This AI Tool Automates Any Task in 60 Seconds (Live Demo)

Hey, welcome to the Next Wave podcast. I'm Matt Wolf. I'm here with Nathan Lands. And today we're going to talk about probably one of the hottest topics of 2025, and that's the topic of AI agents. Today, we're bringing on a special guest, Dimitri Shapiro. He is the CEO of Mind Studio. And Mind Studio is a tool that makes it insanely easy for you to build your own AI agents. So in this episode, not only are we going to talk about AI agents and what they're useful for, we're going to actually jump into mine studio with Dimitri and show off him actually building some of these agents in real time and making them work. It's pretty impressive stuff and I don't want to waste your time, so let's just jump right in with Dimitri.

Being a know it all used to be considered a bad thing, but in business, it's everything. Because right now, most businesses only only use 20% of their data. Unless you have HubSpot, where data that's buried in emails, call logs and meeting notes become insights that help you grow your business. Because when you know more, you grow more. Visit HubSpot.com to learn more.

Before we get into talking about Mind Studio and all the cool stuff that it can do, let's get a little bit of background on you. I know you and I are practically neighbors. You live maybe three miles from me. We've met up for coffee before, and when we hung out, you told me about all of the sort of stuff you did before you built Mind Studio. And I just found it fascinating because you worked at MySpace, you worked at Google, you did all sorts of really, really cool stuff, and you've been sort of in the thick of the AI world for quite a while. So can you give us that quick rundown of what you've done in AI and in the tech world so far?

Ah, totally, yeah. Great to see you again. And I love being your neighbor these days. I describe myself as an old nerd. I started writing code in 1984 when I was 14 years old and I went to see the movie War Games. And I came out, I was like, I gotta get a computer. And we were like, dirt poor, so we couldn't afford a computer. But my high school had two Apple 2e computers. And so I spent the rest of my high school years instead of being a normal teenager in this, like, computer lab. And you're mostly trying to, like, write code and cause havoc in whatever way.

That I could start World War 3.

Mostly like troll my friends. You Know versions of like very primitive viruses and things like that. And then I studied electrical engineering, got a degree in electrical engineering. I've never done a day of it, always done software. And then, you know, had a long career in software before, you know, founding Mind Studio. I spent four and a quarter years at Google on the main campus there I was on the product side. I started on Google when Google felt that Facebook was an existential threat. And like all the efforts went into standing up what they called the social spine or the social backbone of Google. That's what was called Google make all of Google Social. And my job was to build the social graph which is like the most important part of any social network. And even though Google has tremendous amount of information on users, it was actually struggling to put together things like you may know, you may like feed rankings and things of that sort. And so my job was to run product on these like three mostly machine learning teams that were crunching through all this Google data and trying to sort of like compute these lists and also creating a signal to abuse because whenever you open up social networks, you know, abuse goes crazy. And, and all of that as a function of that. Before that I was the chief technology officer of MySpace Music. This is the joint venture between MySpace and the Record labels. So I did that. Before that I built a venture backed company called Veo Networks which was one of the major competitors to YouTube in the first five years of YouTube's life. In fact we had video recommendations before YouTube had video recommendations. So that was very early and of course that was all machine learning stuff. Before that I built a venture backed cybersecurity company, enterprise software cybersecurity company called Akonics raised $34 million for a Conix, raised $70 million for Veo here so far we've raised $36 million, like raised 140 some odd million in my career and then whatever. Before that I spent five years at Fujitsu and other big Japanese company like building their web team. So I'm an old nerd. I love building products. This is the most fun I've ever had, which I think should say a lot because I've been involved in a lot of like really, I thought like really interesting and fun things but like wow, what, what a time to be alive and working on products, to be able to like leverage like these new capabilities. It's crazy.

Yeah, absolutely.

You raised a lot of money. Yeah. Over three different projects. Right. Was most of that from Western VCs like Silicon Valley VCs or like strategic partners or like where did most of that come from?

Both. So akonics. I raised $34 million. The vast majority of it was from VCs. Top tier in VCs. Like Manlo Ventures was in there, which is like top tier VC and sort of second tier VCs. At Veo, I raised $70 million. A lot from VCs, but also like Intel, Adobe, the two former heads of Viacom, John Dolgen and Tom Frost and all of Siffran Brettenham law firm, which like the most important law firm and entertainment. Michael Eisner, the former head of Disney. Michael was on my board of directors. So there was a lot of strategic. Time Warner was an investor. So a lot of strategics in that deal along with VCs. Here it's primarily VCs. And lots of angels, lots of like senior Googlers. Because in my time at Google I knew who I was. I left, I started doing interesting things. They all sort of showed up and said, can I ride with you?

Yeah. I've heard there's like a huge angel network of like ex Googlers.

Yeah. And so, yeah, so it's, it's a bunch of nerds, you know, people that like me are mostly nerds. Yeah.

Well, so now you've got Mind Studio. And you know the, the big buzzword of 2025 is agents, right. Like, everybody's talking about agents. You hear OpenAI and Google and all of these companies basically saying agents, agents, agents. This is what you're going to hear the most about in 2025. And Mind Studio to some degree is basically a tool that helps you build these agents. But I'm curious, like, I feel like everybody kind of defines agents differently. Like how do you define agents? What are you calling what you're building at Mind Studio? Agents. Like where do you stand on all that?

Yeah, I think there's a tremendous amount of sensationalism and hype and noise and I'm mostly allergic to all of it. And so even though we call them agents, we're kind of allergic to the word agent because it's, it's so overloaded.

Yeah. But it's kind of the word that people know and it's the buzzword right now. Right.

So I'm not sure what else to call these things. For a while they would call them workers just to differentiate from agents. But here's how we think of them. I think differently than majority of the people when they use the term agent. I think kind of the popular definition of it is an AI that can sort of direct itself and do what it needs to do in sort of quite amorphous situation. It has access to tools and it has some goal and it goes off and does that magically. It's intelligent and it's able to do that like a human would. Like, I'll figure it out. Basically it takes the approach of like, give me some task, I'll figure out like what to do about it. And as a nerd, like I'm excited about that kind of vision and certainly we're all moving in that direction. But at the same time we think that's a sort of a bit too much autonomy to be just given to AI and that like for most things that you want to do, you generally want to be able to be in like more control. Like you want it to do the heavy lifting and the scaling, but not necessarily to do like all the strategy, for example, for you and like choosing what path to take and you're sort of telling it where to go and it goes there. Rather like it takes you for a ride to wherever it is that it decides to go. And so we think of these things as being workflows that you can say, I want to get something done, I want to create something that does something. And well, how would you do that? Well, you create a workflow like same way that you would sort of like onboard a human to do it. You would explain to this thing, here's what I want you to do. Like every Weekday morning at 9am, I want you to go and search the web and search these other databases looking for developments in this thing that I care about. And you can be quite fuzzy in saying what the thing is, or you can be quite precise. And then if you find that kind of situation, if you find interesting information there, do some thinking, you know, synthesize it, evaluate it from my standpoint, you know what I want and then sort of create a nice report for me and sort of nudge me, send me an email, send me a notification, tell me, hey, something just happened in the world that you should know about. I've thought about it for you. Here you go. And you can just set it on a schedule. That's like one of the ways that you can deploy these mine studio AIs. And it just runs and does that for you, by the way, that's amazing. Like I'll give you like an example here. Like, so Google Trends is a really sort of important living data set of what are people searching for? Like what does the world care about? Or what is like the world that you care about. What do they care about? And more specifically, like, how is that changing over time? Like, what's becoming more popular, what's becoming less popular? So the information is there on Google Trends, but the interface for you to like, load it in your browser and like periodically go look at Google Trends, that takes a lot of work. And so you're going to look at like one or two keywords, but you're certainly not going to look at 100 keywords.

Right.

You certainly wouldn't look at 1,000 keywords. But you see, with a Mind Studio AI agent, you can look at 1,000 keywords and you don't have to do any of it. You just have to go and say, when you build agent, like, this is what I want to look at. It can generate the keywords, or you can generate the keywords and it can at scale every day or every hour or every minute you decide, go and look at the world and synthesize it and look for those changes you want. If you're trading stocks, that's an unfair advantage. Right.

If you're starting companies as well, that's a big thing with like product studios.

Is like, it's an unfair advantage in everything. Yeah, exactly. Yeah, yeah, yeah. And so again before now, like that, how would you do that? But now that's actually really straightforward. Or you find something online, you're reading, you're consuming, but it's hard to tell whether that thing is accurate. More and more things are inaccurate. Misinformation, disinformation. Now we've got a little button you can press and we've got a Chrome extension. And you can have these agents triggered from any web page. And so you can push a button and basically takes that web page and then does deep research, like deep Google searches using very sophisticated Google dorks, Google search operators, and compiles all the data, evaluates and can return back to you. Like, these things are true, these things are misleading, these things are false. And if you're making business decisions on that or personal decisions, it's valuable to be able to have that kind of superpower. That's actually kind of the way I've mostly been describing these things when I talk about them is I call them superpowers, like AI superpower.

Right, right.

Because that's exactly what they are. And they give you an unfair advantage if you've got the right ones and you use them in the right way.

Yeah. Well, let's dive in and actually take a look at mindstudio and some of the stuff that you're building, you know if anybody who's listening on audio, we are going to be sharing this on YouTube as well. So you can go and check out what Dimitri is sharing. But I know you guys just rolled out a pretty new feature that works inside of Chrome, where you can build all these automations directly inside of Chrome. And so I thought it'd be kind of cool to show off some of the types of agents that you're building right now and some of the, like, really simple workflows that you've built to, you know, just make your life easier.

For sure. For sure. Yeah. So I've installed the Chrome extension. This thing is an alpha now, and soon it'll be in the App Store. It's been submitted and it's here. I can drag it around. It can be collapsed or it can be expanded. And these are things almost like triggers for AI agents. So here's a TLDR AI agent. Here's an AI agent called. What is this? This one's called Alternatives, Right. And I can trigger them, or if I want to have these things collapsed, and I can click on this button and it gives me the side drawer. And in it, you get all the agents that are sort of available in our agent store. You get all the agents that you've created yourself. You've got agents that you've starred. So these are like your favorites. And so you can very quickly again take an agent and trigger it. And so, like, what could we do here? Well, one thing we could do here, for example, is do like a tldr. And so again, I can trigger it from here, or I could have triggered it from the extension right here that I've got. And it instantly again grabbed the content on this page and gave me this quick summary. And this quick summary was just a quick AI model looking at this thing. And it did this and extracted these entities. We also have a thing called who's Involved, which is more sophisticated. It'll go grab the page and it'll find all the people and entities on that page and ask me what people or entities I'd like to know more about. Like this theoretical physicist. I can click this, and this is going to kick off, like a really sophisticated workflow that could take, in this case, I think takes about two minutes to run, which is again, generating Google search queries over here. Again, you can sort of see it working. It's executing those Google search queries on Google News, Google images, blah, blah. And then it's going to create a report for us. And by the way, we don't have to wait, we can say, notify me when finished. And this thing is going to send me an email with this like really sophisticated profile executive summary, you know, of this person here that I've done. And again we don't have to stay on this page, we can keep moving. So you know, here you are and I've got a little thing over here where I've got a bunch of buttons. One of them is YouTube TLDR. Right. I can click this thing, it's going to grab the transcript, it's going to process it and it's going to give me a TLDR. Now your video is only 33 minutes here and so I could have watched it and you know, two times speed, that's nothing at all. But some videos are, you know, three and a half hours, you know, all of it is also fine tunable by me. And so if I want my TLDRs to be different, I simply go and I edit this AI agent that I have and then I can use it.

Gotcha. So many of these things like when you were just showing you had the little like sort of tray on the right side where you had a whole bunch of little like apps that you built. So my understanding is most of those apps are like pre built templates that are available in mine studio that you can just add to your sidebar but then also you can create your own and you can also modify some of the ones that are already pre built. Is that kind of the idea?

Yeah, let's talk about that. So again if people sign up and they don't yet have anything started or they haven't built anything, they will see on this agents tab they'll see these agents.

It's almost like a little like app store for agents.

It's like an app store for agents. Exactly. Sorry, these ones right here. And there are more agents coming. Again, this is an alpha now you should expect hundreds of beyond that thousands. And then on any one of these agents you can open it up and it's got sort of like a detailed page and it tells you how much it costs to run this. So on average about 6 cents in inference. This one takes about 45 seconds. But again it returns back like these really awesome summaries, like reports in a sense that are publishable, that are shareable. They're like pieces of content in a sense with just clicking a button. And so you can run it right from here or you can run it from the extension. You can just click a button, YouTube TL Dr. And run it. Now the other thing you can do with these things is if you click here, you can say make a copy. And when you make a copy, you've just made a copy of that agent and you have access to all of the. It's not code.

Are you like forking it?

Basically you're forking it. Exactly. So agents are made out of these things called workflows. And these workflows are basically these things where you can add these little nodes to. And so it's got a start block and so let me make this a little bit smaller. Let's take a look at what this one is doing. It's actually quite simple. It's got a start block you can configure the start block to. Again, how does it run? These things can be run on demand, meaning you trigger them manually. They can be run on any kind of a schedule that you specify. If you click email here, it gives you this custom email address. And whenever you send an email to this email address, it takes that as input and triggers this workflow and passes it variables. Or you can trigger it as a browser extension, which is what this was, and then you get access to these launch variables, as we call them. So when this thing gets called by the browser extension, the browser extension passes it the URL, the metadata encode the metadata, name, description, image, et cetera, the full text which is visible to the user. So stuff you can see on the page, if user is selecting anything, highlighting anything or right clicking anything, it gets passed in this user selection variable. And then like all the raw HTML, CSS, JavaScript behind the scenes. So it grabs that and that's available to us and we can reference that using variables. And here we've got different blocks that you can add. And I'm going to go here you can see there's many different types of blocks like fetch YouTube captions. Like this is how you can grab the captions from a YouTube channel or you know, fetch the channel and metadata about it, or fetch all the comments.

This is all if you want to customize it, right? Like I assume you're like a regular person would not have to do all this capabilities.

Exactly. Yeah.

There's a whole bunch of pre built ones that you can just use by default. You can go build one from completely from scratch if you want. Or you could take one that already exists and sort of fork it and iterate on it yourself.

Right. And so okay, so this is like one. Let's take a look at another one here quickly, let's take a look at this researcher that we have that people love. This one takes an average of almost four and a half minutes to run. Again, you can run it from here. If you run it from here, it says, what do you want to research? I can say I want to research Matt Wolf and then go off and do the research there, or I can run it from Chrome extension and so I could be whatever on cnn. And then I can click this thing and I can go to researcher and run it from here.

And.

And then again it's going to do what that people indexer did too, which is like grab a bunch of things off this page and say, which of these do you want me to research? And I'll present you like a menu. You know, emerging technologies in AI investigating recent developments in artificial intelligence, blah, blah, blah. I can click this and this thing is going to kick off this like very sophisticated workflow. Let's take a look at what that workflow is actually doing. Again, I'm going to say make a copy, which means fork this. You make a copy of this. And so as I said, these things are made up of these workflows. The main workflow here, right here is called main, and it's made up of these automations. What is this thing doing? Well, when it gets triggered, it gets passed into it a variable called request. That's the user's request. And then the first thing it does is it calls a model, in this case 3.5 Sonnet. And again, we support over 50 AI models with new ones being added all the time. All the major providers including Deepsea, Google, OpenAI, Metal, Google, Mistral, whatever, they're all multimodal. So you can call any model you want. And then in this case it says, analyze this request and figure out a plan to use Google Search in order to do that. And then store that plan in a variable called plan and then move on to the next step. And in this step it says here's the plan that was made here. Using this plan, create a bunch of Google search queries to, to be able to research this plan. And then it uses another type of block, which is a run workflow block that triggers another workflow. And this workflow called RunQuery, and it passes into it an array of queries. And then it runs this workflow here over and over again and searches Google and then comes back with all of the data from that. And then it normalizes everything and then it generates the report and, and then it runs another workflow to generate images and then it outputs this awesome report and emails it to you. And it's available for you whenever you want. And so again, you don't need to know any of this stuff in order to use it. Right? In order to use it, you just push this button and say run. And it creates this stuff for you. But you can edit it and do it. Yeah.

Dmitri, off camera, when you talked about the Google trends thing, that was like, honestly the thing that I found the most interesting out of everything you've said. Like just imagining how that would work and I can imagine so many different scenarios where that'd be useful.

Or monitoring your competitors websites and messaging.

Yeah.

Again, the AI can just be checking all your competitor websites and as they change their messaging, which they do, the AI can be sort of analyzing like which way are they going? What might it be implying?

Yeah, why are they doing that?

All the stuff that like you wouldn't have time to do and do it.

It's going to get kind of funny though if every company's like following trends like that. Like oh, based on some data, we changed our website copy and then, oh well, they're changing theirs and then they're all like changing in unison.

Well, it's funny because it's like whenever like OpenAI or Google or Microsoft or one of these big companies like makes a change to their terms of service, it's news, right? People are like tweeting about it and things like that. And it's like you can actually have one that watches for changes of terms of service on AI related companies to see if now all of a sudden this company is going to spy on me even though they told me they weren't originally, you know, right here, simple.

To do, use case trivially, easy to build. You can build it in five minutes. Doing this. Okay, so I'm on your profile on Twitter X and so these things can be contextual. And so when I'm on Twitter the agents that are relevant to Twitter can show up right here. And so one way I could get to know Matt Wolf is by going through and like spending a bunch of time like looking through your profile. But ain't nobody got time for that. And so I've got a thing called Profile Analyzer and it just grabs that and the AI is going to analyze you like it builds up a bunch of things, you know, your primary focus, sort of content themes, content types, key behavioral insights, you know, so again like a bunch of stuff about you. I'm in control, I can modify this thing to do something different because you know, I want different types of things. Or there might be conversations that are happening on Twitter. And again, I can try to like read through this and understand this thing, but that's really hard to do. But I can just click this little button which is X Conversation Analyzer and it can do like all this work for me. It give me again these TLDRs and, and these are artifacts. Like I can share them, I can save them, I can send them, I can do whatever I want with them in order to be able to do that.

Yeah, one cool thing too that I don't think you mentioned, but I know I've seen it playing around with Mind Studio. Is that the sort of visual workflow that you were showing? You can actually use natural language and tell it what workflow you want and using natural language, it will actually go and automate the building of those workflows as well.

Oh, that's awesome.

Well, again, we can make yours here by saying Create Agent. There's actually I wouldn't even use this to do it, but like for somebody novice, they could do it. So again, you can start from a blank project or you can start from this thing called Mind Studio Architect where it says, what do you want me to build? And I'll build it for you. By the way, this Mind Studio Architect was built using Mind Studio. It's just a Mind Studio agent whose job it is to, to be the architect of Mind Studio. So it knows how Mind Studio works, it knows all the blocks that Mine Studio has, it knows the kind of the design patterns that Mine Studio prefers. And then if you tell it what to do, it says, I know how to follow these instructions. And so again, we could say every day, monitor my competitor's website and check terms of service. When they change.

Notify me with a summary of changes. It seems pretty straightforward. I don't need to. Again, you can do that and say generate and this thing is going to go off and it's going to create a plan of how it's going to build this thing. Because there's a bunch of ways you could build it. You can say, I don't like this plan, choose another plan. Or you can say, great, this plan looks good enough for me, push Go. And then it'll build all of the scaffolding for you many times. First time it builds it, it runs well and you don't need to fix anything. So that happens. Sometimes it makes some weird error and you gotta troubleshoot it a little bit, right? And we've got a little button where you can ask people for help and they can that are more experienced than you in doing It So you can do that. But while it's doing that, let's open up another tab and I just wanna start another one here quickly. I wanna show kind of the simplest one that I think will help people grok this maybe better than like some of these ones that I've been showing. So I'm just going to go to build. I'm going to say create new agent and my development, I'm not going to use this. I'm just going to say open blank project. So the system prompt, you can add these or not. I tend not to do them, but if you want to do them, this is again sort of tells the system like what you want it to be. You can write it yourself or again, you can use another agent that we have here to generate a prompt.

I'm still thinking about the whole Google Trends thing. Sorry, my mind's stuck on that. I know. You know another HubSpot podcast, my first million. They've had so many episodes. Talk about, hey, if you want to start a new company, one of the main ways to do it is to look at Google Trends now. You can just automate, you can be automating.

I'm glad that that connected.

Yeah.

Because that is a profound superpower.

Yeah.

It sounds like you can actually leverage Google Trends, leverage that data at scale. It's profound. It's an insanely unfair competitive advantage with anybody who can't do that. Yeah.

Sorry to take you off track. I just like, that was still my mind. I'm like, I want to go try it.

Okay.

So let's build the Chrome extension thing. Okay. I'm going to go here, start. I'm going to browser extension again. We now get these variables. I'm going to say plus and I'm just going to say generate text. This means column model. In this case. I'm just going to change a model just to change it. Why not? We're going to use Gemini 1.5 flash and we're going to tell this model, create a TLDR of this. Okay. And then in double curly braces. That's as complex as it gets. We're going to pass it this variable here that we get again called fulltext. Full text means the full text of the page from which we called this. There it is. It's case sensitive. And here we can either display it to user or save it to variables. I'm just going to leave it display to user because we're done. We just made this agent that does a TLDR again. If we want it to be Fancier. And we want it specifically say how we wanted to create the tldr. What should it look for? What should it buy us for? Like you just add this in plain English over here, right? And then you can click here and you can give it a name. And I'm just going to call this Maps tldr. And then you can give it an icon if you want. And then you can just say publish. And so again, we just built this thing. It exists. It's got a URL. We can run it. We can run it manually and just give it a URL. But that's no fun. That's not how we built it. We want to run it from whatever this one here. And so we can go here and we can open our extension right here. And like a fast way to find it, because I've got so many of them, is to simply do a search, Matt's tldr. And I can just run it from here. And again, we built this thing in what, less than a minute here. And this says Microsoft made a major breakthrough in this. Okay, that's not a very good tldr. We could go back and say, no, no, like these are specific things that I want out of my TLDR and do it again. I'm not going to demonstrate that because anyone can do that. You can play around and then you hit save and you go back and you test it again. Or you can go back here and test it again. And it's that easy to sort of like iterate on these things to create custom tools for yourself.

Yeah. And then you can just go and test different models too. Maybe Gemini 1.5. Flash tends to give you smaller responses where if you were to use, you know, GPT4O or Sonnet 3.5, it'll give you a much deeper response.

Absolutely. In fact, let's do that. Let's use Sonnet here instead. And we're going to publish that and we're going to rerun the exact same thing. And again, like it's super easy to rerun these things. I should take the one we just made so I don't have to look for it. MATL. Dr. I can click on this thing and over here I can say add to start. Okay, so this one didn't do much more on the tldr, but you can change models again if you give it some more instructions. It doesn't matter. A much better job. Another way to get it to do stuff like that, that's even better. I can say generate prom. You know, we like to refer to this thing usually as Assistant, but you can say anything you want, write it in any form you want. Assistant generates TLDRs for a human when provided with content. And I could say from the web or I can say content or whatever and like say generate and this thing will go off and it'll like generate the prompt that will be much more specific about what a TLDR should have. And that tool now bias the TLDR that we get on that. But the point is like all of this stuff is like really easy to do and can play around with.

Yeah, for sure.

Now what's up with the one that you started using the automation? You use the natural language to build that one. Let's check on that one.

Let's check that out. Okay, so here's what it did. It built the terms of service change monitor. Again this is the plan. Automatically monitors competitors terms of service page for changes, analyzing differences, sending email notifications. It says it's going to take, you know, it's input the URL in a notification email and then here are the variables it's going to use. It doesn't need any custom functions that it needs to write. This is what the automation is going to look like. It's going to scrape the terms of service, it's going to clean and normalize content, check if it's the first run or if it's got data about it already, save the version, end the first run, analyze channel and we say great, whatever this means. Sounds good to me. And then you say go and it'll go off and it'll build that entire workflow. We'll get that workflow again. It might work with just we obviously need to put in the URL because we didn't tell it what URL, we could have told it what URL. Okay, so workflow generation is complete. I can watch a YouTube tutorial on how to use it or we can close and start editing it again. It built the system prompt, it built the automations for us that we can then look at and we can edit. And so this is using our scrape URL block and it's taking in here again a variable called URL that we need to define. So here it says this thing is scheduled and it gets passed these like launch variables or if it's the same URL, we can just put it in here explicitly so we can say, you know, cnn.comtermsofservice you get the point of that, right?

Right.

Yeah.

Dmitri, I'm kind of wondering the more you talk about this Long term. I'm curious, like, how does this, you know, obviously for OpenAI agents is going.

To be very, very important.

Like, you know, and I'm kind of curious, is your thesis that even if they really, really focus on agents and of course they get amazing at it, that there's still going to be a benefit of having like a community of people sharing their different use cases and different templates? Is that kind of the idea behind Mind Studio?

Yeah, look, we believe that models will continue to get better and better. And certainly model providers like OpenAI will have sort of consumer interfaces along with API interfaces. And those consumer interfaces, you know, today are primarily conversational. So chatgpt and it's got access to, you know, tools, function calling, et cetera. And like in your chat you can sort of go back and forth with it and it like does stuff for you and there's a lot of use cases for that where like ad hoc, you want to do something and you do that. But for, you know, many other things, and I propose most other things, you will say, well, I wish like something just like I could build it. And it just ran and it did these things like monitor Google trends. I don't want to constantly ask ChatGPT to do it for me or talk to it. Right. I want it to only notify me by the way it does it in the background. It should only nudge me if it found something interesting. Most of the time it won't. And so it just should just be there in the background. I've got a dashboard, I can see these things running, I can see what's scheduled, et cetera. As for like the capabilities of the models themselves to sort of be at a place where again, you kind of don't need anything like this and you just tell them, I want you to be this thing. And it sort of does it. I struggle from a product side to see how you would sort of have the human feel enough in control without sort of creating something that they could understand. Again, you might generate the workflow for them just like our architect did. But I think humans still being able to say, oh, I want to put like a checkpoint in here.

Yeah, modify how it works, except for.

This, or I want to be able to change this in some way. You know, I didn't show you guys our debugger, which is awesome, like, extraordinarily powerful. So, like, our instrumentation for like building and running these things is dramatically deeper than any of the model providers themselves offer. And so if you're going to build anything with models, we Think this is the no brainer way to be. It's the most instrumented, it's the fastest, it's the easiest, gives you access to all those models. Models go up and down.

Yeah, that's what I was going to say. It feels like that's a big advantage too. Right. Is like being agnostic because I mean, in the last month we had Deep Sea come out. Now Grok's amazing. There can be a lot of options out there.

Well, and as you're building these workflows again, you want to, for each step where you want to use intelligence, you might want to call a different model. Right. Where you know, Gemini 1.5 Flash is super fast at being able to do a bunch of stuff, but it's not a very good writer. Claude is a much better writer. So you might get the Gemini to do a bunch of analysis, data collection, all of that, and then sort of hand that off to Claude to put together and do or like generating images or whatever else is doing here. So this ability to analyze different models for each part of the step, use the right model based upon quality latency. So low latency and cost.

Yeah. And you probably could suggest that eventually too. Right. This model's better for writing here. You might want to use this one.

Yeah. Today it's like suggested in those detail pages that each one of them has. You can read them. But sort of where I was going is like these models aren't reliable. These model providers are still building their infrastructure. They've got a bunch of usage, so they go down periodically, in fact, quite frequently. And if you're just like standardized on only OpenAI and all you've got is there, you're kind of stuck. Your AI ain't doing anything here. You can just fail over to any of these other models that could also do the job. They might not be preferred, they might be a little bit more expensive, they might be a little bit slower, but at least you're not down in your workings. Like all of those kinds of things become really important when you get serious about building these things.

So are those like redundancy mechanisms? Are those built in already? Like try Claude 3.5. If that doesn't work, then go do GPT4O or whatever.

Today it doesn't automatically do it, so it can notify you that models are down, which we do, and then you can manually go do that. But coming soon it will do it intelligently and just like do that for it.

Yeah, that's cool. Like here's my tier list, you know, for Writing for coding for whatever, use it in this order.

All the infrastructure is there. You can now just manually choose a different one. Takes, you know, 10 seconds, but soon it'll be autonomous for sure.

Yeah, I did have a question. This was something you showed off on the screen earlier in our conversation where you showed, all right, it's going to cost like $0.06 to run this, right? How does that work? Is it like a bring your own API key? Is it like you guys bill, based on the usage? Where does that sixth sense come from?

Yeah, so you don't need to bring your own API key, although if you want to call your own models, we allow again, function calling and API calling. You can call any model, including on premise, private, cloud, et cetera. The vast majority of people that use mindstudio use our API keys that are built in. So you don't need any API keys. So you instantly have access to all of the models from all of their providers on our billing and we simply pass the costs on to you based on usage. Okay, we take a tiny markup, 2.9%, and then it reduces with scale and goes down to like 0.15% of overage over what the model providers charge us. It's kind of like Stripe. We kind of see ourselves very much in that light. We're sort of like what Stripe is to payments or Twilio is to telephony or MUX is to video, Mind Studio is to intelligence.

Makes sense. Makes sense.

Do you ever consider doing it more like cursor, where like you have like a monthly and that's basically baked in. Like you're kind of assuming they're going to use the API some amount with some limit. And then you go over that. It's like, okay, you got to either pay us more or use your own API key. I always like that with cursor. Then I don't have to worry about doing something and like, oh, right now I'm paying something. I'm like, oh, I'm not paying anything. I'm just using it, you know, and so that's kind of feels nice to me. I don't know.

Great. Well, I'll break the news on your show here. We're about to add that. So.

I can predict the future depends.

On when this publishes. It might already be there. Yes, exactly. That makes sense. So I wanted to mention sort of another thing. So we've got this, Matt, as you noticed, this is like an app store and these agents are put in there today by us. We created all of the agents that are available for you to use. You just pay metered usage form, they're free to use, but other than metered usage, you can make copies of them and modify them and launch new ones, etc. And we will keep adding more to it. The other thing that we're doing is we're opening this up to the world just like the App Store has opened up to the world that people can show up and say, I've got some use cases that I think are awesome and that lots of people are going to want. I'm going to build them in mine Studio and again, as you can see, it's easy to build and then I'm going to submit them to the store and I want to make them available to other people to just use. Eventually also like charge form, obviously we'll implement that as well. And so we're actually looking for people that want to show up and start building these things. And we're venture backed and so we're willing to pay like bounties. So there's an opportunity for enterprising folks to get in early and fill up this App Store with all kinds of amazing utilities that now can be triggered from a browser extension if you're building good ones. We've got, you know, praise and money to make your life better. We're also looking to hire people full time to do this. There's like great opportunities for people to start building these things and taking advantage of it.

Very cool.

When you say App Store, is there going to be like a revenue share or.

Absolutely, yeah. You'll be able to charge for these things and you'll be able to earn revenue from the work that you've done.

Yeah, yeah, very cool.

By the way, people are already building these things and putting them behind paywalls, which you can do. And so there are entire businesses that have been started that simply build a bunch of Mind Studio AI agents, put them behind a paywall and then charge monthly access to all of those agents.

Yeah, I remember early on, I think when we first Talked, you know, 18, 19 months ago or something that was kind of like the original sort of selling point of Mind Studio is go build agents and then sell access to them.

Yeah, so people took us up on that and are doing it. And so yeah, you can make money with them already. By the way, we also now have over 150 integration partners that make money by again, they're experts in Mind Studio and they are building AI agents for organizations. So we've got over 150,000 of these things that have Been built and deployed. Again, some are just like experiments, they're toys. But many are like mission critical, really important AI agents that sit at government agencies. The His Majesty's Revenue and Customs, the British IRS uses Mind Studio and recruiting, you know, ServiceNow uses it in sales. Demo automation company called Advanced Local, which is a giant publishing company, has got over 80 of these running that are automating newsrooms and newspaper publishing. And so some of those are built by the people inside these companies. Others, these companies hire our partners to do professional services work and build them. So again, there have been entire actually agencies that have like started up simply to do Mind Studio implementation for organizations that want the skill. That's where all the budgets are flowing, you know, in enterprises. They're all flowing towards this like AI digital transformation.

Right.

And we make it really easy for them to get this like giant ROI out of very little work, you know.

Very cool, very cool.

Absolutely. Is there anything else about Mind Studio that we should touch on that we haven't touched on yet? Or even something outside of Mind Studio that's really, you know, got your attention right now that you're really excited about? Just in the AI or tech world?

That's a deep question.

It was like another podcast, right? That's another hour.

Yeah, no, I mean, I'm assuming your head's pretty much always in Mind Studio, playing with it, making it better. So, you know, you may not be going and playing with all the other tools and, you know, stuff that's out there.

Well, I get to watch your videos and get to you doing it. I don't want to make it political because there's a bunch of interesting stuff there. Like, what's a fact checker? Well, fact checker is a really important tool that, you know, citizens of democratic countries need to be able to understand what's reality. And now that tool is a click of a button. Like it has broad implications of what that means.

But yeah. Yeah. Well, there's a site called Ground News. I don't know if you've heard of it or not, but it's like a news website that curates all of the latest news from like all of the various news sources. And then when you're actually looking at the site, it'll tell you, does this article and this news source tend to lean left? Does it tend to lean right? Does it tend to be a little more center? And it sort of like analyzes every piece of news and tell you, like, where that news falls on the political spectrum? And I feel like with Mind Studio, you Can build that, right? Go to any news website, click a button, it'll tell you whether this site tends to lean left, tends to lean right and actually look for bias in the news article for you. Right? Like you don't need to get political. Doesn't matter what sort of political leanings you have. That just seems helpful to anybody.

Pick your bias.

Yeah, we have AI agents that have already been built to do specifically that that are sentiment analyzers and bias analyzers. You push a button and it looks at the article and says, hey, here are things that might be inaccurate, here are things that are misleading, here are things that are clearly biased where they're showing only one perspective, there's others that say counterpoint. And so it'll be like great. What's the opposite of that? Like what would be the counter arguments? And so again these superpowers that you now get with click of a button that AIs can do really easily, but once they're like made so easy to take along with you, that's kind of the big unlaw. Wherever you are, you now can like instantly trigger them and get that instead of like going to ChatGPT and asking it to do something. Right.

Super cool. Now with the little like Chrome extension to any plans to make that available on mobile as well. Like will I be able to be reading a news article on my phone and do the same kind of things?

Yeah, we're going to have a native app probably, I'm guessing in a quarter which will allow you to basically make it a shared target. So you can like intent out from any application to any of these agents and get callbacks on that. So like it instantly sort of enables anything that can do share intents to now have gotcha available for it. There are some, I can't sort of announce them yet, but there are medium sized social networks that are starting to integrate MindStudio agents into their interfaces. So they will be within their native mobile apps. There will be buttons on their posts that allow you to fact check things or like do other things. And, and those are simply Mind Studio agents that are doing that work. And so you'll see, you know, certainly a lot more of that kind of stuff that you can do.

Well, cool.

This is, this has been a really, really interesting conversation. I mean I love, I love actually getting to see all this stuff in person. I've been playing around with Mind Studio for, for a little over a year now and have messed with it quite a bit. But it's always really cool to see how like the person who's actually building the tools would actually use it as well. So I really appreciate all the insights and you showing off how how to use it. So the website is mindstudio AI. Is there anywhere else people can follow you? You on x, you on LinkedIn. Where's the best place to sort of get to know you?

I'm on all of them. I prefer LinkedIn. And please feel free and add me on LinkedIn. I'm easy to find. Dimitri Shapiro and yeah, that's probably the best place to follow me. But I'm also on X and Facebook and Radit everywhere else. You gotta be everywhere.

Very cool.

Well, thank you once again for, for joining us today. This has been a really cool conversation. Anybody that's listening, if you haven't already, make sure you subscribe on YouTube for more interesting conversations like this, especially if you want to get the visual element. And if you prefer audio, you like listening in your car or while out walking your dog. We're available wherever you listen to podcasts, so subscribe to us there. And thanks again to Dimitri. It's been an amazing episode.

Great to chat.