, every time you take a Lyft or an Uber, that’s dynamic pricing. Every time you book a flight, it’s dynamic pricing. So there’s a lot of precedent, and I don’t know how we decide what is okay to price dynamically and what is not. I think that does introduce a lot of potential discrimination and biases, unfortunately.

SAFIAN: Who can be taken advantage of by the dynamic pricing? The AI will get better at knowing who it can poach more.

EL KALIOUBY: Exactly.

Copy Link“A race for human intimacy”

SAFIAN: There have also been headlines about personal relationships with AI chatbots, mental health therapy, romance, concerns about suicides and other bad choices. I’m curious how much you saw all this coming. I mean, your personal mission has been to humanize technology, but right now it seems like the tech is almost too human sometimes or addictive, I guess, especially for those in need, even if it’s not really human.

EL KALIOUBY: I heard, I think it was Yuval Harari who said this, he said the last decade was a race for human attention and the next decade will be a race for human intimacy. And you can kind of already see that with a lot of these AI companions/friends. They are trying to not just grab your attention, but build this intimate relationship with you. And I worry about that, I worry about that as a mom of a teenager. I believe that a lot of companies who are building these kind of companions and friends, to our point about building guardrails, are not prioritizing guardrails at all. They’re just maximizing for time on the platform or time with your AI friend. That’s very dangerous.

I’m all for AI that can augment our human connections, but not take away from our human relationships. And I think that happens by these platforms putting guardrails in all sorts of ways. Why should you be able to converse with your chatbot for five hours in a row? It should two hours in, it should say, “Bob, enough. Go talk to a real human being.” But the companies aren’t incentivized to do that.

Copy LinkPrediction #1: The rise of relationship-intelligent AI

SAFIAN: Let’s look ahead to 2026. You sent me some fascinating thoughts about AI’s next phase impact on business, and I’d love to take you through them. The first one was the rise of what you called relationship intelligent AI.

EL KALIOUBY: So everybody’s worried that AI is going to make us less human and take away our human to human connections. We just talked about that, there is definitely a risk of that. But I think the thing I’m most excited about for 2026 is how AI can actually help us build deeper human connections and more meaningful human experiences. And the way this happens is through AI that can really help you organize your relationships and your network and surface connections that you need and maybe make warm introductions to you. I love connecting with people, but my relationship data is a mess. It’s all in my brain. Some of it is in LinkedIn, some are on WhatsApp. I take a lot of notes when I meet new people and I use an AI note taker. It’s just a mess. It’s very disparate data sources.

And I always think of this scene in, do you know The Devil Wears Prada? Have you seen?

SAFIAN: Oh, that’s exactly what I was thinking of. The character is kind of whispering in your ear.

EL KALIOUBY: Exactly. It was Anne Hathaway, and she was with Miranda, Meryl Streep at this gala. And this guy and his partner were moving towards Miranda and she was like, “Oh my God, it was this guy.” And Anne whispers in Miranda’s ears, she’s like, “He’s the ambassador and his new wife.” And I was like, “That’s exactly what I need. I need an AI version of Anne Hathaway.” And it’s now doable with LLMs because it’s all this unstructured, messy data that an AI can take all of that, contextualize it, and hopefully be that AI chief of staff for you.

SAFIAN: Is that like a product? Is that something that you would have to do to turn your chatbot, whatever, your Claude or your ChatGPT into that? Or is it a new product that you think is going to come out that will make that easy for you?

EL KALIOUBY: There are already a number of new companies that are starting in this space. So one company’s called VIA.AI, it’s a Boston-based company. They do this for sales professionals and BD professionals who have to do this for their work. There’s a company called Goodword that I’m very excited about. They’re doing this for just the average person. Like you and I, we have very strong networks, but how can we organize it? So I’m excited about that one. There’s a company called Boardy that does this for investors and founders. So it’s becoming a thing and I’m excited to see how these companies take off in 2026. They’re all fairly new, so it’ll be interesting to see how they evolve.

SAFIAN: Yeah, and whether they can stay ahead of some of the bigger chatbots that may just try to integrate some of this capability into the products they already have. That’s always the case in this kind of evolution of technology:  What’s a feature and what’s a company, right? What’s an independent service?

EL KALIOUBY: Absolutely. When I’m looking at these companies and I’m diligencing them, that’s a key question that I ask. Is this something that the next version of ChatGPT or Gemini is just going to implement? And if the answer is yes, then that’s obviously not a defensible company. But a lot of times there’s this additional moat of data and algorithms that you need to sit on top of these LLMs. And I believe in this relationship intelligence space, I don’t think this is something that just a kind of an off-the-shelf LLM can do. It really needs to know you. It needs to know your data, it needs to know your relationships.

SAFIAN: And you have to trust it enough to share it with that, all that data with it, right?

EL KALIOUBY: Absolutely.

SAFIAN: That’s your proprietary data, whether it’s about your business or about you individually.

EL KALIOUBY: Exactly. And I don’t want this to all go up to OpenAI’s cloud. I want to trust that I have control over these really private relations. If you and I have a conversation about our kids, I don’t necessarily want that to now sit in a general OpenAI cloud and be used to train the next ChatGPT. So that safety and security, appreciating the privacy and the importance of this data is really key.

SAFIAN: Like many things in life, the more we put into AI models, the more we’ll get out of them. Still, I’m not quite sure I trust enough to upload my Rolodex and network details into GPT just yet. As Rana says, that could change in 2026. So what else is she predicting AI for the year ahead? We’ll get into AI’s next infiltration of the workplace, as well as what she calls physical AI after the break. Stay with us.

[AD BREAK]

Before the break, Dr. Rana El Kaliouby, host of the podcast, Pioneers of AI, took us behind the biggest headlines in AI today and why she thinks a personal AI chief of staff is coming in 2026. Now we dig into other Rana predictions for AI in the year ahead, from how to onboard an AI into your org chart to what she calls physical AI. Let’s dive back in.

Copy LinkPrediction #2: The insertion of AI into the org chart

Another business change you expect in 2026 is the insertion of AI into the org chart. This is about who manages AI, like performance reviews and team culture impacts?

EL KALIOUBY: Yeah, so this goes back to this thesis that there’s this shift in how AI is creating value, and it’s not a tool anymore. Well, it is a tool. It’ll always be a tool, but it’s not a tool that helps you get work done faster. It could actually take an end-to-end task and get it done for you. And I’ll give a few examples.

So I’m an investor in a company called Synthpop, and instead of building a tool that helps healthcare administrators accelerate or really become efficient in how they do patient intake, it just takes the task of patient intake. It does the thing end to end. And so if you then imagine what that means for a hospital or a clinic, it will have a combination of human workers collaborating and working closely with AI coworkers.

And so then the question becomes, well, who manages these hybrid teams? Sometimes it’s a human manager, sometimes it’s an AI manager. I’m also an investor in a company called Tough Day, and they sell you AI managers. And then how do you do performance reviews for these hybrid teams? How do you build a culture? Like at Affectiva, my company, culture was our superpower. How do you build a culture when some of your team members are AI and some of your team members are humans?

So I think that is going to spur a lot of conversation around how do you build organizations that are combinations of digital agents and human employees?

SAFIAN: Yeah, well, and it’s interesting, this sort of team culture as you talk about Affectiva, I think a lot of businesses have felt that their culture is one of their competitive advantages and distinctive features. And yet if more and more of your jobs are being done by AIs and those AIs are, those agents are sort of off the shelf, how do you make your culture differentiated or integrate that agent, that AI into your culture?

EL KALIOUBY: Exactly. And my prediction is we’ve already seen that actually with this company, Tough Day, that sells AI managers. Their first client was the state of Hawaii, it was the Hawaii Employer Association. And this company’s based in New York, where it was trained. The Hawaiians basically said, “This sounds, like, very New York. We need more Aloha spirit.” So they actually had to go back and collect data from the Hawaiian customer employee conversations and use that to train their AI. So I think we’re going to see more and more of these AIs and AI agents become customizable to a certain culture or to represent certain values of a company.

Copy LinkIs AI actually killing jobs?

SAFIAN: As you talk about this merging of AI agents and humans in work, I mean, it brings up that looming question about the impact of AI on jobs and employment. And there’s some numbers that are coming out now that make it seem like, “Oh, it’s bad for jobs.” There are other numbers coming out that’s like, “Oh, we’re actually hiring more people because of it.” Do you have a prediction about what is going to happen with that in 2026? Is AI going to take over roles that have been done by humans that quickly?

EL KALIOUBY: We had a really fascinating round table discussion at the Fortune Brainstorm AI conference and the headline was like, “Is AI killing entry level jobs?” And actually, a lot of the Fortune companies and also AI companies that were around the table were basically saying, “No, we’re hiring more entry level jobs. They’re just not the same jobs that we were traditionally seeing.” And also the career ladders have changed too.

So my prediction is we’re going to see an entirely different organization where I think if you are able to come in an entry level position, for example, but work very closely with AI and be AI native and be AI fluent and be able to wear multiple hats, I think that’s going to go a long way. As opposed to this very siloed job trajectory where you come in, this is your little task, and then you do more of it, and then you go up the career ladder. I think that’s going to change. I think young people are looking for different ways of working, and I think AI is changing all of that anyway.

Will there be jobs that will go away? I think so. I can’t remember who said this line, but it’s now very popular. It’s not AI that’s going to take your job. It’s going to be somebody who knows how to use AI. And I believe that to be true.

Copy LinkPrediction #3: Memory will be the new AI moat

SAFIAN: Another prediction for 2026 you’ve shared was that memory will be the new AI moat. You’re not talking about memory like RAM, right? Like computer memory. You’re talking about generative AI having a memory like a human, so understanding context differently?

EL KALIOUBY: Yes, exactly. So because these LLMs are basically becoming commoditized and sounding the same and kind of presenting information in very similar ways, I think what’s going to be different is how much does it know about you? How much context does it know about you? Does it know your preferences, your emotional state, your history? And right now, all these LLMs, their concept of memory is very naive. It’s just like basically a memory dump of every single conversation you’ve had with say at ChatGPT. Which is fine, it’s actually already sticky. I had a problem with logging into my ChatGPT a while ago and I was trying to get somebody at OpenAI to help with that. And they basically said, “Oh, the only solution is you’ve got to start from scratch.” Just like, “Forget about this account, sign up with a different email.” And I totally freaked out. I was like, “Are you guys kidding me? It’s like three years of conversation. I am not doing that.” And anyway, they fixed it, thank goodness.

But that was an aha moment for me because I cared so deeply about the history of these conversations that I was not ready to just, the switching cost is really high and we’re going to start to see more and more of that, not just at the LLM level, but at the solution and product level. So products that are able to really thoughtfully convert your conversations and your history within AI into an understanding of who you are and your preferences and personalized recommendations and whatnot. That’s going to be very sticky and it’s going to turn AI into a trusted copilot.

SAFIAN: And the moat, I guess, as you’re talking about it, you’re giving this example about yourself, you’re not going to switch from ChatGPT to somewhere else if you’ve got three years worth of memory in there.

EL KALIOUBY: Exactly.

SAFIAN: Unless that’s something you could port over, but I don’t even know how you necessarily could do that. So what that means is you’re kind of making a decision now as to whether, am I going to be a Perplexity person? Am I going to be a Gemini person? And then once that has your memory, that’s where you are.

EL KALIOUBY: Right. I mean again, I use several tools, but ChatGPT definitely has a lot of my personal stuff. It has a lot of my professional stuff. Yeah, there’s no way to port it to other systems at the moment and that is a key consideration in my decision around which of these models to use.

I do think, however, as you know, I studied memory for many years, and there is a close interaction between memory and your emotional experiences. And I am excited to see companies implement some of these more sophisticated models of memory into AI. So right now, it’s just a flat history of all your sessions, but memory is weighted on importance. It’s okay to prune memories. We definitely do that as humans, and that’s a good thing.

SAFIAN: Yeah. Some things we don’t want to remember, right?

EL KALIOUBY: Right. We just want to delete. And a lot of these frameworks don’t exist in AI today, and I think we’re going to start to see more of these and that’ll be, again, more differentiation.

Copy LinkPrediction #4: An increasing importance of inference

SAFIAN: You also see ahead in 2026, you mentioned the increasing importance of inference. I had a little harder time following this. Inference is different than memory?

EL KALIOUBY: Oh, so in the world of AI, there’s the training of the model. That’s the thing that most people think about when they think about heavy compute and it’s intensive in terms of its energy consumption and all of that. And people don’t really think about inference. Inference is, every time you prompt Gemini or ChatGPT for a suggestion for dinner or to create a logo for you, that is inference. You’re just calling the model to answer a question for you. You’re not training the model. Training takes months, right? Inference, hopefully, at this point takes a minute to get you an answer.

As the demand for AI increases, the cost of inference is going to go down. These models are becoming more efficient and so we’re going to see more and more demand for inference. We’re just going to be calling more AIs to do more things for us. And that shift will unlock more demand for intelligence at the edge. I don’t want to every time I ask an AI for an answer for it to go to some cloud somewhere. I want it to just run on my phone, I want it to run in my car.

I interviewed Renee Haas, the CEO of Arm, on the Fortune stage, and he talked about how eventually AI will be embedded in our everyday devices, like maybe our fridge or our stove or something. So we’re going to see more of these shifts of the conversation around compute in AI from the training stage to the inference stage.

SAFIAN: When I’m making a call, doing an inference, isn’t the model then going to train on that data later? It’s just not doing it at that moment? Or are those completely separate processes?

EL KALIOUBY: They are definitely separate processes, but actually it’s a great question because there is a mode in most of these LLMs where you can say, “Do not use my data for training.” It’s not like you send your prompt to the cloud somewhere, and it immediately takes this data and retrains the model. This is not how it works. Basically, these training phases, they take months and there’s a lot of planning that goes into it. It’s not something that is updated on – I mean, although the cycles are becoming faster and faster with this AI race.

SAFIAN: They’re aggregating the data they’re going to train on and then they run a training session for however long that may last to sort of shape the next version of the AI, of the software.

EL KALIOUBY: Exactly.

Copy LinkPrediction #5: Physical AI will become more mainstream

SAFIAN: I mean, AI is software. It doesn’t live in the physical world, at least not yet. And that’s one of the things that you see potentially shifting in 2026 from robotics to AI native interfaces like glasses, so that physical AI will become more mainstream.

EL KALIOUBY: Absolutely. We’ve already seen some of that in 2025, like huge investments in humanoid robots, but I actually think that’s just one specific rendition of what physical AI will look like. We will see a whole slew of different robotics, some of them humanoid, some of them not. I’m especially excited for robots that are designed to be in the home, but I think we will see more of these things that are embodied versions of AI and have world models. A lot of AI today, they’re great language models, but they’re not necessarily world models. They don’t have perceptive abilities. So for me, it’s this trifecta of sensor data paired with really novel disparate sources of data, and then both predictive and generative AI that’s going to unlock a lot of use cases in the physical world.

SAFIAN: I’m curious, as you look ahead to 2026, are there lessons from your interviews on Pioneers of AI that you find yourself going back to, being reminded of, harking back to in some way?

EL KALIOUBY: I really liked my conversation with Mark Cuban. We did a mini Shark Tank AI edition thing on the show, which was great. But he also, I guess I knew that, but hearing him say it really solidified it for me, the importance of IP, right? The importance of intellectual property, whether it manifests in data or perhaps it’s in your trademark secrets at a company, patents, all of this IP is differentiating. Because again, as these models look for ways to differentiate themselves, it’s going to go back to what data they’re using to train these models. And we’ve kind of tapped out the publicly available data.

And so it’s going to all kind of center around what do you do with this unique data set that you have? Or if you’re a founder, can you get access to these unique data sets to train or build a product that is truly defensible?

SAFIAN: Because that IP will define what is distinctive in your products moving forward.

EL KALIOUBY: Exactly.

SAFIAN: Well, Rana, this was great as always. Thanks so much for doing it.

EL KALIOUBY: Yeah. Thank you for having me.

SAFIAN: Rana’s perspective always gets me energized about what’s possible in AI. Am I still wary about AI’s impact on mental health? Absolutely. Do I think AI companies are doing enough to safeguard vulnerable users? Not really. But I’m cautiously optimistic that we’ll make strides over the next year to make AI safer and better. After all, progress is rarely linear. For business leaders, new ways of managing our networks, of onboarding AI into an org chart and company culture, these topics are only going to get more important. The more directly we address weaving AI into our businesses and our lives, the more we can maintain our human first priorities. So as we build out plans and roadmaps for 2026, let’s make room for what’s daunting as well as what’s exciting. On both sides, change is coming and we just have to embrace it. I’m Bob Safian. Thanks for listening.