Tiny Fish raises $47M to turn the entire internet into a structured database for enterprise web agents

Ads account to Julius now. Run some numbers there. Love that. Love that. Anyway, we have our next guest in the reream waiting room. He's here now in the TBPN Ultradome. Welcome to the stream. How are you doing? I am doing well, Jordi. And John, thank you so much for having me here. Thanks for joining.

Would you mind kicking us off with an introduction on yourself, your company, any any news you got for us? My name is Sudish. I'm the CEO of Tiny Fish. First of all, I'm very happy to be here. First time I got to know you was saw Parik by the way. That was a great guest. That's amazing. That was a great guy.

What a wild day on the internet. Uh and how fast it's all gone too though. I mean it's all gone. So yeah. Yeah. I mean people were we never got the full postmortem. We'll need to check in. But he said he was going exclusive. There's somebody named Sohan Periq.

Exact same spelling that's been appearing on AI research papers. Really? People are trying to figure out is it the same guy or just I saw that too. Yeah, I think DBP needs to have an investigative arms.

Yeah, I actually did talk to someone that was thinking about hiring him after the fact and it seemed like it was not uh not enough of a red flag and they said they were going to give it a shot and hope that he I thought he was going exclusive. Yeah. Yeah. Yeah. Yeah.

And so and so as far as I know he is still exclusive but I haven't done any research recently. Anyway guys thank you so much well you both have built something in the last 10 months entrepreneurs investors. We appreciate it. Thank you. Thank you Tiny Fish.

We just announced our 47 million funding series which I was not ready for that. I was not ready for that. Absolutely. Needed a big number. I needed a big number warning board. Absolutely huge. Congratulations. 7 million is like a seed round, right? Mango seed. It's a seed round for a tiny fish these days. Tiny fish.

But congratulations. I'm sure you can do a lot with it. So what are you going to what are you going to do with it? Look, I think we are going to invent a category, build a category that makes sense and build a great company. That's basically it. Prior to this uh you know I was at uh thought spot and nutanics.

We have done some large enterprise companies co-founders. They all came from Meta and companies like large. We want to build a large company doing one thing which is to make web browsing like humans do but a superhuman scale for solving enterprise problems. We don't want to be a consumer company.

I know you both have built companies. There are a lot of AI companies doing things in in in consumer space browser agents and all of that. We want to do uh a massive scale web browsing and turn the internet into a structured database for enterprise customers to analyze and act on.

Do you want to sell this to other software companies that are going to piece it together with other uh APIs to create maybe even more enterprise SAS? Like how deep do you see yourself in the stack?

Look, I think nowadays there is no difference between enterprise business and developers/builders because every enterprise product in the AI world starts as a a builder product, somebody trying it out.

We 100% want to be the best way anyone to connect to connect code to internet because look these models can solve math problems at international math Olympics level but when it comes to browsing it's a mess because browsing is so complex that our brains make it look easy. Yeah, like think about going to a movie.

The the seat that you pick is a function of complex reasoning. If I'm going by myself, I can sit in front and read the movie. But if I'm far away with my family, I want to make the right place, right, you know, contiguous seats. These decisions we make without even thinking.

Not to mention the complexity of the website itself, how the pictures are popping, how there's like a pop-up here, these things need to be solved really well. We do that. So enterprise customers who want to browse mass 60% of all time of knowledge work in enterprise is on browsing. Yeah.

And we think we can automate that through what we are calling enterprise web agents. So, we we heard reports that uh some of the big labs were building RL environments with verified rewards, basically cloning Door Dash, creating clones of Amazon. com so that they could train an agent on the full Amazon.

com web app and then an agent would be able to check out for you. Are you doing something similar? What does data collection look like? What does training look like? What does differentiation look like in the world where uh where these agents uh might need to be rldled on a specific website to really get good results?

This is why I love talking to you guys. You know, it's it's you geeks ask the right question to geeks. So there are two models when it comes to uh these. One is synthetic websites like web arenas and others have built this. The other uh is uh paying money to say let us go learn Door Dash workflows.

Both are good but both have downsides because both really create significant challenges with respect to scale when they come out of that environment.

What we are doing is something very we have a proprietary architecture that we have built that frees us and takes a snapshot of a data website in a way that becomes that exhaust feeds ourselves in a very scalable way.

So yes it is synthetic data that we are generating but it is out of actual web and more importantly most of our use cases we launched our company with a use case with Google hotel as a customer dash is a customer great great I try to give this I try to give this advice to to early stage founders if you if you're going to get a customer early just get Google start with a mag start with a mag 7 why not right they built this AI technology I mean it all started with all you need is attention Yeah.

So these customers are giving us access. There's a customer who has 22 million uh web SKs in the product. There's another customer the 40,000 uh subshops as part of them. Class pass we announced them right.

So these sort of things while we are solving for them is giving us a lot of data and that exhaust is mined by our products for uh annotation and creating labeling that makes art.

The second thing I will tell you though models are good at a lot of things but they usually perform really well when the goal is straight and the reinforcement can be built on something where the context doesn't change. Web is exactly the opposite. Everything is context that is constantly changing.

The goals are changing the websites are changing and the context is changing. So what we have done is we use reinforce learning and models for understanding and like sort of see what the heck the we are trying to do here. Then we separate that and execute that code in CPU at massive scale.

And then we use alpha evolve a version of that to make the code better. So separating exploration and execution is another fundamental thing that we've invented that is going to make this browser agents you know perplexity comet and operator and others that is like forcing and feeding us.

But this is where the execution framework. So that's an infrastructure problem. Think about uh the amount of browsers we have to spin up, the fingerprinting, the anti-bot, all of that infrastructure need to be solved. Interesting. Um how do you think uh the competitive landscape will evolve?

Amazon's launching a bunch of stuff. Uh I'm sure all the clouds have something that looks like a browser agent on the shelf somewhere.

But um we've seen you know Silicon Valley has you know probably hundreds of stories about uh companies that have built things that have been competitive with the big hyperscalers and been wildly successful. But what's your specific plan?

You know there is a quote that uh often attributed to Chasty Puller Marine that we are surrounded from the north, southeast, west better watch out. This is one of the space where it is so amazing because there are three groups of companies that we will be competing with from a tech space.

I mean obviously there is sharehing of dollars. One are the browser agents who are doing one thing like jod going on a vacation 20 minutes he'll book a hotel for you. We want to sell to xedia not to jodi because xedia now want to simulate 100,000 jodi transaction from 50 states or 50 countries. How do we do that?

So browser agents to a certain extent will be trying to expand to that space. I think we want to compete. The other is the infrastructure. So the browser base and others who have been building things we want to build something enterprise grade with the surveillance you know the the governance the control.

So that's the second.

Third is going to be another interesting area where the codegen and the developers who builders who just want to connect their code to internet so it can do amazing things there again there are parallel you guys interviewed parag I do think that we'll compete there on some level not because today we compete but all of these will converge because there will be a few massive companies built on how browser access how internet can be turned into a structured database so exciting time.

We've raised enough only because we want to make sure that we build it in a way that is where we can control our destiny. I have a lot of respect for all of these companies, but to me, you know, if you're not surrounded by competitors, you're not in the right place. I love it. Yep.

You're not going after a big prize, but congratulations. Fantastic answer. Good luck out there. Have a good rest of your day. We'll talk to you soon. I hope so. This was such a short conversation. I enjoyed. I'll keep watching. Thank you so much. We'll talk soon. Talk soon. Bye. Bye.

Let me tell you about Adio customer relationship magic. Adio is the AI native CRM that builds, scales, and