Skip to content
a16za16z

Where does consumer AI stand at the end of 2025?

As 2025 comes to a close, consumer AI is entering a new phase. A small number of products now dominate everyday use, multimodal models have unlocked entirely new creative workflows, and the big labs have pushed aggressively into consumer experiences. At the same time, it is becoming clearer which ideas actually changed user behavior and which ones did not. In this episode, a16z consumer investors Anish Acharya, Olivia Moore, Justine Moore, and Bryan Kim look back on the biggest product and model shifts of 2025 and ahead to what 2026 may bring. They discuss why consumer AI appears to be trending toward a winner-take-most model, how subtle product design choices can matter more than raw model quality, and why templates, multimodality, and distribution are shaping the next wave of consumer products. Where do startups still have room to win? How will the role of the big labs continue to change? And what will it actually take for consumer AI apps to break out at scale in 2026? Timecodes: 0:00 Introduction & Market Overview 0:35 Who Won Consumer AI in 2025? 1:22 Major Model and Product Launches 2:24 Image and Video Model Innovations 4:05 Product Sensibility and User Experience 6:40 Advances in Image and Video Models 9:11 Under-hyped Products and Productivity Apps 10:21 Prosumer Tools and Power User Workflows 12:56 Gemini vs. ChatGPT: Distribution and Brand 14:04 Product Nuances: Onboarding and Engagement 16:09 Social Features and Group Chat 19:26 Sora, TikTok, and the Status Game 21:02 Challengers: Claude, Perplexity, Grok 23:37 Meta, Grok, and Rapid Model Progress 26:54 Predictions for 2026: Enterprise, Apps, and Multimodality 32:02 Startup Opportunities and App Generation 36:27 Favorite Products and Recommendations 42:19 Building Consumer Products in 2026 Resources: Follow Anish: https://twitter.com/illscience Follow Olivia: https://twitter.com/omooretweets Follow Bryan https://twitter.com/kirbyman01 Follow Justine: https://twitter.com/venturetwins Stay Updated: If you enjoyed this episode, be sure to like, subscribe, and share with your friends! Find a16z on X: https://twitter.com/a16z Find a16z on LinkedIn: https://www.linkedin.com/company/a16z Listen to the a16z Podcast on Spotify: https://open.spotify.com/show/5bC65RDvs3oxnLyqqvkUYX Listen to the a16z Podcast on Apple Podcasts: https://podcasts.apple.com/us/podcast/a16z-podcast/id842818711 Please note that the content here is for informational purposes only; should NOT be taken as legal, business, tax, or investment advice or be used to evaluate any investment or security; and is not directed at any investors or potential investors in any a16z fund. a16z and its affiliates may maintain investments in the companies discussed. For more details, please see a16z.com/disclosures.

Olivia MoorehostBryan KimhostAnish AcharyahostJustine Moorehost
Dec 29, 202543mWatch on YouTube ↗

EVERY SPOKEN WORD

  1. 0:000:35

    Introduction & Market Overview

    1. OM

      For most of the year, less than 10% of ChatGPT users even visited another one of the big LLM providers like Gemini

    2. BK

      When you open Gemini, it has a pop-up, says, "We got NanoBanana. Would you like to do something with it?" The, the little pane where you have to pick something.

    3. OM

      Yeah.

    4. AA

      Mm-hmm.

    5. BK

      I don't, I don't know what to do.

    6. OM

      Yeah. [laughs]

    7. BK

      These are product nuances that I, I think makes people actually take the first step.

    8. OM

      The models have gotten to the level of quality that you can build a real scalable app on top of them, and so the hope is 2026 will be a huge year for consumer builders. [upbeat music] Today

  2. 0:351:22

    Who Won Consumer AI in 2025?

    1. OM

      we're talking about who won consumer AI in 2025. This was arguably the year that we saw the big model providers, OpenAI and Google most out of everyone, make a major push of their own into consumer, both in terms of new models they released, but also in terms of new products, features, and interfaces that target the mainstream user. Um, you might wonder, why does it matter who is in the lead here? There are some early signs that the general LLM assistant space might be trending towards winner take all or at least winner take most, so only 9% of consumers are paying for more than one out of the group of ChatGPT, Gemini, Claude, and Cursor. And for most of the year, less than 10% of ChatGPT users even visited another one of the big LLM providers like Gemini.

  3. 1:222:24

    Major Model and Product Launches

    1. OM

      If we had to call it now, ChatGPT is currently in the lead by far at 800 to 900 million weekly active users. Gemini's at an estimated 35% of their scale on web and about 40% on mobile, and everyone else significantly trails this. So Claude, Grok, Perplexity are all about 8 to 10% of the usage. But especially in the last three to six months, things are changing very quickly. With the launch of new viral models like NanoBanana, Gemini is now growing desktop users 155% year over year, which is actually accelerating even as they reach more scale, which is pretty crazy to see, and ChatGPT is only growing 23% year over year. And we're starting to see players like Anthropic almost specialize within consumer, owning different verticals like the hyper-technical user. So today we've brought together the a16z consumer team to recap what we saw this year from the big model companies and consumer, and also to predict what might be ahead of us in 2026.

    2. AA

      Cool. Well, thank you, Olivia. It's, it's been a super

  4. 2:244:05

    Image and Video Model Innovations

    1. AA

      fun year. If we kind of wind the timeline back to last January, maybe we should start with what we saw, launches, products, what worked, what didn't. So Justine, tell us what you saw this year, OpenAI, Google. What are you paying attention to? What have you changed your mind on?

    2. JM

      Yeah, those two in particular had a ton of consumer launches, like Olivia mentioned. Um, from a model perspective, I would say... I would argue their most viral models this year, at least among consumers, were in image and video. So for OpenAI, it was the ChatGPT 4.0 image, the Ghibli moment.

    3. OM

      Which is crazy that that was this year.

    4. JM

      It, it seems like that this is this year.

    5. AA

      Three years. [laughs]

    6. BK

      Yeah, it does.

    7. JM

      It feels like it was years ago. Um, and then Sora, obviously, Sora 2. And then for Google, uh, it's Veo, Veo 3, and Veo 3.1, um, and then NanoBanana and NanoBanana Pro in image models, which went insanely viral, probably comparable to if not beyond, um, the, the Ghibli moment for OpenAI. I think in terms of the product layer, what we saw was OpenAI tended to keep more things in the ChatGPT interface, so like Pulse, group chats, shopping, research, tasks, all of these features launched inside ChatGPT as the core. Um, the exception there is obviously Sora as a standalone video app. Whereas Google tended to launch more things as standalone products, so they did ship a lot through like Google AI Studio, um, and Google Labs, and Gemini, and the plethora of Google [laughs] surfaces there are to launch a product. Um, but they would also ship things as standalone websites that you could go to and visit, which basically allowed for a more custom interface for each type of product, not just the kind of chat entry, chat exit, or image/video exit.

    8. AA

      So Justine, I

  5. 4:056:40

    Product Sensibility and User Experience

    1. AA

      have a question for you on that. So it felt like 18 months ago we were talking about Midjourney, and most of the multimodal models were defined by aesthetics and realism.

    2. JM

      Yeah.

    3. AA

      Is that still true? Like, what changed this year?

    4. JM

      Yeah. I think there's definitely different styles still, and I think Midjourney, when you talk to people really deep in image and video, it still kind of stands apart for this like aesthetic sensibility that a lot of the models don't have if you don't know how to prompt for it. But I would say this year in particular, we made a lot more strides on realism and also on reasoning within both image and video, like all of the little details that make an image or a video actually seem real. For example, if you have a person walking and talking, the people in the cars in the background, if they're on a street, should be moving in the correct direction, like they shouldn't be morphing and looking strange. Um, in image we were able to, you know, have, uh, multiple input images and text and sort of reason across all of those uploads to create like a cohesive design or something like that, um, which was not something we saw happening last year for sure.

    5. BK

      Yeah. La... I remember when we were excited about having a letter show up correctly in images.

    6. JM

      Yes, yes.

    7. AA

      Right.

    8. BK

      And now we have insane infographics.

    9. JM

      Yes.

    10. OM

      Yeah.

    11. BK

      We can just put up amazing YouTube video and say, "Give me an image that, image that explains this."

    12. JM

      Yeah.

    13. OM

      Totally.

    14. BK

      That's incredibly different.

    15. OM

      NanoBanana Pro can even generate like market maps. Like I can tell it-

    16. BK

      No, I... You, you generated a market map. It's incredible

    17. OM

      ... generate a market map of this space, and it will... like, it either has or will go do the web research within the image model, which is crazy, to get the correct list of companies and then pull their correct logos-

    18. BK

      That's what we're gonna do now

    19. OM

      ... which is insane. [laughing]

    20. JM

      I know. There's, there's one benchmark left that the reasoning image models have not cracked, not... I tested GPT Image 1.5 yesterday. They sometimes struggle with like both reasoning and multi-step reasoning. So what I've been testing is you upload a picture of an Monopoly board and you say, "Remove the names of all the properties-

    21. BK

      Mm.

    22. OM

      Mm

    23. JM

      ... and replace them with names of AI labs and startups." And, um, GPT Image 1.5 is actually the closest, but it's very hard for them to

    24. OM

      Do all of those steps. Remove it, come up with the new names, put all of the new names in the correct places, make sure there's not overlaps or one thing you mention three times and another big player you never mention. So there's still some room to go-

    25. BK

      Yeah

    26. OM

      ... on the image evals. [chuckles]

    27. BK

      True. And it's interesting, um, that especially from the image model from ChatGPT, where you-

    28. OM

      Yeah

    29. BK

      ... can actually see perseveran- perseverance of like it, it, it carries a character-

    30. OM

      Mm-hmm

  6. 6:409:11

    Advances in Image and Video Models

    1. AA

      me, it felt like the most under-hyped aspect of Nano Banana was the integration with search because it feels like there's realism, you know, which is physics and sort of other things that feel, you know, we're, like we're on Candy Valley. There is reasoning, which is apply modifications that are adherent to what the user asked for. But then there's also sort of accuracy, and for me, a good example of this is product photography. If you say, "Hey, generate-

    2. BK

      Hmm

    3. AA

      ... a photo of this album cover-

    4. OM

      Hmm

    5. AA

      ... or a historically accurate photo of this moment in time," you have to actually have the search integration, and that was sort of non-intuitive, but is actually very useful.

    6. BK

      Hmm.

    7. OM

      Totally. Yeah. It's kind of like the Veo 3 moment when I don't think it was intuitive to people that video would be cracked necessarily by bringing audio together with video in the same place.

    8. AA

      Yeah.

    9. BK

      Right.

    10. OM

      And that ended up being the thing that made AI video go viral.

    11. AA

      Yeah.

    12. OM

      Like since Veo 3, and, and now Sora-

    13. AA

      Yeah

    14. OM

      ... maybe dominates, but like since Veo 3, my social feeds have been like full of really realistic looking AI videos all the time.

    15. BK

      I, I counted. About one-fifth of my feeds are AI generated.

    16. OM

      Amazing. Yeah.

    17. AA

      Wow. What, what do you guys do? There's so many launches this year, and many of them went well, like Veo and Nano. Um, what do you think is under-hyped or products that you think didn't get enough attention? Bryan?

    18. BK

      It's a good question. I think, uh, under-hyped Pulse of the world is probably still under-hyped. And, you know, we're talking about OpenAI, Google, which to me fall under productivity category. So if you actually think about if you go to App Store today, top five out of top 10 productivity apps are all Google. The... It's insane, and, and ChatGPT is number one. So we're talking about a productivity category where it's like helps you do things, and I feel like a lot of people are trying this from a different angle. Like, how do I actually ingest your data or your schedule, your email to make it more helpful and give more proactive, you know, notification to you? I think a lot of people are working on it. Given the frequency of people using ChatGPT, which I think is, what? 25, uh, times a week?

    19. AA

      Mm-hmm.

    20. OM

      Mm-hmm.

    21. BK

      Pretty good. Pretty good. Three to four times a day. It feels like it's a really good position to actually give you proactive nudges and summary and help your life in general. So feel like, um, the everything app was always this myth in, in, in the Western world. Think, uh, OpenAI is trying to move in that direction, where it's ingesting enough, people are going there enough-

    22. AA

      Mm-hmm

    23. BK

      ... to start giving really useful proactive nudges, and I think that's a space that I'm excited about. It's interesting.

    24. AA

      But are you a DAU? BK?

    25. BK

      I am not a DAU.

  7. 9:1110:21

    Under-hyped Products and Productivity Apps

    1. AA

      Of Pulse?

    2. OM

      Well- [laughs]

    3. BK

      I'm not, not, not of Pulse.

    4. OM

      Similarly, I tried Pulse for a while-

    5. AA

      Yeah

    6. OM

      ... and have kind of largely turned off of it.

    7. BK

      Yeah.

    8. OM

      But I would agree with you that I feel like Pulse and a couple other examples that OpenAI launched this year are kind of new primitives or ideas that feel under-hyped-

    9. AA

      Hmm

    10. OM

      ... but because the execution-

    11. BK

      Yeah

    12. OM

      ... is a little off-

    13. BK

      I think it's the execution.

    14. OM

      The usage is off. Another example that I would give, which is similarly like personal contacts, would be their connectors.

    15. BK

      Yeah.

    16. OM

      So now you can... And you can do this on Claude as well. You can connect your calendar, your email, your documents. And so hypothetically, you could say to ChatGPT, you know, "Read all of my memos over the past six months and like summarize-

    17. BK

      Yeah

    18. OM

      ... what's most interesting, least interesting." I think when that works, it's really exciting.

    19. BK

      Yeah.

    20. OM

      I have found it to be a little bit unreliable so far, but I think as the models get better, they have a real chance to kind of own the prosumer-

    21. BK

      Yeah

    22. OM

      ... workspace if they get that right.

    23. BK

      Prosumer is perfect category 'cause like we, we talk about it sometimes, but 99% of people don't run their life on Calendar.

    24. OM

      Yeah.

    25. BK

      We do.

    26. AA

      Right.

    27. BK

      But, um-

    28. AA

      Right

    29. BK

      ... so that, that's when I'm thinking about the actual average frequency of using ChatGPT.

    30. AA

      Mm-hmm.

  8. 10:2112:56

    Prosumer Tools and Power User Workflows

    1. AA

      Yeah. Olivia, I feel like you're the ultimate power user. You know, what are you still using?

    2. OM

      Yes.

    3. AA

      What's your stack?

    4. OM

      It's a great question. From all of the larger model companies, actually, I would have to say the thing that I'm still using the most and was maybe the most impressed by this year was the Perplexity Comet browser.

    5. BK

      Mm.

    6. AA

      Mm.

    7. OM

      And I don't and was not using Perplexity as my core general LLM assistant. I use ChatGPT and Claude much more, but I think they really executed on it in a first-class way in terms of both the agentic model within the browser, but also, perhaps more importantly, all of the workflows-

    8. BK

      Mm

    9. OM

      ... that you can set up that allow you to basically run the same task over and over, either at a preset time or when you trigger it on a certain webpage. So that to me was a really exciting launch, and if you look at the data, like the spike at launch and the sustained traffic for Comet was actually much higher than for ChatGPT's own bro- browser launch, Atlas, which is kind of crazy given how much more-

    10. BK

      Yeah

    11. OM

      ... distribution ChatGPT has than Perplexity. Um, but I think they also launched an email assistant this year, Perplexity did, and they made a couple acquisitions of really strong agentic startups. And so what I would love to see from them next year is like more of these dedicated prosumer interfaces. I feel like that would be an awesome direction for them to kind of double down in.

    12. AA

      They do feel like the startup that has the biggest breadth of ambition.

    13. OM

      Yeah.

    14. BK

      Mm-hmm.

    15. AA

      You know, alongside the labs and sort of big tech.

    16. OM

      Yes.

    17. AA

      Like it's very, very impressive just the number of things they've shipped this year.

    18. OM

      Yes.

    19. BK

      Yeah.

    20. OM

      Definitely.

    21. AA

      What, um, you know... One, one thing I wanted to ask you, Justine, was sort of Gemini feels like it's having a real moment because of all the image and video models.

    22. OM

      Yeah.

    23. AA

      Do you think it can overtake ChatGPT? Is there truly that much demand for these types of models?

    24. JM

      I think yeah. So what we, what I've seen basically is, um, there is always nearly infinite demand for, like, the best-in-class image or video model.

    25. BK

      Mm-hmm.

    26. JM

      Um, because n- then you have a mix of tons of different people seeing it and wanting to use it. You have pro- like, if you're using it professionally, if you're marketing or in entertainment or storyboarding or whatever, um, you always wanna be using what's at the forefront of the field, and so you're totally fine to go somewhere other than ChatGPT and Sora to, to get access to VO. Um, even if you're an everyday consumer, so many new viral trends are created around new capabilities of the best-in-class image and video models, and so that ends up driving users into different products-

    27. BK

      Yeah

    28. JM

      ... that they may have never tried before. Like you might be downloading the Gemini app or accidentally ending up on Google AI Studio, which I know they're trying to make be more for developers to use NanoBanana Pro, which a lot of users, I think, experienced in the past couple of months.

    29. BK

      Yeah.

  9. 12:5614:04

    Gemini vs. ChatGPT: Distribution and Brand

    1. BK

      The interesting thing about Gemini to me is, like, hypothetically, they benefit from the massive Google distribution advantage.

    2. JM

      Mm-hmm.

    3. BK

      Like if you look at Android, Gemini is at, like, 50% of ChatGPT's scale on mobile.

    4. JM

      Mm-hmm.

    5. BK

      Whereas on iOS, it's, like, 17%. So, like, clearly something is working there. They launched a little Gemini widget within Chrome recently-

    6. JM

      Mm-hmm

    7. BK

      ... that encourages you to use it.

    8. JM

      Yeah.

    9. BK

      They're launching it within Google Docs and Gmail and other things.

    10. JM

      Yeah.

    11. BK

      But I think that most... the average person is still just using one AI product.

    12. JM

      Mm-hmm.

    13. BK

      And ChatGPT is like the Kleenex of AI.

    14. JM

      Yes. Yeah.

    15. BK

      Like it is the brand that has become-

    16. JM

      The verb, the noun.

    17. BK

      Exactly.

    18. JM

      Yes, yes, yes.

    19. BK

      And so I think that Gemini still has a pretty big hurdle to overcome-

    20. JM

      Yep

    21. BK

      ... just in terms of that.

    22. JM

      Yeah.

    23. BK

      Um, but if they keep doing what they're doing on these amazing, um, viral consumer creative tool launches and model launches, like, they could get there next year. I was thinking about this. It's, uh, really interesting when you look at Gemini, whi- which is everywhere.

    24. JM

      Yeah.

    25. BK

      But yet nowhere to some extent, right? You don't like, uh, you know, when you look at the actual usage, people still think of the Kleenex.

    26. JM

      Yep.

    27. BK

      Um, and they go to ChatGPT. But the

  10. 14:0416:09

    Product Nuances: Onboarding and Engagement

    1. BK

      interesting thing also is on the product sensibility.

    2. JM

      Mm-hmm.

    3. BK

      So this morning, I had, like, two panes open, OpenAI's image model and, um, and Google's, uh, Gemini, and basically use a image functionality.

    4. JM

      Yeah.

    5. BK

      When you open Gemini-

    6. JM

      Mm-hmm

    7. BK

      ... it's a blank screen. It has a pop-up that says, "We got NanoBanana. Would you like to do something with it?"

    8. JM

      Mm-hmm.

    9. BK

      And, uh, it's the, the little pane where you have to type something.

    10. JM

      Yeah.

    11. BK

      I don't, I don't know what to do.

    12. JM

      Yeah. [laughs]

    13. BK

      ChatGPT, you go in, and it has a very TikTok-like style of like, here are the trending themes-

    14. JM

      Mm-hmm

    15. BK

      ... that you might wanna generate, and you click on I want a sketch pen or whatever, and then just, like, use one other picture, and it creates something amazing. And then it says, "Would you like a holiday card? Would you like a blah, blah, blah, blah, blah." These are product nuances that I, I think makes people actually take the first step-

    16. JM

      Yeah

    17. BK

      ... to generate it.

    18. JM

      Yeah.

    19. BK

      And then once you have it, you have character consistency.

    20. JM

      Yeah.

    21. BK

      So you keep going.

    22. JM

      Right.

    23. BK

      So that's interesting in that I think OpenAI and ChatGPT has proven that there's deeper product sensibility.

    24. JM

      Yep.

    25. BK

      But then I... This is a funny thing, maybe, maybe a little non-kosher thing to say, but, you know-

    26. JM

      [laughs]

    27. BK

      ... m- I, I worked at Snap, so when you look at Meta versus Snap-

    28. JM

      Mm-hmm

    29. BK

      ... famously, uh, Evan Spiegel was chief product officer of Meta.

    30. JM

      Yeah, yeah.

  11. 16:0919:26

    Social Features and Group Chat

    1. AA

      maybe going in a slightly different direction, BK, I'm very curious for your take on OpenAI's social features.

    2. JM

      Mm.

    3. AA

      'Cause it does feel like that's something that you really have to get product execution right on, but also-

    4. BK

      Yeah

    5. AA

      ... network design. You know, there's some efforts around Sora 2. We should talk about that.

    6. JM

      Yeah.

    7. AA

      There's also group chats within ChatGPT. You're our sort of social guy or have been historically.

    8. BK

      Yeah.

    9. AA

      Bullish, bearish? Where's your head at?

    10. BK

      Uh, bearish for now.

    11. AA

      Okay. [laughs]

    12. BK

      Uh, and the reason to me is twofold. Uh, historically, we, we look at sort of... It, it's funny. I, I look at products based on what I call inception theory. You go, like, three to four layers down to figure out what the one-liner is, which is, like, I want my dad to love me.

    13. AA

      [laughs]

    14. BK

      And so, you know, when they, when they think of a product-

    15. AA

      Is that for you or for the world?

    16. BK

      That's, that's for me-

    17. JM

      [laughs]

    18. BK

      ... as well as for a lot of people.

    19. AA

      Okay. Yes, yes.

    20. BK

      Um, and, and so I, I look at some of the, you know, um, products like, like ChatGPT. Ultimately, when you peel the onion five times, I think essentially is help me be better. Like, help me get that information, help me be more productive, help me be more efficient. And then when I think about social features, Meta, Instagram, what have you, or even TikTok, the two layers of information or the, the, you know, emotion that it's trying to address to me is, for TikTok, entertain me. I want my clown. Entertain me.

    21. JM

      Yeah.

    22. BK

      And then the other layer is I'm lonely, I wanna be seen, I wanna connect with people.

    23. JM

      Mm-hmm.

    24. BK

      And to me, these are pretty two different parallels in the product direction.

    25. JM

      Mm-hmm.

    26. BK

      And OpenAI's product is incredible. It's magic. It's amazing. But it's ultimately a see me or, or help me category-

    27. JM

      Mm-hmm

    28. BK

      ... which essentially is why it's the number one in productivity category.

    29. JM

      Yeah.

    30. BK

      Now we're trying to take this and shove it in people's life and say, "Guys, connect. Connect better, and, like, actually-

  12. 19:2621:02

    Sora, TikTok, and the Status Game

    1. BK

      Sora's competition or analogy isn't actually TikTok.

    2. JM

      Mm-hmm.

    3. BK

      It's actually CapCut.

    4. OM

      Hmm.

    5. BK

      It's like a funny way. It's like-

    6. JM

      Yes

    7. BK

      ... almost like a creative tool.

    8. OM

      Yes.

    9. AA

      Hmm, interesting.

    10. OM

      Yes.

    11. AA

      Olivia, what's your take?

    12. OM

      I think, well, I was gonna say, like, I think it goes back to your earlier point, which is, like, the kind of motion that drives social apps is both these, like, positive and negative feelings of, like-

    13. BK

      Yeah

    14. OM

      ... oh, I'm publishing this thing of myself-

    15. BK

      Oh, oh, yeah

    16. OM

      ... that's kind of sensitive, or that I want people to think it's this or that or this other thing. And so that's kind of what drives participation on the app.

    17. JM

      Yeah.

    18. BK

      The status game.

    19. OM

      Yeah.

    20. BK

      All about the status game.

    21. OM

      It is exactly the status game, and when it's AI-generated content and people know it's not real, like a real representation of you as a human being, the status game is lost a little bit.

    22. AA

      Absolutely lost.

    23. OM

      Yeah. I think the status game comes then with, um, can you prompt something very cool?

    24. JM

      Yeah.

    25. OM

      But that's a different type-

    26. BK

      That's a different-

    27. OM

      ... of product.

    28. JM

      Yeah.

    29. OM

      And that's why I think it goes viral on, like, Twitter and all these other existing platforms.

    30. AA

      I mean, my, my sort of counterpoint or bull case for Sora 2 is I actually think the status game was about humor more than anything else.

  13. 21:0223:37

    Challengers: Claude, Perplexity, Grok

    1. AA

      So w- what do folks think of the challengers? You know, we're talking about Sora 2.

    2. BK

      The biggies.

    3. AA

      I mean, Meta, it's crazy to talk about Meta as a challenger.

    4. BK

      [laughs] Challenger.

    5. AA

      I guess in this context they are-

    6. OM

      [laughs]

    7. AA

      ... but I think Claude, Perplexity, Grok are the more obvious names for challengers.

    8. JM

      Yeah.

    9. AA

      Olivia, what's your take?

    10. OM

      I love Claude. I talk to Claude all the time.

    11. AA

      [laughs]

    12. OM

      Claude has somewhat replaced ChatGPT for me as my general LLM. I think Claude is opinionated in an interesting way. Um, I also love Claude because I'm willing to invest time into building out AI workflows.

    13. JM

      Hmm.

    14. OM

      I think Claude actually launched a lot of really powerful things this year around, like, artifacts and skills-

    15. JM

      Mm-hmm

    16. OM

      ... where you can essentially set up tasks or workflows to run, um, over time. I do think the reason it hasn't hit the mainstream yet is even the way they built those things is geared towards a technical user or an engineer.

    17. AA

      Uh-huh.

    18. OM

      It's, I think they tried to make skills as easy as they could to create, and it still was not anywhere near easy enough for the mainstream-

    19. AA

      Yeah

    20. OM

      ... consumer. Um, another example would be they were actually the first of the big players to kind of launch file creation, slide deck creation, editing.

    21. JM

      Yeah.

    22. OM

      And they branded it as, like, file generation and analysis or something-

    23. AA

      [laughs]

    24. OM

      ... and it was, like, a toggle feature within-

    25. BK

      A FGNA

    26. OM

      ... a setting bar of a setting bar or something.

    27. AA

      [laughs]

    28. BK

      Right.

    29. OM

      So, like, very few people used it.

    30. JM

      Yeah.

  14. 23:3726:54

    Meta, Grok, and Rapid Model Progress

    1. OM

      need to hear Justine's take on both Meta and Grok, as I feel like they both had fascinating years in different ways.

    2. JM

      Okay.

    3. BK

      Oh.

    4. AA

      Yes.

    5. JM

      So Meta, um, hired all those researchers. I think their strongest models are actually not consumer-facing models. It's their SAM three series, so, like, they segment anything for video, for image, and for audio, and basically, like-For video, for example, you can upload a video, and you can describe in natural language, like, "Find the kid in the red T-shirt," and it will find and track that person across every- the entire video, even if they're coming in and out of the frame. It will let you apply f- effects like blurring them out or removing them or whatever. And you can imagine a similar thing with audio, um, with different stems, and then with image, with different objects in an image. I think we're gonna see next year, hopefully, some incredible consumer products built on top of those models, but today they're more of a playground for developers than they are a consumer-based product

    6. OM

      Which is surprising given just, like, the DNA of the company.

    7. JM

      Yeah. The-- So the one good consumer feature I think they've launched this year with AI is the Instagram AI translations.

    8. BK

      Mm-hmm.

    9. OM

      Mm.

    10. JM

      Where, um, when you're uploading a reel now, you can opt in to enable translations, and it will clone your voice, translate it into five different languages, um, apply the translation with your voice, so, you know, and then redub with, with the lip sync.

    11. OM

      Wow.

    12. JM

      And so it basically makes it seem like you're a native speaker in, in whatever language. So I would love to see more of that stuff come to, to the Meta products. Um, Grok, I think, has had... So Grok had a crazy year with, like, the companions-

    13. OM

      Yes

    14. JM

      ... with all of the LLM progress and the coding progress. Um, I think their image and video progress is probably the steepest slope I've seen of any of the companies. Like-

    15. BK

      Mm-hmm

    16. JM

      ... um, may- it was probably like six months ago they didn't even, like, have image and video models.

    17. BK

      Mm-hmm.

    18. OM

      Mm-hmm.

    19. JM

      And they're shipping so fast to launch new features. Like, it was initially just image to video. They added text to video. They added audio. Then they added lip sync with speech. Then they added 15-second videos. Like, they're just not slowing down the speed of progress, and Elon has made a bunch of statements about, like, wanting more interactive video game-type content out of Grok-

    20. OM

      Mm-hmm

    21. JM

      ... and wanting movies out of Grok by the end of next year. So, um, let's hope it continues to go at that pace. [chuckles]

    22. BK

      Do you feel like it's a pincer movement, where, like, on one hand there's, like, a very l- uh, infrastructural model layer of, like, let's get to the... let's top the LLM Arena charts, and then the other one is like, "Let's go Ani"?

    23. JM

      I think-

    24. BK

      It's like a little bit-

    25. JM

      Yeah

    26. BK

      ... of like a bifurcated move.

    27. JM

      Yeah. Right. Like the entertainment and the, like, smart.

    28. BK

      Absolutely.

    29. JM

      Yes.

    30. BK

      But entertainment in a way that, like-

  15. 26:5432:02

    Predictions for 2026: Enterprise, Apps, and Multimodality

    1. OM

      from '25 to '26, what are some of all your predictions for next year? What do you think we'll see? Hardware, models, commerce we haven't spoken about yet, so what do we think will play out? I think, I know this is, we're talking about consumer, but one of the things that's been really maybe underrated for me about ChatGPT that we might see more of next year is they've really made a push into the enterprise.

    2. JM

      Mm.

    3. BK

      Yeah.

    4. OM

      Both with the traditional enterprise licenses and then working with specific companies to even, like, train models for them.

    5. BK

      Mm-hmm.

    6. OM

      And I think when we think about the fact that most consumers only use one general LLM product, um, ChatGPT Enterprise usage, they published a big study, but it's up s-somewhat like eight or 9X year over year.

    7. JM

      Yeah.

    8. OM

      And so if we're entering a world now where people have to use ChatGPT for their company or as part-

    9. JM

      Yeah

    10. OM

      ... of their work-

    11. BK

      Yeah

    12. OM

      ... that could really translate into consumer usage.

    13. JM

      Yeah.

    14. OM

      Um, or maybe they become the c- the workspace with the connectors and some of the other things that they're investing in, um, and someone else owns the consumer-consumer use cases.

    15. JM

      Yeah.

    16. OM

      I think to that end, we have to talk about their push into apps, and I think-

    17. JM

      Yeah

    18. OM

      ... whether or not that works is gonna be kind of the defining question for them next year. Yeah, and I think that the... We've all discussed the importance of the apps SDK and the apps directory, as they're calling it, and it's gonna be a huge new channel for consumer. I think what's less discussed is it's hyper-relevant to enterprise.

    19. JM

      Mm-hmm.

    20. OM

      So I think where ChatGPT shines is where it's able to operate across a number of tools for one workflow.

    21. JM

      Right.

    22. OM

      And if you think about the number of things you do in your sort of business day-to-day that operates across many tools, it's, it's most of those things.

    23. JM

      Yeah.

    24. OM

      So I think that will have very interesting implications for the SaaS ecosystem, and it's a part of the App Store we're not talking about as much.

    25. JM

      Yeah.

    26. BK

      Yeah. Um, maybe less of a prediction, but, um, thinking through 2025, and we talked about all the big move from big labs. And from the startup point, um, I think one of the biggest trend we've seen is app generation.

    27. JM

      Mm-hmm.

    28. BK

      A-and I think, uh, there's a real world where we see the big labs, uh, with the distribution and the frequency of usage of people coming in to start saying, "Look, like, maybe there's a common type of product and apps that we could actually help you generate within the confines of the big lab products."

    29. JM

      Yeah.

    30. OM

      Yeah.

  16. 32:0236:27

    Startup Opportunities and App Generation

    1. OM

      the cons- the labs have launched in consumer, they've done a great job with models, and they've gon- done a great job with incremental things that improve the core experience of using like a ChatGPT or a Gemini.

    2. JM

      Yeah.

    3. OM

      In my opinion, we've gone through dozens of things that they've launched or tried as new consumer products or new consumer interfaces like Group Chat, like Pulse, like Atlas, like Sora. Google has had a long tail like Stitch, Gems, Opal, Doppel, Tons.

    4. JM

      Yeah.

    5. OM

      None of those are really working.

    6. BK

      Yeah.

    7. OM

      And I think it's because it's not the core competency of these companies anymore-

    8. JM

      Yeah

    9. OM

      ... to build opinionated standalone consumer UI. Out of all of those, I think the product that's working the most is like Notebook LM.

    10. BK

      Yeah.

    11. OM

      And that's one of like maybe 20 things that Google has tried or experimented with. So I think it's actually very positive for startups in that, consumer startups, in that the models will keep getting better, which the startups can use, and they'll keep, you know, they'll make ChatGPT better and better.

    12. JM

      Yeah.

    13. OM

      But I don't necessarily think that ChatGPT like verticalizes into all of these other amazing use cases-

    14. BK

      Yeah

    15. OM

      ... or products, and there's still room for startups to be building there.

    16. BK

      I have a yes and to that.

    17. OM

      Okay. [laughs]

    18. BK

      Where absolutely, but however, when the input and the output is text-

    19. OM

      Yep

    20. BK

      ... where ChatGPT and Gemini of the world shine the most-

    21. OM

      Mm-hmm

    22. BK

      ... no matter how deeper you go, no matter how specific you think your text output is going to be, essentially given the frequency of use-usage of the main big lab products-

    23. OM

      Yeah

    24. BK

      ... I think it's gonna be really hard to stitch that and get that away from that usage-

    25. OM

      Yep

    26. BK

      ... if your product is mainly text in, text out.

    27. OM

      Mm-hmm.

    28. JM

      Yep.

    29. BK

      So I, I do think you, you have to be creative around what is the angle that you can like go steal people away from.

    30. AA

      You know, I love that you used the word opinionated because I think that for labs, um, certainly for big tech and perhaps increasingly for labs, their priorities get set in their promo committee-

  17. 36:2742:19

    Favorite Products and Recommendations

    1. AA

      specific recommendations. Like, after this pod, what are the products people should download, or the features or the models, what should folks be using today?

    2. JM

      I, I guess on the multimodal point, I think one really, um, under-hyped product that people should check out, not because they'll use it every day, but because it shows sort of what is possible when you combine, like, an agent with image with text, is Pomeli. So this is like the Google Labs product where you put in the URL of your business-

    3. AA

      Mm-hmm

    4. JM

      ... and it has an agent go to the website, pull all of the product and brand photos, summarize what it thinks your brand's, um, aesthetic is-

    5. BK

      Hmm

    6. JM

      ... what it stands for, what kind of customers it's targeting, and then it will generate three different ad campaigns for you. And it will generate not only the text, but it will generate, like, the Instagram posts. It will generate the flyer. It will generate, like, the, the photo of your product in this, you know, w-whatever, wherever it thinks it should be based on your customer. Um, and very cool product. Would be hard to become a giant standalone product within Google, I think. But, um, shows sort of-

    7. BK

      Hmm

    8. JM

      ... the future of what happens if we combine agents with generation models that have sort of really deep understanding of context that an image model-

    9. BK

      Hmm

    10. JM

      ... or a video model normally wouldn't have.

    11. AA

      Startup products, though. Do you have a favorite startup product?

    12. JM

      Oh, startup.

    13. BK

      [laughs]

    14. JM

      In creative tools?

    15. AA

      Yes.

    16. JM

      In creative tools?

    17. AA

      Yes.

    18. JM

      I think, I, I mean, we're investors in Krea, so this is biased, but I think they, they've, they've really done an exceptional job of, um, being the best place to use e-every model-

    19. AA

      Yeah

    20. JM

      ... or every quality model across every modality, and also building more of the interface on top of these models. Like-

    21. BK

      Hmm

    22. JM

      ... I now prefer to use NanoBanana Pro on Krea because Krea allows you to-

    23. BK

      Hmm

    24. JM

      ... save elements, which are essentially characters or styles or objects that you can, like, @ tag to re-prompt, versus having to drag the same image reference-

    25. AA

      Yeah

    26. JM

      ... into NanoBanana over and over again.

    27. AA

      That's a good one.

    28. BK

      Um, I suppose it falls under startup category, uh, again, shilling companies.

    29. JM

      Mm-hmm.

    30. BK

      But, you know, um, the one that I use the most is actually ElevenLabs Reader.

  18. 42:1943:14

    Building Consumer Products in 2026

    1. AA

      Anyone want to throw us home?

    2. JM

      Oh, I mean, the obvious one is we are very actively investing in consumer companies, and I, and I genu-

    3. AA

      Thank you.

    4. JM

      I think a lot of people say this.

    5. BK

      Good one.

    6. JM

      I genuinely believe that the models have gotten to the level of quality that you can build a real scalable app on top of them. Wabi is a great example of this.

    7. AA

      Yes.

    8. JM

      Um, and so the hope is 2026 will be a huge year for consumer builders-

    9. AA

      Mm-hmm

    10. JM

      ... um, not just, like, consumers as, um, c- like, consumers being consumers of a product.

    11. AA

      Yeah.

    12. BK

      Yes.

    13. AA

      Well, thank you all for a super fun year in consumer and AI. We'll be back with more next year, and merry Christmas, guys.

    14. BK

      This is a wrap.

    15. AA

      Yeah.

    16. JM

      Happy holidays.

    17. BK

      Happy holidays.

    18. AA

      Happy holidays. [outro music]

Episode duration: 43:14

Install uListen for AI-powered chat & search across the full episode — Get Full Transcript

Transcript of episode p4-7x6QiYr0

Get more out of YouTube videos.

High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.

Add to Chrome