Andreessen Horowitz's Olivia Moore on top 100 consumer AI apps: agents rise, DeepSeek fades, Sora surprises

will come back.

He goes by rat thing. Thanks, Mike. Anyway, we have Olivia Moore from Andresa and Horowits. She's a partner there in the stream regularly. Let's bring her in. How are you doing, Olivia?

Good. Thanks for having me.

Thanks for hopping on. Sorry about the uh the global chaos in the oil markets delaying this appearance, but I'm glad we had time to actually digest the report uh because there's so many interesting details in there. And whenever you drop one of these big reports, I feel like you sort of need the Twitter hive brain to like dig through it and find all the interesting commentary and and quote each quote tweet each other until there's like a consensus. But uh take us through the actual project. What did you launch? How long have you been working on this? and then we'll go into some of the interesting discoveries.

Yeah. So, we do this every six months. It's it's one of the most fun parts of my job actually because I think the the Genesis was back in 2023 when we were wondering, you know, the the Tech X community has their own group of products that they use and love, but like what does the average person actually care about in AI? So, we do this every six months. We basically pull every single website and every single mobile app, rank them by usage and pull the top 50 on each side that are um kind of AI native or now majority AI enhanced. Um and it gives a really interesting picture of kind of what what normal people use and care about in the AI world.

Great. Let's talk about the data because when I first started seeing these charts of like oh this is this company's winning, that company's winning, I was naturally sort of skeptical. I was like, "Oh, sensor tower are pixels really accurate in the mobile age." But then I actually dug into the Yipit data which is credit card based and that seemed really like way more reliable. So talk to me about are you paying for data? Is this data do they give this to you because you're friendly with them? Like how does the data work? And then how confident are you about the data? What are the pitfalls? What do you like? Where are you seeing the data be reflective?

Absolutely. Yeah. So our website data is from similar web which we have a paid subscription to. Same with sensor tower for the mobile data.

Yippid is something I agree with you. I think it's more reliable and it's something that we want to lean on more going forward.

The other thing that gets kind of significantly underounted when you just look at web visits and mobile mouse is all of these desktop products like cursor, cloud code, granola, whisper flow that people are using. And so I think for the next few lists we're going to have to shift the methodology more towards I mean it's helpful to see what's getting traffic but I think at this point as consumer AI is maturing we also want to see what people are paying for.

One last question on data I mean uh Andre Horitz is a huge firm at this point. Have you have you considered doing uh what Nate Silver was talking about, a Consumer Reports style interview panel with uh experts that are maybe uh disconnected from a particular company and just sort of surveying everyone, getting some qualitative data, some quantitative data, and sort of putting together more uh more adoption data that's maybe slightly less sanitized. and and analytical and and quantitative but paint a bit different picture.

I would love to do that because again I think like we are in our own world of what products we use and talk about um and the rest of the the world uses all different things like people in medicine, people in law, people in retail even are using AI products that like we have probably never heard of or interacted with and so I'd love to get more of that qualitative stuff in there in the future.

Great. So take us through uh the biggest movers, the biggest surprises, the biggest narratives that maybe uh should never have changed in the first place. Whatever your takeaways were,

we're definitely starting to see the industry mature. So there's kind of fewer new entrance than we've seen in the past versions of the list where like every time half the list was new.

To me, there were I think two probably most interesting takeaways from this one. One would be the rise of agents. So, Jen, Spark and Manis both made the list as horizontal consumer agents. Open Claw uh would have made the list if we pulled it in February. Our data was from January, but it would have ranked at number 30, which is like a very strong debut, especially for a product that's only for someone who knows how to use Terminal, which is like 1% or less of the population.

That's true.

The other takeaway, which I feel like is on everyone's minds right now, is the kind of three-horse race between Chachi BT, Claude, and Gemini. And there's the traffic data alone, which is helpful. And then if you kind of tease out some more of the product strategy and the engagement data, like there's kind of different stories happening there. So that was fun to dig into. And uh the the main takeaway from the uh from the the the mainstream consumer in uh just the foundation the chat apps what what are you seeing between the chat GBT gemini claw deepseeek perplexity gro

yeah it's interesting I would say deepseek has completely fallen off in the US it still makes our list pretty high because it's like the number one AI product in China and Russia which are really big markets.

Do you have any do you have any personal theories like things that you can't like my thinking with Deepseek was that all the downloads originally when it just started charting out of nowhere were just 100% it was just all entirely botted. I have no way to prove that other other than other than it was just going up the chart like crazy and there was no nobody was actually using it. no one was talking about it other than the fact that it was at the top of the chart.

Yeah, I think that's totally possible. I think we're actually seeing in different in many ways, but a little bit of an anal an analogous story playing out with Claude right now where like pre all of this press, whether it's positive or negative, no one in the US knew what deepseat was. Pre all of this press, I think there was some survey that Claude had like 2% market awareness in the US. Yeah. And so we see this thing happen where like even if it's like the worst headline of all time, if it's going mainstream, like it will you it will drive people to try and use the product and then we just have to see if they retain and they didn't on Deep Seek.

Okay. Yeah, that makes a lot of sense. Uh what about uh who who who fell off? Uh there there was this kind of big wave of companies maybe like a year or two ago that were just like trying to confuse people into thinking they were chat GVT, right? They were

chat AI. That's the one I used.

Is that not the main one?

It's funny. I I never take like joy in any company falling off the list except the mobile app you the mobile list used to be full of all of these like I think they call them fleeceware apps that are developed in like Eastern European app studios which are basically like charging you for the free version of ChachiVT and pretending like they're the premium version and Apple took a while but they finally cracked down on them um thankfully. Yeah.

The other category we've seen a little bit of a decline in that we talked about in the report is standalone image generator products.

Um, and that's largely a result of the fact that the image models within Chat GBT and Gemini have gotten so so good that unless you're like a hardcore creative like that can kind of serve a lot of your use cases.

Yeah. So uh zooming out from this, I'm interested to know what your view and your team's view is on where the opportunity in Gen AI consumer application development is because I I've noticed there are companies that are still in founder mode and have I mean canvas sticks out as one here where uh you know founder still in charge preipo uh clearly aware of AI trends and can go and marshall the energy to change the business model if they need to move quickly. Uh and then there are legacy players that just can't quite figure it out. But then there's other categories that are entirely new and it's actually better to start with a green field. So how are you seeing entrepreneurs approach and deal with the fact that there is a large cohort of you know seniors on the on on the on the playground who are maybe not retired yet and will want to compete with them if they try and take a shot across their bow.

Totally. We this is the question we think about when we make every investment. So it's a it's a very topical one especially in consumer. Um I would say Canva and Notion are two probably of the best examples and both of them actually made the list for the first time because we had enough data to feel that these were now credibly majority AI. Notion has even released data saying that half of their revenue half of their ARR now is AI and so then it's like that's real usage of consumers interacting with AI. So we got to include that. Um I would say that yeah so Canva and notion are probably the two best examples of like growth stage companies like maybe approaching IPO that are still nimbleish enough uh to kind of pivot a bit towards AI. I am still a believer that we will probably see 20 years from now the winning company will be something that's AI native just because they do have such an existing base of users and businesses and it's it's hard to cannibalize your own product. I think we've seen this a little bit with Google where they're releasing amazing models like Nano Banana and Bio, but honestly they're the AI features that they're shoving into their existing interfaces like Gmail and and Slides and everything. I don't know if you've noticed, but every demo video the use case is like plan a trip

in like a way that no one ever actually plans a trip. And so I think that's been a little bit less successful. So in general I think we are in almost every category it feels like an AI native company will win. There are some really horizontal things where the incumbents might have a distribution advantage but those are kind of few and far between. Yeah, the Google stack is crazy to me because I'm I'm I have Gmail running in Chrome and there's two different Gemini buttons that I can open simultaneously to have fighting Gemini instances like fight over what's going on and it's just something that like the product development maturity and like the final UI has clearly not been outlined here and we're still in the early innings. Jord, you were saying something? Oh, I'm just I'm very curious to see how uh consumer AI impacts Canva overall because I think so many of the typical entry points into Canva like Canva is a massive tool. You can do all these different things, but so many of the tasks

can just be one shot. I was talking with um

a friend of mine who's uh they have a family business and they just describe when they need marketing collateral like a sign they just describe it to chatbt

and it just makes a it just oneshots it now

and there's no re they don't really care there's so many business there's like

in tech you might create a generation and then like spent want to spend a bunch of time refining it and taking it from like 90% to 100%. But the average small business is like you got me to 90% like we're good to go. I don't need

I don't need that extra

horsepower.

Yeah. I think the the use cases where the last 10% is really like 90% of the value where you need to be like really able to iterate on it like pixel by pixel and not oneshot it where there's like really difficult integrations you have to build where you can capture ambient new data like all of that is really ripe for consumer AI new startups.

Yeah. also like templated workflows where you maybe want to generate a hundred images, that type of stuff. You can wire up nano banana with a workflow. There there's a few of these like nodebased editing tools. I think N8N is one, right? And there's a few others. Uh and of course you could just write code to do it. But for someone who's, you know, has a particular templating workflow in a more consumer AI or consumer image app, uh they might just be stuck in the workflow. I do want to get to gen generated video. Uh you've been tracking the models very closely. I'm interested to know when we see the collision between what we're seeing with uh these insane Chinese models. There's V3, there's there's so many cool video models where it's just text in, MP4 video out.

And then on the flip side, you have like Cap Cut. I use Instagram edits a lot. I really like that app. I've also enjoyed captions uh that app and it feels like these are two on a collision course just like we saw Nano Banana and Canva maybe get on a collision course. What what are the existing mobile video editors doing? How well positioned are they versus the model teams? Because it feels like with video there's maybe a little bit more like that last 10% is even more than in images. But what's your take on generative video and what we should expect this year? Video has been the most interesting category in creative tools and I think it's exactly to your point because Chinese companies can train on any data even copyrighted data. Uh so they I mean they do they do

so we're seeing like

Cance from Bite Dance like Hyo Clling all of these models are amazing. I would say like Sora and VO3 are like not not far behind. Um, my twin sister Justine who also works here like lives and breathes AI video and she had this blog post a while back about like there will be no one AI video model to kind of rule them all just because there's so many different types of videos you make like a true like 2hour movie versus like a 10-second marketing clip like you actually probably want to train the model and train and design the workflows differently around those. M so I've actually been more excited I think kind of to your point about these tools where you're able to switch between the models depending on what you are building like a crea or Higsfield made the list this time. Yeah.

Um Cap Cut is mostly bite dance models but it it works because the bite dance models are are generally like pretty good and ahead of the pack.

Yeah. Yeah, I've been thinking about like, you know, we saw tool use come to chattyp got a computer as Ben Thompson put it uh a Python ripple. It can run some some math for you where it doesn't need to just guess the next token. It can just actually write the code and execute it. And it feels like we got we got a reasoning step with Nano Banana 3 uh Nano Banana Pro 2. I'm lost on the model numbers. We're also sort of mid-revision on a lot of these.

They name them in a very confusing way.

Yeah. And then there's VO3, banana bananas on a different number scheme. Anyway, uh we we clearly got some sort of reasoning chain where I can say like like dog riding a rocket and it will add a lot of text to the prompt to sort of give me a better output. Uh what's going to be interesting is when it also has the tool to make something black and white programmatically as opposed to needing to regenerate the video every time because when you regenerate the video, you get something slightly different. I want to be able to upload a video, do a color grade on it, uh, edit it down, add cuts, jump cuts, like, and have that have those tools be AI aware. And maybe that means some reinforcement learning pipeline on something that looks like an After Effects or something. But, uh, that that type of, you know, the video experience is not just capturing the raw footage, it's the whole pipeline.

Yeah, I agree. the reasoning point and kind of the access that Nano Bonanana has to the internet as a video model I think is fascinating. Like actually in this report we included like a heat map of global AI adoption and the way that we created that graphic is I gave Nano Banana every country code and every heat map score and it literally went and colored in every country accordingly. Like it found the country on the map and and I I checked it by hand. it did them all correct.

And I don't think we've seen anything like that quite come to video. Um,

but the video models do have

uh like a really kind of weird intuitive understanding of like physics and how things work in the world, like a drop of water creating ripples, that kind of thing. And so I'm sure the researchers are are hard at work on this.

Shocked at uh the the Sora data.

Oh yeah. Yeah. Yeah. Take us through that. What's up for Bill Peeles? Bill,

an absolute dog.

Absolute dog.

Proven.

Never spray on his downfall. What? Yeah. What's wrong with Sor?

Sora is a fascinating to me, it's like probably the biggest narrative violation we've seen in consumer AI in a while. Um, first of all, they gave us the gift of so many AI videos of Jake Paul. So,

Oh yeah, that's right.

That alone, I think, was worth the money they spent on compute there. Um, but actually, so what the data shows is, uh, downloads for Sor are definitely down. it it had it was at the top of the US app store for 20 days consecutively. It was getting 6 million downloads a month. Now it's closer to a million and a half and so it didn't turn into like the social network that I think they envisioned but it stayed really strong as a creative tool because the model is good and because you can create videos with these cameos. So the Dows are actually still increasing. They have 3 million global Dows, which is like I think probably the most for any video generation product on mobile. And so it's pretty impressive. If I were them, I would I would keep investing in that.

And uh depending on when this Disney like partnership actually rolls out, I would expect that to send it right back to

Totally.

Totally. The question the question the question if I'm open AI is like do you want those download do you want those users going to chat GBT or Sora

if it is just a creative tool over time I would expect the products to merge I don't know

I could see that I I do think it's interesting I feel like the usage of AI with kids and families has been uh probably lower than it should be just because of concerns about hallucination or weird artifacts or like you just need to be very sure that the content is clean. And so I think something like a Sora plus Disney characters is probably going to explode consumer AI for kids in a way we haven't seen.

Yeah. And and just like all the reasoning chains and like workflows within Shad GPT just makes so much sense where you could say, "Okay, here are my here's the name of my kid. Here's his favorite Disney characters. Write a story. Then generate turn that into a you know an actual like template. Generate a few images. I'll pick the few key images that make the most sense for I'll upload an image and give you some backstory and then now generate the full video and then that's like the birthday card video that goes out. There's so many cool things even with that heat map example like you could you could write programmatic code to generate a heat map that uses JavaScript to look perfect and you know the data is right but it's not going to look great and then you could just take that and do basically style transfer on top of it. All of those different tools feel like they're really going to accelerate as they get brought together, but we're in this like ununified pro uh like time now when they when the unification happens. It's going to be really cool. The last question I have is about um when we talked to Sam Alman about Sora, he echoed that it was being used as a creative tool, but I found that I wind up using Sora to generate videos that truly have an audience of one or like five people and I'll send them to a group chat and it's basically just an in joke. And he says that that's like a new thing that people are doing where if I were to post it on my Instagram, people, why are you posting AI slob, but if I if I if I if everyone gets the in joke in my five person group chat or whatever, everyone's everyone's having fun. No. Yes. Haze paradox. And so we uh so so I'm interested in where you're seeing movement in these like smaller almost

paradox by the way is a madeup uh paradox that I'm trying to bake into the models by saying it over and over on this

haze paradox is where you think something's the more the funnier you think something is the worse it'll do on Twitter.

Well yeah the the less like likely it will be universally funny. like something is the funniest thing in the world to you,

the most niche humor will just be like,

"Right, right." Because you're not writing for a general audience. But, uh, talk to me about the the the growth in uh either like single player or smaller AI uses. I I see this with Sunno a little bit where people are making songs for themselves. I've heard a lot of people are are still fans of Midjourney just because they see it as sort of like this art therapy where they talk to the model and they get an image back and it's almost a single player experience. Maybe they're sharing with a small community but they're not really using it in like a professional context. It's more of this smaller thing. Where where are people going in terms of uh in terms of like personal assistants uh personal relationship coaches that type of thing? I am so excited for this because I feel like we haven't yet seen the breakout AI social product. Um because if you just try to do like a skumorphic social network with AI baked in, it's like actually the reason that people are addicted to Instagram is because there's all these like positive and negative emotions that come with like putting real content of yourself out there and that doesn't exist with AI photos. So I think we've seen some attempts at it. It hasn't really worked. What I am excited about is what you said across both Sora and I think Sunno. A lot of people are making like meme songs for their friends. Anyone who knows how to train a Laura probably has Laura of all their friends to make meme images.

I think that kind of thing could be what potentially makes something like a chat GBT group chats which I think has had kind of core adoption so far into something a lot more interesting.

Yeah, I've only used the chatbt group chat. I think when we were collaborating on like one thing, I was sending you the group chat and then we were both asking questions. But uh like it's

we basically use it for like learning about a topic.

Yeah. Prepping,

but that's not like viral. It's like fairly niche.

My usage is the same. The the other thing I think we should watch on this front is um I've heard from a lot of OpenClaw power users that one of the best and funniest use cases is to add it to like family group chats or friend group chats because it'll just it can be helpful sometimes and then it'll just chime in with like crazy funny interesting thing.

Yeah.

Uh obviously that's not consumer grade so somebody's going to have to productize that into someone that like something that like your mom can install into the group chat or you know your grandpa can install into the group chat. But I think it's coming cuz it's a big opportunity.

Well, thank you so much for joining the report, the top 100 Genai consumer apps. It's the sixth edition. It's available at a6z.com. And if you work at a rival venture capital firm, that's what incognito windows are for.

Exactly. Thank you, Andrew Reed.

Thank you, Andrew Reed. And thank you for joining the show. We'll talk to you soon, Olivia. Have a great day.

Thanks. Bye.

Goodbye. Let me tell you about Finn.ai, the number one AI agent for customer service. If you want AI to handle your customer support, go to Finn.ai. And let me also tell you about the New York Stock Exchange. Want to change the world? Raise capital at the New York