Delegating Executive Tasks to AI Administrative Agents

Hey everyone, welcome back here to Techstrong TV. I've been trying to get this next guy on here now for the last three, four months. His name is Nir Sabato.

Nir's the co-founder and CEO of a company called Catch AI. I became aware of Catch AI, I don't even know if I saw an ad or someone told me, whatever it was, but they're specializing in AI agents for busy execs, EA, admin, AI admins. And I had played with a bunch of them, and the Catch AI came with a free trial, and I used it, and I instantly thought it was the best one I ever used.

And I reached out to Nir, and we've been talking ever since. I continue to use them, let me say that up front. For those of you out there who have ever interacted with me over the last three, four months, and Crystal Williams wrote back to you, that's a Catch AI persona that I made myself with it, and it's pretty cool.

But hey, Nir, welcome to Techstrong TV. It's great to have you on here. Hey, Alan.

Thanks so much for inviting me. It's a real pleasure to speak on camera. No, it's great, man.

So Nir, we're going to talk about Crystal, we're going to talk about Catch AI, but I wanted to take a moment or two to talk about you. The audience likes to know who am I listening to? What's his chops?

What's his cred? Absolutely. Tell them a little bit of yours.

Go ahead. My name is Nir Sabato. I'm based in Tel Aviv, Israel.

And for the last couple of years of my life, I was a VC, I was an early-stage investor. And for the last, I'd say three or four years, just really trying to focus hard on AI companies. I invested in a bunch of them and went really deep into what AI can do in this world.

But eventually, on a personal level, I felt like I can't stay on the sidelines. I can't stay a VC. Life is too short.

You got to do something meaningful and like- It's too passive. Yeah. I was just sitting on the sidelines and seeing- Mm-hmm ...

all the action going on, so I had to do something. And basically, the ambitious goal that we're trying to achieve here in Catch is to create software using AI that is trusted, is sophisticated, is smart as a human being. Now, obviously, it cannot be done across all different verticals at the same time.

Cannot create the professor, marketeer, and any other all at once. We're focusing on real-world execution of administrative tasks. " So, we started with simple things, or what we thought was simple, like scheduling or reminders.

And now we have hundreds of users and hundreds of companies that use us for really all in the admin space, but just a long list of different workflows and playbooks and type of administrative tasks that we're handling for them and freeing up their time to do more meaningful work and more strategic work. Absolutely. And as I said in the beginning, I'm a customer, and I probably have one of the busier, noisier mailboxes in the world, I think, from doing the Techstrong thing and PR companies and companies reaching out.

And I do. I followed the same path you said, Nir. I started out basic scheduling kind of stuff, right?

Mm-hmm. Handling things. But then I started developing more complex workflow with a lot of branches.

If it's this, do that. If it's that, do this. And building it up.

And now I use it to help me distribute because I wear a lot of hats. I'm the CEO of Techstrong Group. I'm editor-in-chief, so I write a lot of articles.

I do a lot of research for special reports. I'm the president of the HOA where I live. I got tons of crap from that I have her dealing with.

And I say "her" like it's a her because I gave it a female kind of persona. But it handles, it really does, it handles my calendar, my schedule, my admin stuff, and I'm looking to do more. We were talking offline.

Travel. Booking travel is something you guys are working on, and that's harder than it sounds. Yeah.

I will say that every executive have their own preferences, their own world, their own small nuances that matters a lot to them. And what we realized is for busy people like you that have hundreds of emails in your inbox every day and your calendar's full of meetings and different priorities and different obligations, and it is actually unrealistic to think that someone like you will stop everything and start writing a very long document describing all the nuances and all the workflow that you want your agent to take care of. So what we did was we said something like, because a lot of people are trying to do those things with open source projects that force you to write MD files and kind of give your agent all the instruction and all the knowledge that they need, which is really cool for builders, but less for executives who work in real companies.

Mm-hmm. So we build Catch, we give it persona, it lives in your messages, you can text it, you can call it, you can speak with it on the phone. And the interesting part is that it learns your preferences and pick up things as you go.

So you start with the small things, but it creates the habit and trust to do more and more. And then- Yeah ... " No, not at all.

The moment you connect us to your environment and give us a little bit of context from your digital tools, Catch can catch up and start serving you in the admin space really, really fast. And we found that this is actually really important for executives. People don't want to spend time building their own agents.

They want something that just works. And this is the kind of experience that we're aiming for. Now, travel is also a big thing.

And again, the whole concept of AI agents that can do things for you in the real world versus just drafting documents and things like that is complicated. But we are committed to build an agent that really takes things off your plate and really solves admin problems for real executives, and travel is part of it, so there's no other option. We're working on it and working on it really, really hard.

But actually, the value proposition in travel is working in a really cool way because once you build a proactive agent, it sees in your calendar or in your inbox that you bought a ticket to Black Hat Conference. It doesn't wait. " It works proactively, saying, "Listen, you got tickets to Black Hat.

You always go to the same hotel. I already checked the rates. " And that's it.

You don't need to do anything else. So it works really, really nicely, and it's one of those features that we see great adoption among our user base. So we're very happy.

Excellent. I want to talk about other things that you have on the drawing board, but before we do that, Nir, two things I wanted to bring up and get your answer here on video. The two objections that I hear when people ask me about Crystal, my EA, and I tell them it's Catch AI, I tell them how much it costs, I tell them everything.

And there have been two pushbacks I get from people, and I want to hear how you deal with them. One is, well, why can't I just build this using Claude or ChatGPT or something, on Perplexity Compute or something like that? And I suppose you could, as you said, if you want to take the time of really specializing that agent, and using MCPs, and to give it the access it needs and all that, though some of them have, let's say, Google access.

Mm-hmm. " Yeah, no, I think it's a very common question that we get a lot. And I think the most important part to understand is that Claude is not, or ChatGPT or any of these tools, they're not built specifically for this task.

Right? They're general assistants, they do a whole bunch of things. So somehow they're not really, really good in execution of administrative tasks.

Now, obviously, we use all the best, the frontier models. Under the hood, we use all of that. But we realized that for many tasks, they just don't work.

They just fail. They fail miserably. And so we started training our own models and build our own models that are dedicated and specific for administrative tasks.

Now, here's the thing with how people work with AI. So usually when you work with AI, if you're working with Claude or ChatGPT or whatever, it's very interactive. Right?

" And then you get something in return, and you give feedback to the tool you're using. "No, I want you to speak more about these companies, and make it shorter, make it less formal," whatever, right? You can kind of steer it the way you want to do it, which still delivers tremendous amount of value.

But it's the kind of work that is interactive between you and your AI software. In our space, you really don't want to work with the agent about administrative tasks. So it's not like you're creating an invite and you're working together and making sure that the time and title and email address is correct.

You want to delegate it, and you want it to be 100% accurate. And for that reason, and when you think about it, tasks that you and I probably think are very, very simple. " Actually, there's a lot of small decisions that the agent needs to take, right?

Who's Alan? How do I reach out to him? Where is he based?

How long is the meeting? Is it in his office or my office? What happens if he doesn't reply?

Do I follow up? Do I escalate? Do I call him?

Does he have an executive assistant that I need to reach out to? There are many, many different questions, and if- Sure ... your AI agent will fail in just one of them, then the whole situation will be out of control.

It stops. Why did she schedule a meeting in my office when we were in Florida, right? Right.

So you actually need to be an expert in this domain to deliver experience that is really amazing. And putting aside all the maintenance and security issues and everything that is the infrastructure layer that supposedly maybe people can fix or not, but putting that aside. Even from a pure experience perspective, it just wouldn't work.

So a lot of our users, actually, the vast majority of our users, use us side by side Claude or ChatGPT, and they use them to analyze data and write code and write content and write marketing stuff and everything, but use us to run their admin. And this has gone on to a situation where I think it's going to continue this way because they're just not focusing on that, so they're the experienced much. I agree.

I agree. I want to come back to that in a moment, but I want to hit you with the second objection I get from people. Let's go.

And that is, are you worried from a security point of view? What happens if it wipes out your... If you gave it Google Access, it could wipe out your Google Drive.

It could be hacked, security, governance, control. Yeah. You know.

I think that's a great argument not to build something yourself without the security mindset in place. But we're selling only to organizations. So from a compliance perspective, we're SOC 2 compliant, and we build our entire infrastructure with enterprise buyers in mind.

So all the data is in AWS, backed, encrypted, with key rotation, everything. So, we're trying to really build high bar of security. Our employees are former engineers in the best cybersecurity companies of Israel.

And- Right ... so we have a lot of emphasis on security, both from an infrastructure layer, but also from the application layer. We don't want that someone that is interacting with Crystal will be able to get out of Crystal your entire schedule or personal emails that she holds.

So we're putting a lot of emphasis on the architecture, and we're monitoring things 24/7 and making sure that we're handling your data in a secured way to make sure that there are no breaches and nothing of that nature. Trust is really core to our business. So if we will fail on that, the business will fail.

No one will trust us. So we're putting a lot of emphasis on it. And it's also, it goes back to the same idea of why not building it yourself.

The architecture and compliance and governance around how to give AI agents the ability to speak with the outside world is actually really, really important and really complex. And I wouldn't trust agents that you vibe coded yourself to do that. I think that's a big, not only from a business perspective it's not a good use of your time, but also from a security and data privacy perspective, actually, it's a lot more dangerous than- No ...

choosing an approved, experienced, focused, verticalized player. Excellent. Nir, I got two other questions, not objection related, but just I wanted your opinion.

Sure. So when I came into security 25 years ago, the rage was what we call purpose-built appliances. Right?

Mm-hmm. Maybe there were ASICs in the silicon or they weren't just off the shelf. They were purpose-built for tasks.

Do you think that's the future of agentic AI? That, hey, you could use the general models for a lot of stuff, but eventually we're going to migrate to where you're going to have specialist agents that are really specialized for particular tasks, for particular areas. Absolutely.

And this is really what I think around AI agents. I think the idea of, there's an interesting distinction between AppSec and website builders, where a lot of people use Lovable or Base44 or many of the other platforms out there to create their own websites. And somehow a lot of people think, oh yeah, maybe by the same logic, I can create my own agent.

And what we're seeing in the world is while websites are really, really taking off, and Lovable is a huge company with a lot of clients, and so is Base44, the industry of build your own agent for work environment is actually not working so well. And there are a bunch of reasons for that, but I think the main one is that with websites, kind of what you see is what you get, right? We all know websites for 20 plus years, and the UI is very straightforward.

With agents, there's so many things that go in under the hood that it's actually really, really hard to build. Give them the tools, give them the approval mechanisms, the security, the older logic behind it is actually kind of complicated. So for sure, not only in admin space, I think also in marketing, also in security and in many other spaces, I actually think that specialized agents are going to dominate the world and not build your own or general agents because, as we're getting more sophisticated and the verticalized agents just have a Gap in experience and performance, which will only grow with time.

Because we need to still remember we're in the early days of agents, right? Agents are here for a year or two years, nothing more. And as time goes by, the vertical players are getting really forward on a specific path, and the gap is only getting wider.

So I actually really agree with you. I think verticalized agents will probably dominate the world. Yeah.

That's just the way I see it. Yep. And they'll have the general underneath it, right?

You could act-- You could- For sure ... access the- Yeah, for everyday tasks, for things that don't require specialized- No, but even the vertical agent, if their RAG, if their training data doesn't contain an answer, they could always go back, right? Yeah, sure.

To our underlying- Absolutely ... data structure. Absolutely.

All right. I got one more issue I want to bring up. It's not an issue.

It's like, again, I want your opinion. I think it was Sam Altman was talking about persistent agents versus- Mm-hmm ... ephemeral, and agents that will just keep working on a task until you tell it stop, basically, right?

To me, that's kind of what you have now, right? My agent is persistent. Right.

" So what's the big deal that they're making that persistent agents are going to be so difficult or different from- Yeah ... let's say what Catch is now? Yeah, so I think the best way to look at it is to say persistent agent work really well if they have a defined scope of responsibility.

So if I'm only taking care of your admin stuff and your travel, or let's take a different domain, right? If we're only taking care of your marketing activity, then it makes a lot of sense to work with a persistent agent that have a limited scope of expertise. Where it gets complicated is that with general assistants is that we're using them for a bunch of things, right?

I'm a heavy user of a few engines, but in my personal life, I use ChatGPT a lot, right? And I use it to make bedtime stories for my kids and to plan travel, reserve vacations, and find restaurants, and figure out random things that I once would go to Wikipedia and stock market, and the context that I'm giving it about me is just completely random. I think, it would make the work of a mastermind to puzzle it all back and realize who's Nir out of all that.

And in this case, it's a lot easier that I have separate chats, right? I have a separate conversation for traveling and a separate conversation for bedtime stories and stories for my kids. Where we see a lot of advantage in persistent agents is that they're focused on a domain.

It's always the same address. " So in our domain, where it's really focused, it makes a ton of sense, right? Every request that we're getting, we're trying to map the admin space of the executive, and everything that we're speaking with the executive goes into this map because this is the way we make decisions, right?

I need to understand your context. I need to understand your relationship and understand your priorities, and it does make sense for me to look at all those things together when I'm acting on your behalf. It wouldn't be if I would be completely generalist, and then there was just no need for me to understand this whole world in such a way.

Get it. Hey, Nir, I promised you 15 minutes, and we're probably over 20. I apologize, but I've been waiting to talk to you for a while, so I had all these things to go over.

You know what we didn't tell people, though? What's the website so they can go check it out themselves? Absolutely.

ai. Okay. And just for your listeners, we're offering 10 days free trial.

In this free trial, what users are doing is that they're starting to work with the agent, and there's a learning curve. It takes about three or four days, and then the agent learns how to serve them. The most impressive and ambitious thing we're trying to do with Catch is that you don't need to learn how to use it.

There's no guide. There's no nothing. You need to tell it what you want, and it's our responsibility to make sure that the agent is useful and proactive and serve you in the best way possible.

So in this 10 days trial, and but again, it's usually for people that work in businesses, right? It's not a B2C product. I don't think it's extremely useful in a B2C, kind of like consumer-ish.

It's a product for managers. And as a manager, you don't need to do anything. Just start a free trial, give it a few days, give it a few tasks, and then it will kind of learn you and learns how to be useful in your working environment.

I love it. ai. Hey, Nir, I'll be in touch with you.

We'll have you back on here maybe in a couple of weeks or months, and we'll talk more. It'll be my pleasure. Thank you, Alan.

All right. Nir Sabato, co-founder, CEO, Catch AI. ai.

Check it out here on Techstrong TV. We'll be back in just a moment.