Top 100 Consumer AI Apps, Explained

The most interesting new trend is that of personal agents. We’re in a world excited by what ChatGPT achieved by turning everything into conversation, and now we can add these agents to our messaging apps to make them even more conversational. This is an incredible paradigm.

Approximately 50% of Americans say they use AI. Around 4.5% of U.S. consumers pay for a subscription to an AI product. The wealthiest 1% of users spend $903 per month on AI using their personal credit cards.

The reality is that people are now creating software. They’re creating truly comprehensive products. They may use their agents to write some of that code, but when they offer this entire experience to a user, it’s much more than just a simple model fix.

What does the execution rate that OpenAI has achieved through advertising tell us? Previously, it would take a company years and years and years to reach it, even after launching an advertising product.

I’m Elena Burger, and I’m joined by Olivia Moore, author of our “Top 100 AI Apps for Consumers” report, and Josh Elman from our consumer team. Today, we delve into the 7th edition of this report: who pays for AI, why are the most important assistants heading in different directions, and can personal agents turn everyday tasks into the next big consumer business? Team, thank you so much for joining us.

Thank you so much for inviting us.

Olivia, let’s get straight to the point. I’d love to hear your main conclusions from this report.

This is the 7th time we’ve made this list. The 1st time, we ranked all global websites that were AI-native based solely on traffic, such as website visits. That worked because the way almost everyone used AI was as a dialog box product in a browser.

We’ve had to evolve that a lot in recent years as AI has expanded, so this list is the most extensive to date. We added revenue data for the 1st time. We worked with the data to specifically obtain card spending, without taking into account companies or small and midsize businesses. We classify products according to what consumers actually pay for.

For me, the most important conclusion of this report is based on traffic. It seems that many of these products are becoming established. We only had 11 new products on the combined web-and-mobile list. That’s the lowest number we’ve had in any previous edition of the series.

But the spending data really introduced a lot of new variables. Of the 50 products ranked by spending, 29 were not on any of the traffic lists. I think that also contributes to, or relates to, the other big conclusion of the report: how much consumer AI is a power-user game right now.

Yes, that’s kind of funny. We also recently published our “State of the Markets” report, and something Sarah Wang directly commented on was, “Listen, there’s a power-law game going on right now. It’s happening at the enterprise level and at the consumer level.”

One thing that’s probably worth noting is that we have the fewest newcomers on this list, and it seems like the space is becoming stagnant. But I think things have improved a bit more in recent months, and the most interesting new trend is that of personal agents.

What are we seeing? What are the big changes that are taking place?

It’s curious. The last report we did was finished just before the Open Claw traffic entry deadline. We did a sort of retrospective before publishing the report and found that it would have ranked very highly. This time, OpenClaw is nowhere in the rankings because its traffic has completely decreased.

As I understand it, the team was acquired by OpenAI, so they’re probably working on agent-assistant products there. But I also think it has been surpassed by other attempts at prosumer agents, such as Crock-Bot. Town has some fantastic agents. Now, OpenAI dots, which is new.

On the consumer side, we’re finally seeing things that I would feel comfortable giving to a family member and telling them they can use safely. Most of these consumer assistants have emerged in just the last 3 weeks or month, so many of them weren’t yet appearing in the rankings. But we extracted the data to see what’s happening with them, and it’s been very, very rapid growth. It hasn’t quite reached mainstream users yet, I would say.

I think we’re in a very interesting area that changed at the end of last year and the beginning of this year with OpenClaw. AI went from being something we used as a tool to improve our productivity to something that would actually help us get things done faster.

I think OpenClaw was an incredible pioneer in this, and I know they’re still working hard through their open-source foundation to continue moving it forward. But it really made a lot of people wonder, “Wait, what could these products actually be?”

We saw people buying Mac minis. We saw people managing all their data with their own local OpenClaw, and there was a large influx of people. It was truly inspiring to feel this new energy around what’s possible.

As with most trends, once the spark ignites, it takes time to create the 1st products, time for them to gain momentum, and time for them to be launched on the market for users to see. In a way, this report shows what we talk about a lot on our podcast in Silicon Valley, but it really represents very well what’s happening in the world in general.

I think the truth is that AI is really popping up everywhere. Even if there aren’t many new participants, all the participants are getting bigger.

I think the biggest developments when we talk about pure consumer assistants have been Muse and Instinct. There are many other fantastic products that we listed in the report that have also done this. Poke was one of the early pioneers.

Tomo has a very large audience and is doing incredibly well. Instinct had announced 100,000 users, growing by 10% day over day. They announced that 40% of users connected a credit card within the 1st 3 weeks and spent more than $1,000 on average in their 1st month.

Muse is interesting because, in many ways, it was a very successful launch, especially within the tech community. I believe the figure was 500,000 downloads and 250,000 active users in the 1st 12 days or so. But if you compare it to Threads, which was another big, recent, eye-catching release, it’s still insignificant in comparison in terms of appeal to the average person who doesn’t work in technology.

If we look at the downloads, Muse is still only available in the U.S. and Canada, so we only took the Threads downloads from the U.S. and Canada. In the 1st 22 days or so, Threads had around 16 million downloads, and Muse was still around 5 million in that same period of time.

Is the distribution mechanism similar to Threads? Threads was already linked to your Instagram account, and there was a very natural distribution. Are they trying a similar strategy with Muse, or is it simply a different distribution strategy?

I’d be curious to know what Josh thinks, but from my point of view, it’s not that aggressive, that’s for sure. Part of that is also because there’s a much higher cost to serve these users, so they don’t want it to grow too fast.

I also think that Threads had a great critical-mass effect: if you didn’t get a lot of people communicating quickly, you didn’t really get to take advantage of that new network starting to form.

Muse remains such a personal tool. They’re pushing it, and it’s very exciting to see even this early adoption and for people to have these experiences. Again, if you use Open Clone January, you’re having amazing experiences. When Grok Bot was first launched earlier this summer or spring, you were having these experiences.

Now we’re seeing, as Olivia said, our mothers, our families—people who don’t just live off technology—asking what’s next in the AI cycle. I think that’s what’s really exciting about this consumer space. I expect that when we do this again next time, we’ll see mass adoption.

I don’t think it has been necessary to put in so much effort to have that early network effect. But I also think about what will make these things last for a long time. It will be things like network effects, marketplaces, and all the businesses that you can interact with through those agents. That will also be truly transformative.

This is one of the things we talk about a lot when it comes to consumer agents and assistants: they are getting the cumulative effects of building a platform, or at least it seems that way, in terms of the ecosystem of other applications that are available and that the agent can more easily use.

We’ve seen that it works both ways. For example, Amazon says it won’t allow Muse to browse and buy products on its site, but Muse already has hundreds of other partnerships with Shopify to be able to buy things on Shopify.

1. The privacy wall for personal agents

Exactly. Yes. Well, I still don’t feel that any of the assistant products have unlocked person-to-person network effects. Part of this, for me, is that this software knows us more intimately than any other product in the past. It knows us based on our email, but I also find myself telling my AI extremely personal things.

There’s a natural tension between the more useful the product is to you as a consumer and the more it knows about you. Do you really want to introduce that in other contexts with other people? It’s the same as many people having a separate work GPT chat account and a separate personal account. Nobody has managed to decipher that yet.

No, and I think that, as we talk about consumers and AI, privacy, security, and protection are becoming increasingly important.

Obviously, there are general trends of trust in AI and this kind of belief that AI is good for the world that, in some way, affects adoption and enthusiasm around these products. But on a very personal level, if you don't trust something you're giving more intimate access to—your email, your life, or your credit card—and it might do unreliable things you wouldn't expect or share things with other people you wouldn't expect, that will be the biggest obstacle to these things growing into mass-market applications.

So we are in a fascinating time where we are all learning these new rules. And these are rules that, as people, we are very good at. We are very good at knowing what information we feel comfortable sharing, what information, if someone gives it to us, is entrusted to us to pass on to someone, or what we should protect and keep for ourselves. And suddenly, we expect the software to do this too. It's a collection of new and interesting things that we're learning.

Yes. Well, they both just mentioned 2 interesting aspects of this. One is that there is this endogenous force—all the platforms and how they allow agents to interact with them. And then there's this exogenous force: how comfortable do I feel using this? Perhaps I confused endogenous and exogenous. One is outside and the other is inside.

We knew what you meant. One is what the platforms do, and the other is how you feel about it or how you feel about the agent. And I'm curious to know which one you think will be resolved first and whose responsibility it is. Whose responsibility is it to answer those kinds of questions?

I believe that platforms are capable of improving faster than the actual willingness of consumers to use them, something we've seen so far in AI. It's interesting because many people see these figures published by the founders of these assistant products, who say that it costs hundreds or thousands of dollars a month to support a user, and they wonder how that's possible. I was perplexed too, but I've been working with David Pulan, who runs Assistant Benchmark, which is fantastic.

She recently did a podcast with him.

Yes, it's great.

Basically, what he's done is create a website where they track, I think, more than 170 agents. I checked it last night; it's probably around 200-something this morning. It has users use them and then rate them according to their success in different tasks.

But the other really interesting thing it does is organize a kind of group chat and community for the first users of these agents. It has a group of over 1,500 users, I think. And it has provided us with some interesting information, such as: What do they do with the agents? What do they say they do with the agents?

The vast majority—the number-one use case—was coding and technical automation, even for these consumer assistant products like Music Instinct.

Correct.

And then, within that framework, it's like, "Oh, of course it costs that much." This is like when we talk about some agent or assistant products that are aimed at the more conventional consumer; they're looking at costs of tens of dollars, not thousands of dollars a month.

So I think that's perhaps a representation of the fact that we have a long way to go in terms of consumers feeling comfortable with these products. And I think the people who have been coding and working with AI have found it to be an incredible enhancer, and that's reflected in a lot of the data. You're in the top 100 here; it's incredible.

Those of us who have been able to do that have had these great ideas for all the other things in our lives. But when you simply give this to someone new, it feels like a blank box. You think, "What should I do about this?" There may be examples of how to get a reservation or book a flight, but if you're not actively looking for a reservation or booking a flight, it's difficult to figure out how to customize it.

A big part of what we do as people is learn from others and get these great ideas and try other things that people tell us. A big part of how social media has worked is that people post things on social media, other people see them and try the same thing, and then it creates these effects where I try to make a small modification. Those modifications become cascading snowballs, and that's how great ideas bloom.

With AI, because so much of it is so personal, we haven't seen any of that flourish yet outside of areas like coding and productivity—all the creative work. Someone says, "I created this amazing video. How did you do it?" "I used AI."

Well, there are many tools that do that, that are growing and are already here. But it is very interesting that we now need to figure out how to generalize this—not for the millions and millions of people who are already using them and paying for everything, but for the tens, hundreds, and hundreds and hundreds of millions of additional people.

2. 4.5% pay, $903/month at the top

Looking at current usage data, nearly half of Americans report using AI. And around 25% of Americans believe they interact with AI probably almost daily. But spending data is much more concentrated. Therefore, it depends on the source: our Yipit panel shows that probably less than 5%—about 4.5%—of consumers in the U.S. pay for AI.

Other sources indicate that it is only 2% or 2.5%, or perhaps a little higher. It's growing, so it's roughly double what it was a year ago. But that spending remains extremely concentrated. Even within that top 4.5% of spenders, the top 10% of them account for more than half of the revenue. The top 1% accounts for 20% of the revenue, while the bottom 50% accounts for 16%, right? So the top 10% of users are driving this market.

Another interesting aspect of the revenue data is that we were able to analyze it almost person by person, observing what it looks like and exactly who these people are who are investing in AI. The top 1% of users spend $903 per month on personal credit cards for AI. And the median expenditure is $25.

Again, all this is limited to only 4.5% of those who make some kind of expenditure. However, it was also useful to us because we were able to see what products they actually spent money on at a personal level. Unsurprisingly, most of them were developer tools, productivity tools, creative tools—things used to create, manufacture, and sell. Products like n8n, Granola, Higgsfield, and Manifold were all overrepresented among the biggest spenders.

I mean, in a way, this is really cool. Even though these numbers still don't encompass everyone, we're making room for a whole new generation of creators: people who don't just use this for work, but use their personal credit cards to program things they've never programmed before, to make art, videos, and content. They really have this creative power.

In a way, even though it's still very early, it's great to see so much being created—that people are so excited about it that they're willing to spend that much. If you spend $900 a month on a suite of AI tools, you're actually using it to, hopefully, get things done faster and better and bring more ideas to life in your head than ever before. It's similar to when I was analyzing the categories that exist between creativity and coding, something that, hopefully, turns out to be extremely powerful.

3. Muse and Instinct's launch numbers

Yes. Yes. Well, I think this is a very important point, and that is that developers and coders are people who are naturally attuned to the idea of how to take advantage of this. How can I create a product that allows me to make the most of my time? AI, obviously, enhances that.

But I think it's a specific kind of mindset and a specific worldview. Part of my understanding of these figures is that it remains a worldview that is reflected only among a small group of people. The question might be, how do we broaden that worldview? How do we help people understand all the ways they can make the most of their time and their lives?

We talked a little about this in this report in terms of areas where we haven't yet seen native AI startups emerge, or categories where we haven't seen them. Eugenia, one of the directors of our portfolio, runs a company called Wabi, which is fantastic. She has a great date. I'll probably mess it up, but it's something like: Most people aren't looking to save time; they're looking for ways to invest it.

That's why social media, entertainment, Netflix, TikTok, and all of these YouTube products are the most-used consumer products we have. I think a lot of what we've seen in AI so far, as you were saying, is, "How do I make this a little faster, a little easier, a little more impressive?" And that's not an incredibly compelling daily or hourly active value proposition for most people.

No, I think saving time is still very much about hyper-productivity—people focusing on working and being creative. But I'm also seeing it in this new wave of people who maybe were used to programming but aren't anymore. Perhaps people who want to make videos but never mastered Pro Tools, Adobe Premiere, or something else.

They can suddenly sign up for these things and take the ideas in their heads and make them a reality. So I think we're starting to see this increase.

And again, as you know, I look at all these figures because it's still very early.

Yes.

And the fact that we're here and talking about how big some of these companies are, and that 100% of people spend $900 a month—I see this as a harbinger of things to come as more people feel empowered, as more people feel they can create.

It's not necessarily about saving time. It's about the joy of wanting to make this a reality. There were all those jokes about when people were in that peak phase where everyone was using their OpenClaw for the first time.

The agents were walking around with their computers and laptops. They had to keep them open, and they were saying, “I’m not sleeping. I’m programming. I’m doing things—more than I’ve ever done.” I think that’s starting to show up at the top of the data, but it’s also starting to show up everywhere.

4. Why subscription is the wrong model

On this list, I don’t necessarily want to see the 4.5% of people who pay for AI products expand directly. That may be controversial because I’m invested in consumer AI products, and I’m a consumer AI maximalist. But the reason I say that is because virtually all consumer AI revenue so far has come from direct subscriptions.

In the list we observed, about 85% of the websites on our list monetize through subscriptions, and another 62% monetize through credits or token payments for additional use. Only 13% had ads or other options where you are the product instead of paying for the product.

If you look back at the history of the consumer internet, it's a pretty unnatural investment. Almost all of the large consumer technology companies we have now get the vast majority of their revenue from advertising or transaction fees, not subscriptions.

The truth is that not everyone is looking for a coding tool to help them start a business or create and sell a product. Most people don’t have the funds, or don’t want to spend them on software. If we look at the top 20 consumer subscription products globally, ChatGPT is already there, which is crazy after about 3.5 years of growth.

Almost all the others are media companies or large platforms like Amazon or even Uber—things where you have very frequent purchasing behavior. So I firmly believe that we need to move forward. We’ve moved beyond the need for everyone to buy an AI subscription, and we need to see how these other business models make a comeback.

Yes, I think we are still in the early stages of reducing inference costs. We’re still in the early stages of people finding ways to run lower-cost models to create great products. As they do, many of the other business models that have always been on the internet, such as transaction fees and advertising, will finally start to work.

But we need that to happen, and I think that’s where we still are. There’s a lot of learning to do—so much so that, early on, the default became, “Well, if it’s going to cost me a lot to provide the service, at least let me charge money so people will pay for it.”

You can probably see this by looking at many previous eras of the consumer internet. At least for me as a consumer investor, if a company made money in the first 5 years, it was like, “Wow. What’s going on?”

The rule was, “Let’s build user density.” Since the cost of providing the service was almost zero, we could wait and make money through ads or start charging for transactions once we had liquidity and density. that's not the case given how high the gears are.

No, I think this issue of costs is a very interesting challenge, and companies are afraid of growing too fast. If they grow too quickly and don’t have metrics for how they charge for it, they can have uncontrolled spending that can be very, very difficult to manage.

Therefore, it has created a very different picture from what we see in the top 100 apps, in terms of traffic and everything right now. But again, it’s still too early. There’s much to build and learn—to reduce costs, increase value, and introduce advertising and other ways of monetizing.

And, by the way, if we subscribe to these services that we love and get ongoing value from, that’s amazing too.

5. OpenAI's $1B ad run rate

I think this is a great transition to talk about ads.

Yes. The report on the top 100 consumer apps included a very interesting piece of data: the execution rate that OpenAI has achieved through advertising. They now have an annual advertising expenditure of one billion dollars.

What does that tell us? I would emphasize that the figure is from August, so it’s likely to be considerably higher even a month later. They’ve been rolling out announcements very slowly, using a sort of partner network, and they already have an annual expenditure of one billion. Billions.

Previously, it would take a company years and years to reach that figure, even after launching an advertising product. I think there are 2 interesting things. First, they now have real user density. I believe the latest figure, reported a few days ago, was 1.2 billion weekly active users. That’s real density.

6. Inside the 7th edition

The second thing, and what fascinates me about the evolution of the advertising product, is that hypothetically they should be able to charge more and have higher conversion rates because they know much more about you as a user. ChatGPT has launched things like ChatGPT Login, and I think there will soon be a wallet and things like that.

You can imagine that it becomes a much more attractive value proposition for advertisers, and that this further boosts the advertising business. But we’ll see. It’s still early days, and there have been some funny conversations about ads and AI.

If you talk about something so personal and say, “Let me talk to you about health and introduce you to this crazy product that can help you lose weight fast,” that doesn’t build trust with AI. Another interesting aspect is how we, as an ecosystem, manage these products, which are so different and much more personal than ever.

They offer personalized answers based on what you’re doing, what you’re asking at that moment, and everything they know about you. In fact, the rollout of ads has been done very skillfully. I’ve been very impressed by OpenAI’s implementation of ads.

When I see an ad, it’s very clearly labeled and feels like a natural fit in the conversation I’m having. I think we’re also seeing another change: people are starting to rely much more on ChatGPT for many more queries.

Before, it was more like looking up a fact, asking it to tell me the story behind something, or helping me with my homework—not necessarily solving problems or anything like that. Now people turn to it for real advice, to buy things, or to figure out exactly where they want to go or what they want to do.

Those are great opportunities when you’re in discovery mode, where ads are often a great answer to help you figure things out. Many more of these business queries and this commercial intent also create opportunities. But it has to be done with a lot of skill, not through a strange, disruptive, and very personal experience where ads are introduced into the conversation.

I would be very curious to know how the unit economics work in advertising. Presumably, it's a matter of lower or lesser cost compared to, you know, encoding or something like that or image generation. For OpenAI, what kind of margins do they see in the ads?

I’m curious too, Josh. What kind of ads do you see? I don’t receive any ads. I assume I’m using the business version, which is business-oriented.

I was doing some research on travel recently. I was going to be in New York for a few days, and I started seeing ads for a couple of things happening in New York that I might have wanted to do. I found it very useful.

I started using it more for business purchases. I was looking for a special key ring that could be easily detached for a valet parking key, and I was having trouble finding one on Google. Finally, I thought, “Let me ask ChatGPT.”

I went back and forth, describing what I wanted, and it started giving me some really good examples. In the end, I thought, “Great. I’d also love one that’s made in the United States.”

I found this amazing option from a factory in Detroit that does many other metalwork jobs. They had a key ring that was really great for me. If I had never had the opportunity to exchange ideas and really have that conversation with ChatGPT, it would have been different.

There were no ads presented in that conversation, but I was completely open to ads that would have helped me find things. I went to the company’s website, and I had never heard of it before. It was a great experience.

Very interesting. I think that, as OpenAI delves deeper into this, it will be able to do incredibly good segmentation, even better than Meta, which used to be best in class for this simply because it knew so much about you. It’s like you’re having this conversation live with them.

Yes. I think it will help ChatGPT, but I also think it will support many of these more focused, almost vertically integrated consumer AI products, to use the enterprise term.

Another company that has had early success in advertising is OpenEvidence, which is an AI product for doctors. It already has 50%—I think the most recent figures were between 50% and 60%—density of all American doctors on the product.

When you have that kind of density and a really valuable audience, it’s a segmented audience because you know they’re doctors. You can advertise very effectively, and there will be—and should be—more companies like that.

7. ChatGPT vs Claude vs Gemini

Is it worth taking a moment to talk about OpenAI, Anthropic, and Gemini? In a sense, these are the companies, these labs, that have been on the list practically since the beginning of this whole process.

Are we seeing any kind of evolution in the way people pay for these products? What can you infer about the different types of users of the 3? I’m very curious to know what you’re seeing.

Yes, of course. From the first time we did this report, the main conclusion was that ChatGPT is the dominant global player in both usage and revenue on the consumer side. That remains true. Everything below ChatGPT has experienced, I believe, more movement and more change. ChatGPT now surpasses Claude by about 6 times on the web and Gemini by about 2 times on the same platform.

Probably in the future, given that much of the usage is concentrated among these advanced paying users, revenue figures could be even more useful here. This is probably one of the biggest surprises for me in this report: We have a panel of U.S. consumers, and Claude has overtaken Gemini in terms of the number of paid subscribers, which is crazy given that Gemini has a much larger installed base and user base overall, as well as the natural distribution mechanism of something like Google.

I think it's a great achievement for Anthropic. I think it has come after a wave of very successful products they have launched in recent months, such as the Claude design. It has also come out of much more press coverage than they've had, starting with the War Department, I think it was in February.

The other big difference, I would say, is in how these products are monetized. Again, ChatGPT is still far ahead in terms of monetization. They have about 3 times more paid subscribers in the U.S. than Gemini and Claude. But Anthropic has, as is well known, said no to advertising, so they are much more aggressive with subscriptions.

They have around 7.5% of subscribers on the maximum plan of more than $100 per month. That's about 1% for both ChatGPT and Gemini, so those users look different and do different things in the product, even within that paying user base.

Should we talk about creative tools? I think one interesting thing is that all the big labs have become really good at creative tools. One thing that Anthropic has refrained from, but somehow manages to do anyway, is creating an image or video model. Now people are creating images and videos because coding is very good for creating images and videos.

It's interesting that, as you know, they haven't created a distribution model, but they're still able to generate images. But in terms of pure creative tools, what are we seeing from companies that have vertically integrated into that category?

8. Where creative tools still win

The distribution of creative tools is very interesting, depending on who succeeds and where. There are some categories of creative tools—or perhaps modalities would be a more accurate term—in which the labs have not focused their energy or attention for many reasons that we could analyze. Many of them deal with audio, such as text-to-speech or music.

ElevenLabs and Suno are now very close to the top of our traffic lists and also rank very high on our spending list. The labs have a lot to do. They are creating coding agents, and they are creating AGI. Does it make sense at this point for them to deal with all the intellectual property headaches that Suno has had to go through and launch a competitive music model? I don't know. Meanwhile, Suno has managed to come out on top with a really solid lead.

Yes, and I am firmly convinced that the labs will follow their own path and continue to grow, as they are evidently excellent benchmarks for everything. But in many of these markets, you see really specialized products that are very focused, that know who their audience is and what they are doing, and that are able to create really personalized things.

Whether you want to make music just for fun or for something you're actually producing, Suno is an amazing tool that speaks to you from the moment you start using it until the very end. ElevenLabs, if you need voice and audio within the product you are creating, has a way of working and interacting with it that is completely different.

When you look at a lot of video tools, and when you look at design tools, even something like Figma, they're designed to be used deeply for what you're trying to express yourself with. They speak to designers, not to people who want to skip that step. I think we're going to see much more value from these truly customized interfaces, where the software layer—all the things you can do, the things you save, and the way you interact with them—becomes where the real value lies, not just the models.

Yes. Well, I think this is the second category of creative tools, which are images and video. That looks different from audio. Regarding images, I think, as you said, OpenAI has really leaned into images with considerable success. Images 2.0 is very, very good for both creating and editing images.

Then, of course, Google has the various versions of Nano, Banana, and Bio. I think they've frankly eliminated a large portion of the high-end consumer traffic that used to go to these independent image generators. For example, Midjourney was quite high in our early rankings, but it has completely fallen out of the traffic rankings.

But on your point about power users, if you look at the revenue rankings, Midjourney is back. Therefore, advanced users still want the model they consider to have taste and aesthetic sense, and that they can adjust in a very specific way.

That brings us to video, which for me is like a wild frontier right now, because Chinese companies that can train with any data have a real advantage. Video is so complex to generate, with visual and sound elements synchronized visually and audibly, and it looks so specific that I think the more data you have to train with, the better. So far, they have been successful in that respect.

More generally speaking, in creative tools, I'd like to know if any of you have experienced this, but recently I've been asking Astra and some other really powerful models to generate or edit creative material for me. They do it very, very well, and it will be used in other places on the internet. Therefore, I believe that creative tools that are easy to use and accessible to agents will increasingly see very attractive tailwinds.

Yes. It's crazy what you can do if you simply give power to an agent and tell it, "Yes, install whatever you want to install." I had Astra and 5.5 competing head-to-head to animate some things for me, and it was like it knew exactly what to do.

What I love about this is that handy people, the people who are at the forefront, are discovering these capabilities. We're going to create products over the next year that will appear on these lists, because they're going to take all this complexity of what you can do and package it in a way that's really easy to use.

When I think about this gap between general AI and consumer AI, it's exactly this move to package all of this and create great product experiences that really makes it into these things that endure.

Yes. I guess it has to start with those who do experiments and then publish on X, and 6 months later we get a real product and a startup from it—or 12 different startups competing against each other.

Outside the web, there has been a great proliferation of mature products. There's Granola, there's WhisperFlow, and there's Superhuman. What do you see there? And then, obviously, those that exist in chat and messaging—so, Instinct[?]. What do you see in terms of distribution, and what do you see in terms of the types of products and users they target?

I think there are 2 interesting threads to discuss here. One, I wrote an article called “The Great Expansion” about 6 or 9 months ago, which is basically about why consumer companies are becoming businesses so rapidly. Products like Creasia, Replit, 11 Labs, and Gamma are examples of this. These products, like Canva, took more than 6 years to generate real revenue for the team and the company.

What's happening is that they're very successful in growing through PLG. WhisperFlow and Grainola are other great examples. Then, because they are related to labor productivity, they are incorporated into companies, and the founders think, "We have to add privacy, security, and an enterprise plan." But it is a very efficient and very low-cost distribution. Again, it works very well for these tools that can be used personally but are kind of adjacent to work.

I think things like WhisperFlow, Grainola, and Superhuman are another good example of this trend we've seen in general, where traditional operators don't want to or can't cannibalize their existing interfaces. The classic example of this is Google. We haven't seen them reinvent Docs, Gmail, or Calendar for the AI era.

I think it's even happening with OpenAI and Anthropic, which have had far less success with anything other than a release within their existing interfaces: ChatGPT, Codex, Claude, or Claude Code. Many of them have tried to incubate revolutionary projects for many of these startups, but I think it's difficult for them to innovate beyond that from a product perspective, especially when they have so many other things on their hands.

I think all of that is true, and I think that now that we're getting used to some interfaces, we're still in a period of incredible reinvention. A big part of the reason we've seen so much growth in the last year or year and a half is that the models have improved a lot.

The people who had dedicated their time to creating these amazing interfaces that were easy to use, felt natural, were easy to share, and were ideal for engaging your team—considering that WhisperFlow and Grainola were so simple to start with—would say, "My God, now they have improved so much as the models have progressed that the usefulness of the product has also improved in a significant way."

Then we thought, "Incredible, you create something great, but it may not do everything you want because the model's capabilities weren't there." The models improve, and your product explodes in demand and utility.

This kind of rising tide raises everything. To think that we only have 1 or 2 companies that can create that layer of expertise, I think that's what this type of report discredits. Those are certainly some of the major companies, and these are very general use cases, but the reason so many other things are growing is that there are many interfaces being reinvented, and the rising tide elevates everything.

The value of the infrastructure can continue to go to the model providers, or it can go to other things that serve inference, and there's a completely different conversation there. But what consumers are adopting are things that work for them, are easy to use, and incorporate into their lives.

Perhaps, also, to continue defending the case for startups in this instance: OpenAI and Anthropic are startups in many ways, but they have been so successful that they have become large companies with thousands of people. There are many more considerations, such as the fact that a larger ship is more difficult to steer and turn around.

One of the companies that appears on our revenue list this time is Plaud, which is an AI note-taking device that started in China and the Middle East and has now arrived in the U.S. You can buy the device, and you can also buy a subscription. OpenAI can manufacture a hardware device, but as we've seen with them, everything takes longer when you're a large company. You have to integrate it with other existing products, which means more checks and balances.

Therefore, I believe that large labs will continue to be very successful and have much more to do, and I believe that startups will continue to find these great windows of opportunity.

It also seems that the question for startups is how to create a product that, as you said, Josh, improves as the models improve but doesn't cannibalize itself as the models improve. Which seems like an interesting sweet spot in which to exist.

Part of the reason I was so excited to join Andreessen Horowitz a couple of months ago was this belief that value is now really returning to the software layer. Now that we have these incredible models at our disposal, you can create truly comprehensive products that leverage those models to do so. And there are new innovations like Jeff from TypeSafe that allow developers to create things in even newer ways.

We haven't even seen all the amazing new ideas that can be created when you have access to a model like that—one that's fast, easy to integrate, and gives you highly structured code or answers. That's why creativity now returns to the experience: the network you can build, the ways you can serve many people and unite them using models to make something powerful. Context and community really become the asset.

There's so much invention that I think almost every category has the opportunity to reinvent itself. Some large companies will be able to make the leap and become the relevant company for their sector in the age of AI, but not all of them will. That's where startups come in, getting ahead of them and creating a truly new possible experience.

I think a lot about the accumulation of personal value and, hopefully, the accumulation of multi-value that becomes something of a bond with some of these startup products. An example I would give is Town, which is more on the prosumer end of the agent-assistant spectrum. It can connect to all your systems—Slack, email, Google Drive—understand your context and your organization's context, and then do things for you.

I use it to write many of my emails, and it helped me with a large part of this report. I know it's easy for people to dismiss things like this and say, “Oh, it's just a wrapper,” which is now a slightly less derogatory term for a container, I suppose. But Town has created these manuals of who I am, how I sound that no other product has, and that I can't easily take and migrate to something that comes out tomorrow.

The difference between an email that sounds 99.9% like me and even an email that sounds 85% like me is the difference between spending 10 seconds fixing it and more than 10 minutes. That's why I think we'll see more products like Town that do a great job of gathering context and acting on that context to create playbooks around you, and that will continue to be successful.

I think that's absolutely correct. I try to avoid using words like “harness” and “container” because I think the reality is that now people are creating software and really complete products. They may use their agents to write some of that code, but when they offer this whole experience to a user, it's much more than a simple insertion into the model or a harness that provides context to the model.

It's actually a really complete product and experience that uses a model as part of what allows it to be offered. I think that's really the excitement of where it's going to take us and what we're going to see on this list. Many of them are already doing it, and we believe that it's truly the next wave. They're all going to be companies that have created much more value, and the model is just one piece.

9. The white space: dating, shopping, entertainment

Olivia, I think you've referenced this chart you made, but could we mention it? All the blank space that still exists for all the product categories that existed in Web2 and in the pre-AI era? Maybe you can talk about some of these categories and where the big opportunities are.

The vast majority of pure consumers still use AI primarily as a replacement for search products, things like Google. Perhaps especially in educational use cases, children already use it to write their essays. A slightly broader use case.

Help with homework, exactly.

Yes. Or people in the workplace writing emails. But if you look at this chart, which shows almost every area where we've seen startups succeed so far, it's been around that theme of getting things done, such as productivity, design and editing of photos and videos, search, and answers. There are many open categories.

Many of those categories are network categories where you need a multiplayer product. Dating is one. We haven't seen any focused on dating. There are certainly startups there that I hope and believe will appear in future editions of the list, but there aren't any on the list currently focused on dating.

The hiring process is different. I think dating and recruiting are similar networks in terms of trying to make a match. As for social AI, we haven't seen it take off at all. Most social AI is people posting AI-generated content on existing social networks.

Then I think there are all these other categories that are more like markets, such as shopping, home buying, and retail. We'll have to see if they end up living on agents like Instinct or Muse, or living as independent products, or in the next wave, which will include many more. They might live as standalone products or perhaps as a mix of both. Maybe your agent will call it the AI native market. I'm really excited about those things.

I think that in categories like gaming and entertainment, the models simply haven't been good enough to produce content that the average person enjoys watching, with the exception of the AI microdramas that are exploding. But I think that will come soon too.

I think that's exactly right. We're in a world where we're excited about what ChatGPT has done to turn things into conversations, and now we can incorporate these agents into our messaging apps to make them even more conversational. It's an amazing paradigm, but when you're shopping, you want to see visual options, explore and customize things for yourself, and be able to say, “I really like it, but I wish it had some trim on the neck or something.”

AI has helped you realize that, and perhaps even have it tailored to your needs. Shopping can have a lot to offer. In the world of entertainment, I think we've only just begun to scratch the surface of the ability to tell better stories and empower people to be better storytellers, which creates these incredible networks of people who consume stories.

TikTok was an amazing place where people gathered and created short videos. You see people from all over the world doing that now, but there's still so much captured in people's heads that I wish they could express their ideas even better. AI will help us in that regard, and we will have an incredible new network.

Therefore, I believe we are at the peak of all these categories, and we needed the models to improve so that you could start to explore some of these ideas and invent them. But as that happens in the coming years, with the new capabilities of the models, I still think the value will ultimately lie in the software layer of the product.

For me, this brings us back to what we were talking about: products that help people save time instead of wasting it. Almost everything we've seen in consumer AI saves time, and that lost time usually ends up being among the largest companies, if not the largest. We need that to exist too, and I'm sure many developers are working on it right now.

Of course. I think it's a great note to end on. Thank you both for joining us, and don't forget to check out the 100 best consumer AI apps, seventh edition.

Thank you for inviting us.

See you at the eighth edition.