Dreamer：面向所有人的 Agent OS——David Singleton

Obviously, it's really cool to have software that will work on your behalf. But it's only useful if you can trust it. Privacy and security are very important to us, so as we started to think about this problem, we realized that we actually had to build something that's a bit like an operating system. The sidekick is like the kernel, and the agents and apps are like users.

Yeah, this is different rings.

Exactly. Because if you try to pick off just one piece of this, you can't actually make it work for people at scale. You could build little bite-coded apps, but they're going to grab all your data willy-nilly, and they won't be able to work together. You actually have to invest in the fundamental core in order to make it work well for people, and that's what we've been doing. It's been a lot of fun.

Okay, we're here in the studio with David Singleton. Welcome.

Hey, Swyxster. It's great to be here.

It's great to have you. We're very sympathetic to the fact that your company color is the same as Latent Space's color.

That's right.

Dreamer purple. It used to be Dev Agents, which I thought was very cool because you call back to Dev Payments. And you were obviously CTO at Stripe. Talk to me about the origin or the thinking process behind Dreamer, and then maybe start with: What is Dreamer?

Yeah. Dreamer is a new product that everyone can come and play with today. It's a place where everyone—literally everyone—can discover, build, enjoy, and use AI agents and agentic apps. We designed it for consumers, for folks who don't necessarily have any kind of technical background. It's really aimed at everyone.

I often think of my sister. She's very smart, but she's not in the slightest bit technical. She has lots of problems in her life that she'd like to be able to have great, intelligent software solve. But even with the rise of tools like Cloud Code and so forth, she's got no way to get started.

Dreamer is a place where she can come in, grab some intelligent apps that other people in the community have built, start using them right away, and solve real problems in her life. At the core, we have a personal agent called the sidekick. You can give your sidekick a name and its own personality, and it really helps you across your entire day and your life. It helps you use all of the agents on the platform, and it also helps you build anything you want.

We've been working on this for a little while, and we recently launched in beta. We have many, many people in the community now who are building really fun, really powerful, really useful agents and agentic apps for themselves.

I think we're going to go right into a demo, but I just want to make an observation: You put “discover” first, before “build.”

Mm-hmm.

But actually, at least with the engineers in the audience—we are primarily engineers—and you're primarily targeting consumers, right?

Yeah.

For engineers, there's a huge full stack of stuff, which we're going to dive into. It's so impressive. I'm like, “Holy shit. This is what I always wanted.”

Cool.

So I think that's really good. In some ways, given your background and given Hugo's—is it Hugo or Hugo?

Hugo Barra, yeah.

Hugo. It's not surprising that you basically built an app store for agents.

Yeah. Hugo is my co-founder. Hugo and I met with our other co-founder, Nicolas Jitkoff, in the very early days of Android at Google, where we were building Google's first mobile apps. We then contributed to very core pieces of Android itself.

You're right: We were really excited about building two things. First, solving a bunch of problems that this breakthrough technology—in this case, mobile—needed to have solved in order to make it work for real people at scale. And secondly, building this ecosystem of third-party developers using the Play Store, able to deliver way more value on the platform than we could have delivered on our own.

We think about Dreamer in exactly the same way. I was working at Stripe, as you mentioned, and we had the opportunity to put some of the very first AI agent systems in the world into production. From the moment we did the first of those, I was struck with a strong sense of conviction that this is breakthrough technology that's going to change how all of us work with computers and phones, and all of the technology in our lives.

But there are a lot of problems to be solved for real people to be able to make this approachable. It really is a direct analog for what we were solving back in the early days of mobile apps at Google and on Android. It's been fun to bring that to life.

Yeah.

Let's look at it. Yeah, let's take a look. So, Dreamer.com, this is our homepage. This is where you can come and watch some videos of what is here and sign up for the wait list.

I just want to say, for those listening, switch to YouTube and look at the animations. There's so much care.

We really care about this product being fun and interesting to use. Obviously, a lot of people are using it to do real, important stuff. You can do real work here, but you can build fun things, too.

Once you get off our waitlist, you'll come into the product. The first thing that happens is you'll have a conversation with your sidekick, which is this little friendly character here. The sidekick will seek to get to know you and understand you: What do you care about? It will help you discover and build your first AI agents or agentic apps.

After that, you're going to have a dashboard. This is my dashboard; everyone's is different. You can see I have a few things here. I have a feed, so a lot of our agents do things in the background when you're not looking, and the feed is how they let you know what they've been up to.

I have some widgets from apps that I have built. This one is called Calendar Hero. It's something that I installed from the gallery, so it's built by someone in our community. It's a really powerful calendar app because, for each of my meetings, if it's with someone I don't already know well, it'll go off and research them. It gives me both a history of my interactions with those people and a bunch of useful public information to get started.

One of the things I love about this particular app is that every day it generates a podcast—a daily briefing. One of the things that we've done with the platform is make it possible for all the things that agents do to show up in places that you care about.

If you look over here, this is the screen of my phone. If I go ahead and open Apple Podcasts, you can see right here: “Your daily briefing podcast is ready.” This was produced by an agent running in my Dreamer account, and it was very easy, by scanning a QR code, to connect it to my Apple Podcasts. That's what I listen to in the car every morning on my way to work. It preps me for my day.

One of the immediate questions I asked you after seeing this was, “What if I want to talk back to my agents?” You said you actually started with voice and then went to podcasts because it's nice to have it pre-downloaded.

That's right. You can talk to your sidekick. On mobile, we have a Dreamer app, and you can talk to the sidekick right here. But we've actually found that making things show up in the other apps that you already use in your life is incredibly powerful.

Let's take a look at what's under the hood here. I already mentioned that we have a gallery. This is where you'll find a lot of agents from our community. There are hundreds at this point, and they're solving all kinds of use cases.

I'd say that the top use cases are personal productivity, but also a lot of information management. That can range from managing personal information, like documents and so forth, to managing your emails. It also ranges out to public information that you might be interested in but need something to help manage—the firehose of stuff that's coming at you.

For instance, I have an agent that looks at all the AI news all the time. There's a lot of it, and it finds the stuff that I would actually be interested in. I find it incredibly useful.

These are agents that you can install that other people have built. With anything you install on Dreamer, you can say, “I want to start making some changes,” and we'll look at that in a second. In natural language, with the sidekick's help, you can change any of these experiences to work just the way you want them to.

At the base layer of the system are tools. As you know as well as anyone, Swyx, any AI system is only as good as the quality of data that it can pull in and the quality of action it can take.

Before we launched our beta, we worked very hard to make sure that we seeded our tools with a bunch of high-quality, powerful integrations. For instance, this is real Google Search. This is actual Gmail. You can do very useful things with those.

But this is also a platform for everyone. As we started talking to people in our alpha community, a whole bunch of sports use cases popped out. We realized that if you want to build something cool for sports with AI, you need really high-quality live data.

Look at these: Formula 1, MLB, NFL. These are tools that we've built. These are not data scraped off the web.

This is a direct data feed integration, and because it's live and high quality, you can build really powerful stuff. But tools aren't something that we're just going to control ourselves. The platform is open for tool builders to contribute tools that anyone on Dreamer can use.

This is actually the place in the platform where I think software engineers would love to come and play with it. Software engineers are really going to build a lot of powerful stuff into the system. We're actually sharing something for the first time on this podcast: tool builders on Dreamer get paid. If you publish a tool to the platform and a lot of agents use it, you'll actually get paid in proportion to their usage. We'd love for folks to come in, give this a try. We've got good docs that help you get started, and you can build things that scratch your own itch. For instance, someone built this Ski Bum tool, which provides live snow conditions for a bunch of ski resorts. I'd love to show you how I've used that in a second. We also have some tool partners where the tools themselves are pay-per-use.

Parallel Web Systems is a premium tool. You can do really cool stuff with it. It's an agentic web-research tool. That one, because it's expensive to operate, is paid on a per-use basis. But if you're coming in to build agents on the platform, even the premium tools come with a free trial. So you get a chance to actually try them out and make sure that the use case is good for you before you decide to sign up.

So, that's tools. We have the gallery, we have tools, and then the Sidekick helps us put all of this together to build agents. We do that in Agent Studio. You can also do this on your phone, but if I open up Agent Studio here on desktop, Sidekick is just going to start a conversation about what you want to build together. I'd love to show you one that I made recently.

Let's do it.

Let's look at something that's hopefully near and dear to your heart. One of the things I love about Dreamer and this kind of moment in technology is that, if you think about it, there are all these things in your life. Have you ever gone to a conference? I know you have, right? Big conferences have apps. These apps are usually built by agencies, and they're usually quite expensive to build. I've been involved in running some of these myself.

How many conferences have you been to where the app was good? Honestly.

Exactly zero.

Maybe one. I've been to one conference that was pretty good.

Was it Stripe Sessions?

But the point is that they're rarely great pieces of software, right? They're also expensive to build. They're interesting because they're episodic. They last for this one thing, and then they're not relevant anymore.

It's the worst feeling to invest in them because you know it's got a limited date.

Absolutely. So, I decided to build a conference app for your AI Engineer conference on Dreamer.

One of the things that Swyx has done, which I thought was very forward-looking, is actually put a whole bunch of data about the conference on the webpage in an LLM-readable way. There's an llms.txt file. There's a feed of all of the sessions in JSON.

So, I used the data from your conference last year and built this intelligent app just by talking to our Sidekick in Dreamer. To give you a quick tour, this is my dream conference app. What I always want to do for conferences is search for speakers. I'm usually there because there's a speaker I care about. So, Swyx, you're the speaker I care about. I can actually see here who you're on stage with.

Here's Greg Brockman from OpenAI, and this is his session. Look, Greg and Swyx are the speakers. Let's add that to my schedule. Great. Maybe there are a couple of others I might see here. On day 2, I remember there were some keynotes. Building the Open Agentic Web sounds fun, so I'll add that to my schedule.

She's now CEO of Xbox.

Awesome, which is interesting.

So cool.

So, I've gone through and picked out a couple of sessions that I care about. That's as far as I usually get with any conference app. But of course, you've got the whole rest of the conference to figure out what to do.

This is where the native intelligence of these things you build in Dreamer can come in. I'm going to click “Guide me.” Dreamer's Sidekick actually parsed out the whole schedule and figured out what some of the themes are. I can choose what I'm interested in here. I'm definitely interested in agents. I'm definitely interested in code generation, and also reasoning in RL.

Now I'm going to say, “Build my schedule.” What this is doing is going across every time slot for the conference and choosing, among the things I could go to, which one it thinks is best for me based on my interests. It also uses its own memory of me that's part of Dreamer to understand what I might like best.

There's an LLM prompt running for each one of these time slots. It's not super fast, but it'll be done in about 30 or 40 seconds, and I'm going to have a special custom schedule for the conference. This, like I said, is my dream conference app. It's exactly what I've always wanted.

I was able to build this yesterday morning. I did it between some meetings. I think I spent a total of 25 minutes of wall-clock time on it. I did it over the course of a couple of hours. Here is my schedule for the conference. I can see it in a calendar view. This is what I should do on Tuesday. This is what I should do on Wednesday. There are no conflicts, but I may not go to every single thing. There you have it: built in Dreamer.

Let's take a look at what the building experience actually looks like. This is the actual account that I made it on. Of course, I should say that anything you build on Dreamer also works on your phone. Here is my AI Engineer conference app right here on my phone. It's got all the same functionality, and of course, this is the best place to jump into my schedule.

Shawn Wang

Yeah. So you could generate a podcast about it. Just a completely multimodal thing, right? To me, this is why I outsource. I posted the llms.txt and the JSON because you cannot run an AI Engineer conference in 2025 and not let engineers do whatever they want. Since all conference apps suck, I'm just going to put up a minimum viable app and let people do whatever they want.

Guest

Totally. The cool thing about this on Dreamer is that I published this to the gallery, and you can use it. So you've got one built to my taste of conference apps. I think it's pretty cool. But you might want something different, in which case you just start telling the Sidekick how to change it.

Shawn Wang

What's great is that you can also fork it, right? I can publish my own one and say, “This is the base starter. It's got good defaults, but go customize whatever.”

Guest

That's right. So take a look at how I actually built this. This is real. I'm going to say, “Make changes.” This experience we're looking at now is our agent development studio. Like I said, you can do this on your phone as well.

This one I started out on desktop. Let's look at my actual prompts. I said, “Let's make an agent called AI Engineer Schedule Planner to be a custom schedule planner for the AI Engineer conference.” I'm not going to read this all out. You get the point. I told it where to get the data from. That was the first prompt.

After I gave it that prompt, I actually had a simple version of this app working after the Sidekick took 1 turn. The Sidekick is like a professional software engineer. We worked very hard to make this work and build functional apps for folks who might not have any engineering experience whatsoever.

Down here we have build logs that are technical, but you can hide those away. As Sidekick is building, it will actually translate everything that's coming out of the harness into English that you can actually read. By the way, this English is in the personality of your Sidekick, which is fun.

The way that we build agents and agentic apps is a little different from what you might have seen on some other platforms, for a couple of reasons. One is just the build process. The very first thing that Sidekick does is understand all the agents you've got set up. It understands all the tools, and it will come up with a plan for how to realize your goal and make sure it actually has the data and capabilities to complete it.

It will occasionally refuse. If it can't do what you're asking, it will tell you, “I can't do that. It needs another tool.” That's a good jumping-off point for any of the tool builders out there to build a new tool.

It'll first figure out how, then it'll build it, and then it'll actually test it. It will make sure that the thing it has generated is realizing your goal. You probably know as well as anybody that anytime you can get any modern, state-of-the-art coding model into a loop where it can make changes, perceive its own output, and then fix bugs, magic happens.

The first build will often take 10 to 15 minutes on Dreamer, which is a little bit longer than you might have seen on some other platforms. But the first thing that it creates will work most of the time. Then, of course, as you start making smaller changes, you can ask it to tweak the UI in any way that you like. Those are much faster.

Just to give you a sense, here's something I asked: “Put a logo.”

I gave it a logo file in static files and said, “Use that as the title.” For folks who really want to dig into a bit more detail, we've provided a powerful IDE here. I can actually see the code that was generated. Some pieces of the code are more accessible than others, like the prompts.

This is the prompt that's used by a powerful LLM to do the schedule picking, and I can read it here directly. I can edit it without having to ask the Sidekick if I want to do that.

Shawn Wang

Which is very nice.

Guest

For the more sophisticated users.

Shawn Wang

Yeah. This is other people's entire startup: prompt management.

Guest

This is true. The other thing that's different about Dreamer is that once you've built something here, it's ready to go. We host it, so you don't have to worry about getting a database from a database provider, signing up, or getting API keys. You don't have to worry about your LLM provider tokens. All of that is hosted on the platform.

You can use it yourself. You can share it to the gallery for other people to riff on it. You can also share it with your friends and coworkers to use your instance of the agent or agentic app. We're seeing that happen a lot in our community. We've seen a whole bunch of folks who built little applications for their personal lives and shared them with their significant others.

We've seen people who are building little productivity apps for their teams at work and sharing them among themselves. We actually do this a lot inside the company. At this point, we pretty much run the company on Dreamer agents for all kinds of important things. Maybe a good example of that is our waitlist.

People are signing up, and every time someone signs up for our waitlist, a Dreamer agent will actually research that person. We're looking for folks who are builders, not super technical, to build agents and come in and give us a lot of feedback. We'll prioritize bringing those people off the waitlist first.

Shawn Wang

Just a quick question on that one—it may not come up again. Do you find enrichment APIs useful, like ZoomInfo or Clearbit?

Enrichment is a very common use case on Dreamer. Any application on Dreamer can kick off a subagent to do a particular task. This is actually a powerful agentic harness that runs inside its own VM. We call them Sidekick tasks because they run in the context of the Sidekick. I'll talk more about Sidekick in a second.

Enrichment is a very common use case, and the cool thing about a Sidekick task is that it has access to all the tools on the platform, as well as public data. Very frequently, enrichment on our platform happens using public data that can be found on the web.

There are some tools for getting people data from various bespoke systems, and those work pretty well. But you'd actually be surprised. We would love it if someone out there would like to build a ZoomInfo tool. We don't have one today, but we'd love to see that on the platform, and I'm sure it would be very powerful.

We're also seeing that this powerful agentic harness can pull in a lot of data. On that note of tools that make experiences better, we're constantly adding more tools because people in the community are building them and publishing them. We review the tools carefully, and then they go live for everybody. Yesterday, we added Granola, and that was pretty cool.

Sarah on my team was talking to someone building on the platform this morning. They have an agentic app that they built, which is kind of a magic to-do list. They put stuff on their to-do list, and for each thing, it kicks off one of these Sidekick tasks to figure out how to move the ball forward on it. Sometimes it will complete it entirely—

Shawn Wang

Yeah.

Often by calling another agent on the platform, and sometimes it just researches it and helps them take the first step.

Shawn Wang

Do you know that this is Sam Altman's number one ask for an AI app? It's the self-completing to-do list.

Yeah. The self-completing to-do list is something that a lot of people have built on Dreamer. I'm getting a lot of use out of it, and I'm finding it genuinely—

Shawn Wang

I actually should try that.

Mm-hmm. Please do. You'll even find some in the gallery that you can remix.

He was saying this morning that he had already built this self-completing to-do list on Dreamer, but he connected it to Granola yesterday. Now something really magical happens: when he says in meetings that he's going to do a thing, it magically shows up on his to-do list, and then it can magically get completed.

As I mentioned, all the agents and apps on Dreamer can actually work together. Our coding agent does something very special as it builds them: it exposes the internals of each experience to the system. Then Sidekick can manipulate those to get stuff done.

He has built another agent that he uses for recruiting. It keeps track of candidates, and it also has a kind of mini-CRM function, so he's able to introduce candidates to each other. He told us this morning that something he committed to do in a meeting recorded on Granola yesterday showed up on his magic to-do list. His magic to-do list had a task to introduce a person for recruiting, and it used his recruiting agent to get it done.

Shawn Wang

Ah.

This is the dream. This is why we started the company. It really is the case that you can build and use these very powerful bespoke experiences that can automate your life by working together. I'd love to talk a little bit about how they work together.

Obviously, it's really cool to have software that will work on your behalf, but it's only useful if you can trust it, right? Privacy and security are very important to us. Making these things accessible and useful while also being trustworthy is hard.

The model that we have, which is working very well, is that the Sidekick is at the core of everything here. It's both your companion and your helper, but it's also the traffic cop in the system. When one agent wants to work with another agent in Dreamer, it doesn't do it directly. It does it via the Sidekick.

I ask the Sidekick to do the thing, and the Sidekick understands everything—all the expectations that have been set with me as a user about what agents can do and which tools I've given them permission to use. It makes sure that whatever is going on is actually aligned with my own interests.

That's part of the background that I bring to this problem domain. I've worked for years keeping very important information safe and secure. As we started to think about this problem, we realized that we actually had to build something that's a bit like an operating system. The Sidekick is like the kernel, and the agents and apps are like users.

See how this is different, right? Exactly. Because if you try to pick off just one piece of this, you can't actually make it work for people at scale. You could build little vibe-coded apps, but they're going to grab all your data willy-nilly. They won't be able to work together.

You actually have to invest in the fundamental core in order to make it work well for people. That's what we've been doing, and it's been a lot of fun.

One other thing I wanted to mention is that I've obviously talked about 2 things: tools and agentic apps. We really designed Dreamer to be an ecosystem and a platform. One of my favorite quotes about platforms—I think it's from Bill Gates—is that you can only be a platform if you create more value for the folks participating in and using the platform than the platform itself creates.

That's our goal here. At every step, we've been thinking about how to make sure that other people are deriving even more value from Dreamer than we are. In that vein, I already mentioned that tool builders get paid, and people can build agents that solve their needs and share them with others. We're already thinking about ways that they can monetize those as well.

Against that backdrop, one of the things that we're launching today is our Builders in Residence program. There are tons of people building really cool stuff and contributing it to the gallery already, but we've been really inspired by programs we've seen at other companies where artists might be in residence—people who are very creative and might have ideas outside of what the folks at the company or in the ecosystem already have.

We're looking for creative people who have fun ideas and, you know, want to really figure out how to apply their creativity at the cutting edge of technology today to come and work with us. So, if you go to dreamer.com/latentspace, you'll find a link both to our tool builder information and our builder in residence program. For builders in residence, we'll let you in off the waitlist quickly, build an agent, and then for a small number of the most creative folks, we're going to pay you to build agents. You can work directly with our team. You know, this is like building Legos. We've got some of the basic blocks together already, but if you need a round steering wheel and we don't have one already, we'll build it for you. We really want to be inspired by these builders in residence.

Shawn Wang

And this LEGO thing is pretty common as an analogy. There's a thing I call the master builder. The actual LEGO company has master builders that they employ to inspire people and post on social media.

That is exactly what inspired us as well, honestly. We talked about the LEGO Master Builder program. That's our builder-in-residence program.

Yeah. And then, finally back on tools. Like I said, anyone can come in and build tools today. If you follow the latent space link dreamer.com/latentspace again will get you off directly off the wait list so you can build right away. You can monetize by publishing onto the platform. That's for everyone. The very best tool that gets added to the platform by mid April. We have a $10,000 prize that we want to give out really because we just want to seed the creativity of everyone out there. So we're excited to do that.

Yeah, and you know this is completely a flywheel, right? The more tools, the more builders, and the more agents—it just feeds into each other. That's right. Yeah.

Just on the payment thing, because we probably won't touch on that again, but I have to ask the former CTO of Stripe about payments. Presumably, you're using Stripe Connect. Any pain points? People are very interested in agent-to-commerce and micropayments and all these things. Presumably, stablecoins get into the conversation at some point, but maybe not now.

Yeah, we're really excited about agent-to-commerce. The first step we're taking is to help people in the world who have never been able to build these kinds of experiences in software before build stuff that meets their passions, share it with the world, and get paid. So that's all commerce that happens on our platform, and we don't need anything new to facilitate that. Stripe Connect has existed for quite a while and is the perfect solution for this kind of stuff. So we're excited about that first and foremost.

However, a lot of the things that people are already doing on Dreamer—we just talked about a self-completing to-do list—a lot of the ways that you want to complete to-dos is by actually closing the loop in the real world, and that's going to involve the exchange of value. We have some folks who are building tools already that actually do have money move in order to complete that loop.

So far, we just want to be open and agnostic to all the protocols out there. I honestly think this moment in time is a little bit like the early web. I personally started coding as a kid, and I think I got access to the internet in about 1995 or 1996.

Back then, the web existed—HTTP was a protocol—but there were also other protocols I was using all the time, like Gopher and UUCP and various others. The point is, the web, HTTP, and HTML were just one among many protocols, and of course it became the winner, and it's awesome. But the others were also kind of interesting and viable at the time as well.

Shawn Wang

I think the world of agentic commerce is like this right now.

ACP.

Shawn Wang

ACP, exactly. All the CPs.

On Dreamer, we hope that folks will build tools that make use of all of these things, but I'm sure that at a certain point one or two will emerge as the winners, and then we'll be able to build really deep support for them.

Yeah. This is maybe a complete tangent, but I do think about how a lot of these AI companies, in particular, have to switch from C-based to usage-based, because of course. But then they end up having to obscure the margins a little bit, and then they end up inventing their equivalent of Robux. They're like, “Well, okay, every company should have its own currency.” And it's a very short leap to a token.

Dreamer

Yeah.

And I'm like, “Okay, well, where does this end?” I can't really play out the next step. Is this chaos? Is this okay?

Dreamer

Well, I think it is like the Wild West. I don't mean that in a completely disorganized way, but there are so many things that could happen from here. The Overton window is very wide right now for where this might land, and I'm just very excited to be building a platform that can take advantage of all of those opportunities. We're just going to be there, working for our users to make sure that things that emerge work.

You're going to own the consumers. You're going to be the OS, the app store for everything.

So, one of the ways to think about this is that Dreamer actually uses all of the state-of-the-art models. As a user, you don't have to think about should I be using Opus 46 or should I be using the 54 model from Open AI. We are continually doing evals and so forth to make sure that the best things are there for you.

You can just build on the platform and know that, as the world shifts around, you're going to get the right stuff for you. I think that's something that is needed to actually have folks take advantage of this technology at scale.

I'd love to show you another example of something I built.

Let's do it.

This is another example of software that just lasts for a certain moment in time. Recently, I went on a ski trip with a bunch of friends.

So we use Ski Bum?

Yes. I went on a ski trip to Big Sky. I'd never been there before, and I made this little intelligent app for us. You can see it says, “Loading Big Sky conditions.” It's actually calling the Ski Bum tool that I just showed you, which is published in our gallery.

What is this? This is a little app that was just for our weekend trip. It shows the current status of all the lifts at Big Sky using that tool from the ecosystem. It shows the forecast for the upcoming weekend. It shows our accommodation—this is just where my group was staying. This is just for us. It also shows a bunch of dining information that one of our friends put together, who's an expert on Big Sky.

I was able to take this app and share the link with my friends. They weren't on Dreamer yet. I just sent it to them on iMessage, and they got a version they could use on their phone.

Of course, here's the real kicker. I've been on ski trips before and other weekend adventures with our friends. People pay for different things, and at the end of the weekend, it's always a pain to figure out who needs to pay whom to settle up.

So we used this during the weekend. We added all of our expenses in here, and then at the end of the trip we pressed “Split,” settled up, and were done. There's another reason there. This was all through Dreamer.

The actual payment happened because we paid for stuff in the real world. It was like, “Okay, this person needs to pay that person $20. This person already paid.” So it just helped us all settle up. We didn't move the money on Dreamer. You could do that, and in fact if you're a tool builder thinking about this and getting excited, come build a tool to do that stuff. We really think of our tool builders as design partners.

Yeah, I got the tool. I use Bank of America. I hate the app. I hate the Bank of America banking website. It's just—

Yeah.

—horrible.

Dreamer

Yeah.

So, just build me a thing on top of Plaid—my banking app.

Dreamer

There's already a tool for that. Attain Finance is a tool that a builder in our community built, and it uses a secure system like Plaid to access your financial data. You can build powerful personal finance agents on Dreamer today using this tool.

Like I said, we review tools carefully. When bringing Attain Finance onto the platform, we did a detailed security review with that company to make sure that, if folks build stuff with it, it's going to work well. I'm pretty certain it connects to Bank of America, so you'll be able to build the app that you wanted already.

Yeah. There are a couple of points I wanted to dive in on, maybe highlight to folks, because I obviously spent more time with Dreamer. The point is that you choose on behalf of your users because they're meant to be consumers, so maybe less technical.

Right.

But obviously, our users can override if they need.

It's not just LLMs. It's also the transcription. There's a first-party curated set of—here's the house opinion on what the thing is. What's the list? Is there, like—

Dreamer

Yeah, so actually, if you look in the tool gallery, the first-party curated set are all the ones that have these grayscale icons. We have a built-in tool for image understanding, image generation, RSS exploration, text-to-speech, and so forth.

Recipes?

We actually do have a built-in recipes tool. It turns out that a lot of people in our alpha wanted to do stuff for cooking. You can scrape the web to get good recipes, but we were able to quite quickly find a good repository of recipes that works great here.

Yeah.

Dreamer

The point behind these, though, is that we'll keep the interfaces stable, so they'll always work. The best translation model today might be different from the best translation model sometime next year. There are people using this translation tool to translate Chinese podcasts into English. It's pretty powerful; it can deal with very long text.

We're just going to make sure that the translation tool is always pretty close to state of the art, so you can build something and know it's going to continue to work well. Of course, some of our tools are branded. You may actually have a preferred way of buying groceries—maybe you prefer Instacart—and that's great. You can use the Instacart tool specifically.

Yeah. Your partnerships—I mean, I don't know if you've ever had a partnership, but this is going to be a bonanza for anyone wanting to do deals.

We have an amazing person who works on all of our partnerships. It's part of what you have to do to build a platform like this that's going to work for people.

We've gone and done that schlep. It is a lot of work. We've talked to lots of different companies in order to make sure that you've got good tools at the core. And then, of course, because we're open to tool builders contributing to the platform, this is only going to get better and better and better.

One observation I have: this is kind of a massive thesis I've been pursuing. What I've been calling an agent lab is different from a model lab, in the sense that you never train your own models, but you are the router and evaluation layer, and the subject-domain expert for choosing between models. And then you're explicitly doing these things. In my sort of construction, every agent lab does some version of this: here's the image-understanding endpoint, and we will route for you and not worry about it. So, I think it's kind of cool.

I think it makes total sense. Again, to make this work for folks that don't follow the AI news every day, it's actually a really important thing to do.

It's been a real pleasure. I'm personally a total geek for this stuff. I love it. Being able to dive into all those details in order to make it work well for other people is a true pleasure. I can't imagine working on anything else right now. It's just so much fun.

The tricky part is multimodality, when some of these things merge.

Mhm.

You're sort of imposing structure on things that fundamentally aren't structured, and sometimes that might work against you. But for 99% of these cases, this is fine.

Dreamer

I think it's going to be very interesting to see how the world matures. A lot of the power of Dreamer is the ability to kick off these sub-agents, these powerful agentic harnesses, which can actually change how they work based on the data. I actually think that we'll be able to keep up with and stay at the forefront of the changing landscape of how our tools and systems work together.

That's new. Software didn't used to work like this, and now it does. Even just figuring out how to design the right primitives to make that possible has itself been quite fun.

This is a sort of two-part question. Why can't Dreamer make its own tools? And then why don't you let tool builders maybe stand up their own routing group? I call this a routing group, right?

To some extent, Dreamer does make its own tools, in that agents appear to the system as tools. They can be used to accomplish things. You can build an agent that is essentially a tool.

Which is, to me, very useful for reuse.

Right.

Right? Because I like this; this is the way I like it. For my next 5 apps, I don't want to do this whole series of back-and-forth again.

Dreamer

Right. Yeah. Then, at the tool layer of the system, it's open to anyone, so it's actually quite powerful and flexible.

If you wanted to add a tool, imagine that you were your own foundation model, swyx. That might be fun. Imagine you wanted people to be able to play with it. Maybe you make NanoChat or whatever, and you want to let people play with your own NanoChat and see how it goes. You could publish a NanoChat tool with image generation behind it, and it could be your own router if you wanted it to be.

I see.

Honestly, if that's the kind of thing that gets you excited as a builder, please come and do it. We really are believers in this idea that we aren't going to figure out every single detail ourselves. We're going to make sure it's a safe and fun place to build this stuff, but we're really open to these ideas coming from other people. I'd like nothing more than for you to come in and build a tool that does some of that cool stuff you have in mind.

Yeah, awesome. And just as a reminder, if you'd like to do that, the way to find the links is dreamer.com/latentspace. And for a limited time on that page, anyone who's listening to this podcast will also get directly off of our waitlist. It's quite long right now. We are working hard to bring Sequel out.

Wait, so skip the waitlist. You know, I think that's fantastic. I think it's really sort of pro-builder way to do it. Well, I wanted to jump back to the bar. You know I get excited about this.

Yes. Okay, let's shuttle it back in there.

Shawn Wang

Let's get as technical as you can on everything you've built. Show off.

Dreamer

Okay, so let's go wild in the Agent Studio. As you can see over on the left here, there's a conversation with the Sidekick where you ask it what to do, and it will explain in English that anyone can understand what's going on. But if you want to pull back the covers and look under the hood, if you're an engineer like me, then we have this debug drawer at the bottom.

You can see the full build logs here, but you can also dig in and see the files and prompts that have been generated. You can upload files from your computer and static files.

Shawn Wang

Very important.

Dreamer

Indeed. You can actually read the prompts that have been generated for you. We intentionally put an example in here just so you can see what the format looks like. We already looked at this one that was generated for this particular app. But if you actually want to bring the code out of Dreamer and work on your own local machine, you can.

At the core of everything here is an SDK with a powerful command-line interface, and we built that first. It's actually possible to build agents on Dreamer without talking to the Sidekick. You can write code with your fingers on a keyboard if you want to. I know that's very antiquated now, but actually this can be a lot of fun.

If you want to pull it out onto your laptop, you can use our CLI. You can edit it in Cursor or in Cloud Code; you don't have to use our Sidekick. The CLI actually has full access to the rest of the platform with you as a user, so obviously it is secure and privacy-sensitive. This is how some of our most technical builders build stuff on the platform.

The really cool thing is that when the Sidekick is in coding mode, it uses exactly the same CLI. The way it builds stuff on Dreamer is by using the same tools that you might as an engineer. That's actually a very powerful abstraction, because it turns out that the right way to give a lot of context to agents to use CLIs is to write great documentation. Make sure that all of the things that you could do are actually possible. And guess what? That makes it a delightful developer experience for real humans as well.

Shawn Wang

Yeah. So, that's pretty cool. We've been telling developers to do this, and they ignored us until now—

Dreamer

I've been saying this for a long time.

Shawn Wang

Let's see Stripe docs. I mean, come on.

Dreamer

Absolutely. Come on. Absolutely. But actually, I was chatting with folks at Stripe last week and saying, "Hey, you've got to make the Stripe CLI actually tell agents what they can do on Stripe, because that way they're going to use more stuff on Stripe." I think this is a real trend for the entire industry.

Shawn Wang

Yeah.

Dreamer

So, we've been doing that.

Shawn Wang

To me, this download-and-get-pushed-everything approach is complete confidence that you're not hacking it. There are other, let's call them AI builder platforms, that impose their stack on you. They don't allow you to do this because they cannot. They impose some degree-of-freedom restrictions so that they can get it to work. Yours is a fully general VM running the full code, correct? Do whatever you want, in any language you want.

Dreamer

Yeah, correct. In terms of language, if you use the SDK, you could build stuff in other languages. We've actually found that TypeScript is the best language for building these experiences. It's strongly typed, so you find out at compile time if you've made mistakes, and there's nothing better than getting an agent coding agent in a loop where it can see its mistakes and fix them. TypeScript is the language that everything gets built in by default here.

Shawn Wang

Do you see TypeScript overtaking Python in the game?

Dreamer

Yeah. For what it's worth, when we started the company, we started writing stuff in Python, and I love Python. If I do have vendor code, I always write it in Python. It's my favorite language as a developer with my fingers on the keyboard.

But TypeScript is an amazing language for AI because there's tons of training data in the models, and it's strongly typed. At the company, we built most of the stack in TypeScript, and we have this amazing property: we have type safety all the way from the database to the front end. There's nothing better for working with coding agents than being able to have them check their correctness at compile time. The same ideas behind building the company's codebase we've put into the agent SDK here as well.

Shawn Wang

Do you know if you use one of those tools like Prisma or whatever, or is it too low-level for you?

Dreamer

We actually have crafted most of our own tools. For instance, we had LLM-driven code review before the thing that got published from Anthropic this week.

Shawn Wang

You know, we all pay $25 per review.

Dreamer

We pay a lot less than that. However, I hear that those reviews are excellent and possibly worth $25.

Shawn Wang

Yeah, it's an option, right? It's good to have it.

Dreamer

Just to give you a tour of some other stuff here. I can also see all the versions. This is not Git; this is built into Dreamer. I can see all the versions that have been pushed before. Why is it not Git? It's not Git because we can make it work more efficiently than Git, and we actually do some work behind the scenes to understand what's in each of these.

Shawn Wang

One of the things I'm pursuing, and I have a lot of pieces, right? One of the pieces is: Does Git go away? Does GitHub go away? And what is the AI-native direction, for what it's worth?

Dreamer

To some extent. In anything you build, there's a lot of path dependency. If we started over, we might make this Git. Within the company, we use Git for our platform source code, and we like it. It works well with coding agents as well. The very first versions of this, we wanted to make it possible for the Sidekick to manipulate it easily, and this was an expedient way to do it.

Shawn Wang

Yeah.

Dreamer

You can also see all the activity that has happened in the workflows that you build. A lot of agents you'll build on Dreamer do things in the background. They run on triggers. These are stimuli from the outside to kick them off, and this is a nice way to see all of the things that might have kicked off your agent.

You can have an agent that kicks off on a webhook, so you can plug it into external systems. You can have an agent that runs when you receive certain emails that match filters, including LLM filters. Here you can see when it ran and what it did. If I open up one of these—Guide New Prompts or Guide Me If/Else—

Shawn Wang

God.

Dreamer

Well, I told you it was calling an LLM for every one of those time slots. Here are all of the LLM calls. Here's the actual prompt.

Shawn Wang

And you don't mind exposing all of this, right?

Dreamer

No, we want builders to see what's going on under the hood.

Shawn Wang

I want to see. Okay, yeah.

Dreamer

So, right now, that one was Haiku. Like I said, we work with all the models, and Sidekick will actually pick the best one for the job. You saw that it was pretty high quality and pretty fast. Haiku 4.5 is the one it picked for that job.

Shawn Wang

That's it.

Dreamer

We also have logs. As I mentioned, there's a database spun up on demand for every agent. You don't have to go and figure out how to do your own hosting. It's a SQLite database.

Shawn Wang

Yeah.

Dreamer

It's a multi-user SQLite database. Each one is a database that's unique to this agent. But if you share the agent with multiple people, we take care of who the owners are in each row and all of that stuff is just there out of the box.

Shawn Wang

And, again, in-house?

Dreamer

In-house.

Shawn Wang

Oh my God.

Dreamer

Yeah. We do work with a bunch of infrastructure providers, but the technology for how to manipulate this is in-house. Fun fact: we actually did a lot of our own infrastructure development early on at the company and realized we needed to spend our energy on the stuff that we're uniquely doing in the world. So, we're very delighted to partner with a bunch of great partners on some of this stuff.

Finally, I mentioned that agentic apps expose all of their internals to the system, so the Sidekick can manipulate them and use them just like a user can. You can see how I've decided to break this problem up into functions. Some of the functions—the ones with the little eye here—are exported.

Shawn Wang

Probably the user level from my side, exactly.

Dreamer

Others are internal. If you want to, you can dig right in here and call individual functions and see what happens. But mostly, you don't need to think about that at all. You can keep that little drawer closed, talk to your Sidekick, and build really powerful and enchanting experiences.

Shawn Wang

To me, showing this gives the engineer a complete mental model of what you've done and what you can do with the app. For example, the first thing I look for is a mental checklist of things, right? Is auth in the database? Auth looks like it's not, right? So that's a separate layer. That probably means it's hard to do multi-user apps in the same app, right?

Dreamer

Actually, we've solved that. The platform builds in auth. As a user, you sign into the platform. If you're using an agent that was published by someone else, your identity is taken care of by the system. When you query the database, you're going to get the stuff that's for you, unless the builder specifically said, "This is public data that everyone should see."

They actually get a chance to think about that. Again, Sidekick can guide you through building agents and apps that work that way. You're right; that's another thing that people have to think about when they're trying to figure out how to build software experiences. On Dreamer, it's built in. You talk to the Sidekick as if it were a human being about what you want, and that's what you get.

My Big Sky app that I just showed you was designed for multiple people to use. The things that we were putting in as expenses were supposed to be visible to everybody, and I just told the Sidekick that's the way I wanted it. But by default, if I'd built an app like that, the data from each user would not have been visible to the others.

Yeah. This is, I presume, a moot question, but basically you've had to build your own coding agent, right? Which is Sidekick, or whatever is inside Sidekick. Obviously, there's a lot of people with a lot of desire for Claude Code and Codex and attachment to them. I know under the hood they basically reduce to a loop. But would you let people use Claude Code and Codex, or is the harness too specialized?

If you want to use Claude Code and Codex, you go into the area. Hit "Get the SDK." We even say this right here: "Edit to your heart's content. Zed, Cursor, Claude Code."

But people want to use it inside of Sidekick, right? They want to switch the engine. That's the coding engine.

We're not doing that right now. Again, the goal really is to abstract the complexity, because the real target for building agentic apps is folks who can't do this already today.

I can't tell you how many users in our community I've spoken to who are like, "Dreamer has changed my life because I used to have all these ideas. If only I could find an engineer to help me implement them, I'd be able to get them done." Now I can talk to my Sidekick and get it built.

I think that's really how we think about the people who should get a ton of value and fun out of the platform. They're not asking to be able to plug in their own coding agent. For those folks, the opportunity is massive. If you've never been able to do stuff in code, now you can build stuff for yourself, your friends, your family, and your co-workers.

There's also a huge opportunity for folks who do build stuff in code to contribute to this ecosystem. That's how we think about it.

Yeah, amazing. That's most of what I wanted to cover, Dreamer-wise. I think personalization and memory—yeah, it's probably the single most important job of an OS. Maybe we could talk about that, and then I wanted to zoom out on company-building stuff.

Yeah, sounds good.

How do you handle memory? What have you found? What have you tried and failed?

At the core of Dreamer is the Sidekick. The Sidekick gets to know you and builds up a memory about you over time. That turns out to be very important. Dreamer gets better the more you use it.

For instance, a lot of agents on the platform, when you start using them, the first thing they do is show you, "Here's what I think is relevant to you for this particular use case." A very popular kind of agent on Dreamer is a weekend activity planner.

Just tell me what to do.

Tell me what to do, especially if I have kids, right? I have 2 kids and a dog, and my wife and I often spend a lot of time trying to figure out, "What are we going to do with the crew this weekend?" We have interests that are very consistent. It can actually take a ton of work during the week to figure this out.

There's an agent on Dreamer called Weekend Activity Planner, and it helps me find things to do with the family at the weekend. In fact, this morning, I got a message from my Weekend Activity Planner telling me about the St. Patrick's Day parade on Saturday at Civic Center.

I'm Irish. My kids are technically Irish as well. I mean, they have multiple citizenships, but they're Irish. What better thing to do than take them to the St. Patrick's Day parade?

Why did that get recommended to me? Because the Weekend Activity Planner has a profile about me, and it knows that I'm Irish, right? All that comes from the memory that Sidekick builds up about me over time. We have invested in this a bunch, and we'll continue to invest in it more.

We've tried many different techniques. As you know, the cutting edge of agentic memory has changed over time. Very early on, we were putting lots of facts into vector databases, doing embeddings, and playing them back out using reverse-lookup embeddings—RAG.

That actually worked, but it turned out to be much more complex than was required. Today, we've replaced it with a different system. I think we use a system that's pretty similar to what you'll find in lots of other products, but it's an area that we're actively investing in. There's more than 1 person at the company specifically working on memory, so expect us to continue to make it better.

Did you try knowledge graphs?

We've tried knowledge graphs. The system that we have now is not a knowledge graph, but we've probably implemented most of the papers you've seen out there on agentic memory. The current system is working pretty well.

Excellent. Zooming out just on the company stuff, this is your first time as CEO.

Correct.

You were CTO before.

Correct.

What's different?

The difference between being a CEO and a CTO really is just making sure you're looking across everything. I have run products before. For instance, Android Wear—

You're basically a CEO of that product, yeah.

I was running that as a general manager.

However, when you do it for your own company and the buck truly stops with you, it definitely raises the temperature a little bit. It's been a lot of fun for me to think about a lot of go-to-market topics. I spend a lot of my time these days meeting users, talking to folks about what they could do on the platform, and being very active on X and LinkedIn, which, by the way, is a huge pleasure. It is so much fun to be able to engage with users and potential users directly and understand what they would like to do. That's the biggest difference between this role and being the CTO of a company.

At the same time, I am someone who always likes to look for why we're doing this and who the people are that benefit from it. Even as a CTO, I was always paying a lot of attention to topics across the company. I feel very grateful for all I learned in my previous roles that got me ready to do this at this kind of scale.

To me, this is the natural lead-in to when I went into your office.

Yeah, surprisingly small.

Yes. I have another thesis I'm pursuing for Latent Space, which is the emergence of tiny teams. The classic image is teams with more millions in revenue than employees, right? That's a revenue-efficiency definition, but I do think that, as a CEO, you're going to run a smaller team than you used to.

I believe very strongly in the power of small teams. The more people you add to a team, the more communication overhead there is, and it doesn't even grow linearly, if you think about it. The more people you add, the more everyone cares about getting to know everybody else and sharing what they're doing with everybody else.

That's great. I'm not saying they shouldn't, right? I want to work in teams that are fun, where people are talking to each other and sharing ideas and so forth. But there's just a kind of gravitational weight that comes from larger and larger teams. Inherently, large organizations are less nimble than small ones. If you run a large organization, you have to keep thinking about how to prune things so that it can act like a small team.

At Dreamer, the core team that built everything I just showed you was honestly about 6 people. We're larger now. We're about 17 people at the company now.

But still, for everything you just showed—

It's still a small team, which is great. It's a very, very high-talent-density team. We've been very careful and obsessed, as we grew, to make sure that everyone who's joining the company is joining a team that they're going to get a lot of learning out of, but also that they're actually going to help everyone else a lot as well.

There's something very special about that, too. Every single person at our company, I'd be delighted to do any project with at any time because they're all just great. The smaller you keep the team, the easier it is to make sure that the talent density is there as well.

Of course, it's a real luxury to be building a company. We started this company in late 2024, but it's a real luxury to be building a company today because we can build with agents. We're using coding agents and Dreamer marketing agents. In all of our operations, we're looking at how we can accelerate what we're doing using our own tools.

Are there any agents that you don't build that you want to shout out, that you just love?

Yeah, is it—

Other people's?

Agents that we built for the company?

No, other people's stuff. You shouted out Granola.

I showed you Attain Finance. Attain Finance has an agent as well, which helps you manage your money. I find this really amazing. I always have this lingering feeling that I've got a whole bunch of subscriptions that, if I just had a bit of time to go across them, I could figure out how to consolidate them.

The person who built Attain Finance doesn't work at our company. They were part of the early alpha group, so they got a look at how all this stuff works pretty early on. They built this really amazing experience that actually helps you save a lot of money because it helps you analyze your purchases. It's almost like a financial fitness coach.

He's called Andrew, and he built it. He came and showed it to us. The first thing it did was recommend that he should buy fewer burritos. He was like, “It's true. That is actually how I could save the most money.” That's a pretty cool example.

I noticed he was first because he's usually the first in there, or it could be alphabetical order. I'm wondering how come they're not, like, Aardvark?

If you're a builder right there and you're wondering how to SEO yourself on the Dreamer platform, swyx suggests you name your tool Aardvark.

In all seriousness, those are the tools I have connected, so they're in alphabetical order. If you haven't yet connected them, we actually put them in the right order for you, so Sidekick understands you and puts them in the right order. But I'd say Aardvark is going to come before anything else, right?

Shawn Wang

Yeah, exactly. How has hiring changed?

You've hired plenty of software engineers in your life. I assume something's changed.

Suhail Doshi

Absolutely. One of the main things that I look for now when hiring engineers is how well you work with coding agents. Our team is quite experienced. Everyone at Dreamer—well, I guess I write a lot of code, too—everyone's an IC, an individual contributor.

Many of the folks who work on the team have previously been managers, and it turns out that being an engineering manager, as long as you stay very close to the code and are able to continue to craft it yourself, is actually a great skill profile for being able to make agents work for you and for your team in this age. That's definitely something that we look for quite intently when hiring engineers.

We still have folks write some code with their fingers. It's just important to know that the core of the craft is there. But the vast majority of what we spend time doing is building quite significant and elaborate stuff together in a fun, collaborative environment with coding agents.

Shawn Wang

Right. What is the interview loop like? Do you sit there with Codex and do something?

Suhail Doshi

Our interview loop is, first, a coding screen to make sure that the base is there. Then we do a couple of short projects with an engineer on our team and whoever is thinking about joining.

We'll put out a very fully formed product idea, riff on it together, and make sure that we test product sense a little bit. Then we'll try to build the whole thing with Code X or Cloud Code or whatever the person is most familiar with.

We watch how someone thinks about prompting the agents and what they do while the agent is working. You can actually see this is a kind of interesting dynamic in the industry: anytime I'm working on code these days, I always have more than 1 agent going at the same time. While 1 agent is going, I'm reviewing the output of the next one. If you get them in a nice round-robin, you can be very, very productive.

You can also chain agents together. You can have 1 agent producing code and another agent reviewing it. Seeing how folks have adapted their workflow is a big part of what we're looking for in our interview process.

Shawn Wang

Amazing. I guess this is the last question, but I'm also open to you bringing up any topics that I haven't touched on. What have you wanted LLMs to do that they still cannot do today?

Suhail Doshi

That's a great question. It's amazing because the capabilities of LLMs just advance so quickly. If you'd asked me a year ago, I would have said music generation. I like music, and Suno's amazing, by the way. With previous generations, I thought, “Yeah, I can kind of tell that that's AI-generated.”

Today, I listen to the latest tracks made by Suno, and I'm like, “That's pretty impressive.” If we went back 6 months, I'd be asking for better image generation. The latest Nano Banana, which, by the way, is a tool on the platform that you can use on Dreamer, is producing spectacular infographics and spectacular painterly images when I ask for those as well. That's quite impressive.

I think that as we go forward into the future, there is still a lot of room for human creativity. That's also maybe where I'm going to have to say that LLMs are most lacking. When you think about building software, the thing that's really important and that we all need to bring is taste. You have to truly understand people, their motivations, and how to build something that's really delightful.

We had to do a lot of work on Dreamer to make it possible for the experiences that we build not to look like AI generic slop. We've done that by putting a lot of our own taste into the templates, the prompts, and the harness. I hope you have fun playing with it. I think Dreamer today generates experiences that don't feel super generic.

But that's a ton of work, right? The LLMs do not do that by default. In fact, if I see a simple to-do list app or something built by the models, I can tell which model built it just by how it looks. Taste, creativity, and a sense of individuality are still things that I think the LLMs are not producing out of the box, and I think that's going to be an interesting frontier. What do you think?

Shawn Wang

Usually, this is my from-builder-to-researcher question because we do have researchers listening, and I'm just like, well, that's it. But taste, for me, is a very broad topic. What do I think? I agree. I just think that it's too big of a topic to break down, particularly because there's a lot of “I'll know it when I see it” type of thing, which is unverifiable for researchers to do.

Suhail Doshi

Yeah, I do talk to researchers quite often, and we talk about this topic. I think most people agree that one of the great things about building models to generate code was that it's so verifiable. It's very tractable. They agree that the next problem is how you step up that hierarchy of needs and get into these taste questions, and quantifying taste is hard.

But I'm actually excited that some people are going to start doing this. That's when I think that some of the really iconic companies in the world will start to become places where folks really try to debug and understand the creative process, and I get pretty excited about that.

Shawn Wang

Yeah, I think we are slowly uncovering what intelligence really means. The benchmarks that we adopt and then abandon because they're solved are basically us evolving machine intelligence in a way that's different from how we evolved, but we are slowly understanding what it means to be intelligent, right? It's interesting. I wonder how they surprise us in the future, but we're not even there yet.

We just get it to a place where we like what we get from the machine. It used to be at 30%, and now it's 95%, but still there's that 5%.

Suhail Doshi

That's right.

Shawn Wang

Yeah. Any other topics we should have touched on?

No, I think we've covered everything, but I want to remind everyone

dreamer.com

dreamer.com/latentspace.

Yes. No, it's a very good deal. I mean, like, come on. Like, yeah. So, thank you for offering that.

Cool, Swyxster, thank you so much. This was fun.

Yeah, thank you. We'll get our honcho to put flashing neon signs on the YouTube.

Cool. Wonderful. All right. Thanks so much.

Cool. Awesome. Thank you.