S15E2: Secure AI Agents for the Future of Work with NanoCo Co-Founder & CEO, Gavriel Cohen

All my agents I call Andy after Andy Bernard from The Office. He's friendly, he's competent, but he's not your best buddy, right? You're not going to fall in love with him. So that's, I think, the right dynamic to have with an agent. We're not building boyfriends or girlfriends. They get work done.

Welcome back to a new season of The Room Podcast. Hi, Mads.

Hi, Clauds. As you all know, we are opening the door to the room where it happens for today's technology founders. I mean, I can't believe this is Season 15. Crazy.

Probably one of our most exciting seasons telling stories up and down the stack.

After 6 years in the room, we're just delighted to be back together in San Francisco more and more, both being based here live from Vercel HQ. I think you have an update as well.

Yes, I'm also now at OpenAI.

Wow.

Which is fun, pulling us deeper into this revolution we're seeing globally.

Thank you everyone who's been listening to us since the beginning, 6 long years ago. And for those who are new here, please like, follow, subscribe, Wherever you listen to podcasts, or of course at theroompodcast.com.

On today's episode of The Room Podcast, we sit down with Gabrielle Cohen, the co-founder and CEO of NanoClaw, the open-source AI agent platform that exploded to over 250,000 downloads and 30,000 GitHub stars by solving the one thing that scared enterprises away from autonomous AI: security. With that, let's open the door to today's conversation. And now a message from one of our sponsors, Rippling. One thing we constantly hear on the room is that grid operators don't just need more data, they need a faster way to turn that data into decisions. As a founder, I know how painful it is to wait days or even weeks for a report while a real business problem is unfolding. That's what makes Rippling AI different. You can ask Rippling AI, is our sales team at risk of missing quota? And Rippling AI instantly pulls real-time data from Rippling and your CRM into presentation-ready dashboard data showing quota attainment, headcount trajectory, and monthly revenue to quota by region. But it doesn't stop at the insight. Rippling AI can act on what it sees, recommending and building an automated recruiting workflow when open roles and slow time to fill are putting quota at risk, all while following your existing permissions and approval processes. Question answered, action taken, problem solved. Ready to rule your business? Head to rippling.ai/theroom and sign up for exclusive access.

And now a word from our new sponsor, Gradial. Whether you're building a company, launching a product, or growing a brand, you know the is just the beginning. And marketing's bottleneck has never been creativity. It was always the work it takes to get an idea out the door. Gradiel handles the operations so that marketers can stay on craft. An agent will learn your process, work across the systems you already use, and delete point solutions you don't need. Designers will ship on-brand pages without counting prompts or tokens. Product marketers will launch across markets in days, not weeks. And digital experience teams test customer journeys Catch what's not working and fix it all before launch. Gradio learns that your brand, business, and process and brings that continuous context to every workflow. Every task gets done your way with brand, quality, and approvals built in. No more token maxing, slop, or runaway AI bills. More time to think bigger and create the impact people remember. Gradio, the system of work for marketing. Ashurst Perkins Coie is a leading full-service international law firm known for providing high-value strategic solutions for startups and investors. Lawyers at Ashurst Perkins Coie support the most innovative entrepreneurs and investors in fast-moving and high-growth sectors across the country while addressing the myriad of legal needs that arise when fundraising and scaling their businesses. The firm's tailored legal services are informed by understanding of each of the client's discrete needs. To learn more about Ashurst Perkins Coie, visit their website, ashurstperkinscoie.com. First of all, Gabriel, thank you so much for joining us today in The Room.

It's great to be here. Thanks for having me.

We're delighted to have you, and we love to start with the beginning for all of our guests. So to start us off, could you share a bit more about where you grew up and how that shaped your view of the world?

Yeah, I grew up in Brooklyn, New York, in a big family. Big family can mean different things. When I say big family, I mean I was one of 11 children. So I grew up in an ultra-Orthodox Jewish community and actually went to a religious school where we didn't really learn much in the way of math, science, English, anything like that. And when I graduated from religious high school at 18, I knew basic arithmetic. So then moved to Israel and started catching up on my studies at 22 years old. And spent about a year just catching up on high school level studies. Started university at 23, studied physics and computer science. And it was an experience that really gave me the confidence that I could, I wouldn't say do anything, but if I put my mind to something, I could accomplish a lot because I was able to catch up for all of those studies and then do well in university doing double major. not the easiest subjects. So I think that was a really important part of who I am and how I got to where I am today.

You're very casually talking about an intense double major, physics and computer science. So we're going to come back to talk about that. But of course, today you are a founder. So curious on that journey of both education and upskilling yourself, did you always think you were going to become a founder?

I think I've had it in the back of my mind that as I was going to university and working in different jobs, that these things, there was no linear path. I wasn't plotting it out, but I was choosing things that, that I thought might be good sets of experiences that would lead me towards being able to be a successful founder. Yeah.

Incredible. It's an interesting thing to learn how people view who they are, who they want to be in relation to becoming a founder. It's oftentimes people say that they always thought they were going to be one, or someone is like, absolutely not. I actually never thought this would be a thing, but then there was this problem and it just needed to be solved.

Something I remember in the decision to study physics is I remember reading an article, must have been in 2010 or 2011, the first time I saw Elon Musk. And I read this article and it was like billionaire physicist Elon Musk. And I was like, those are 2 strange words to put together. That was probably something that motivated me to study physics.

Absolutely. I'm curious about then the schooling journey that you went on, because I believe you went to Tel Aviv University. to study, and that's a long way from New York. So tell us a little bit about how you found yourself on the other side of the world starting that part of your life.

The community I grew up in, I made a decision that I was gonna go off on my own way. I grew up ultra-Orthodox and at about 15, 16 realized it works for some people. It wasn't the path for me. And it's a big change. It's a big decision to leave the community that I was growing up in, and I had to find my own spot, my own place. For me, that decision at 18, leaving the house, looking for different, really a different community, a different place to settle in, I made the decision to move out to Israel. And one of the nice things about Israel and some other countries is that you can study at one of the top universities in the world without taking out loans. And actually, I was able to do remedial studies all With scholarships and whatnot. That's the next 15, 20 years here. Just year by year, good beach, met people. How I ended up in, in Israel, I didn't necessarily think I was going to spend. Met my wife, got a job at wix.com, and I was working at Wix. One thing led to the other. That's how I ended up where I am now, almost 20 years later, and I'm still here.

That's incredible. You're back in your building. And so I'd love to dig a little bit into the early aha moment behind NanoClaw, and we'll dig a little bit into the ecosystem, what you're building, what the product is. But Gabriel, tell us what made you believe the world needed another personal AI agent framework, and maybe even stepping back a layer, how did you get interested in this domain?

Actually, after finishing at Wix.com, I went and worked with my brother, who is also my co-founder, he had a PR public relations agency and was working with a lot of Israeli tech startups, promoting them, helping them tell their story, really technical founders in cybersecurity, software development, machine learning space who were building amazing things, but isn't always that easy for them to explain to people what they're building in a way that's interesting and engaging. So I was working as a developer at Wix.com, stopped doing software development for about 4 years. And worked with 50, maybe 70 different startups and founders, helping them do their launches, launching from stealth, announcing their first funding. Worked with great companies, Run AI that was acquired by NVIDIA, Pinecone from their launch from stealth all the way through investment by Andreessen and whatnot. And it was very successful, but we had an understanding that public relations is probably one of the hardest things to scale and automate. There's a few companies, a few agencies that have done it. But it's, it really doesn't lend itself very well to, to automation, to building out workflows. And so about a year ago, we started to build together a new business, an AI native marketing agency. It was just before that became really popular, but we had a sense that AI native service businesses were going to be a thing. I was building that, showing up every day, opening my laptop and firing up Cloud Code, working with agents, not to do coding, but to do everything, to do research, content optimization, basically everything that we were doing for our customers. And it was working well. We were scaling, we were ramping up. By January, we were pretty close to a million in ARR and ramping really nicely. And then using agents for everything, but I was missing this piece of how do they connect together and how can I have people on the team who are not technical? Engage with them. And I tried OpenClaw and that was this big aha moment. I was like, okay, this is the missing piece. Now, first thing we did, put it in a messaging channel and WhatsApp group with myself and my co-founder and just said, okay, you're now managing our sales pipeline. I called it Andy. Andy, you're a sales manager. The sales data is in the file system that you're in. Good luck. And within a day it was like giving us briefings and updates and reminders and basically doing the work of the person. And I was like, this is different. I've been working with agents now for about a year, but this is something new. And I just saw this massive unlock, but from day one getting it set up, I also started to see huge security issues. And as I sort of dug into the code base, putting on my developer hat again and looking at what was going on there, I realized this is super valuable. This is something new, but it isn't production ready. And that's where I sat down and said, okay, everybody, you know, OpenClaw was this big unlock, showed what value you could get if you took agents, put security aside and some other concerns aside and just connected it to everything. And then there were other people who were doing similar kinds of thing, but having fun with it. And it was fun. It was a cool experiment, but I was like, I actually need this for work. I need to do real work with this. So I sat down to build what I think was the first agent framework that was trying to take those kinds of agents that are connected to your email, connected to your calendar, connected to real tools, and tackle the hard challenges that make it actually work for business.

That's fascinating. I think a lot of the shift towards agentic building and agents completing tasks on the user behalf has very much been framed from like a hobbyist developer perspective or consumer trying to get something done and playing around with this ecosystem. And so I think what Open Claude did a great job at was open everyone's eyes to what was possible. But even at work today, I can't use a lot of the bleeding edge tools because of privacy and security concerns, especially when you could have an agent that accesses data and could potentially share it or do goodness knows what in kind of a sort of risky situation. And so it's really cool to hear how safety and security was top of mind for NanoClaw. Why do you believe the market was getting security wrong with agents? And tell us a little bit about sort of the product approach in remediating that.

The market at that point, everybody was kind of sitting around and knowing that there are these big challenges with prompt injection, with security. We need human-in-the-loop approvals in order to make this work. But nobody was jumping in and taking it serious and saying, okay, how do we actually solve this problem by problem, piece by piece, build the right architectures, build the right tools? It seemed like everybody was waiting around for a big tech company or a big lab to put forward, here's the framework, here's how you do it, here's how you solve this. But it's really hard to do that as a big lab, as a big tech company, because they're also trying to solve the whole problem. And I think really Going back to my experiences at Wix and earlier parts of my career, I think one of the things, and also with public relations, working with tech companies that are doing really complicated things, one of the things that I felt was my superpower, I was technical and I was able to build things, but my superpower was simplifying things and going into a kickoff meeting with product managers and UX designers and all the stakeholders. And they put forward this whole plan and say, okay, this is going to take 6 months of development. And then the perspective that I tried to bring was maybe we can cut this and cut that and remove this and scope it down and find that 20% of the effort that brings 80% of the value. So being able to just scope things down, simplify them. And that's how I approached these security challenges of saying, I'm not going to solve the whole thing. And we still haven't solved the whole thing. But there's so much value to unlock that if we can solve a piece of it and just get it to the point where I can put this into a company, maybe it's not connected to everything yet, but it's connected to some things and start unlocking that value that's there. There's a tremendous amount to unlock. So the architecture and the security architecture started from that point of let's take an agent, not limit what it can, what its capabilities are. So not nerf it. And reduce its capabilities. Full-powered coding agent connected to all the tools, but limit what it can touch and what it can access. Limit the damage it can do by putting it in a box. Just put the agent in a box. It can go wild in that box. That box is its world. Can do anything that it wants to do in that box, but then have a clear boundary around what the agents can touch and then controlling What we put into that box and what we let it touch that's outside of the box.

Could I just ask a clarifying question for our listeners who might be less familiar with the concept of CLAWs or the concept of even just an agent at work? Could you maybe articulate this movement that's happening with that concept of a CLAW and how it might relate to a workflow that someone has in their daily white-collar life?

There have been really powerful AI agents out there. But only developers have been using them. About a year and a half ago, there was a company called Cursor, if you guys know, and then Claude released Claude Code. And these are extremely powerful. These are agents that can do anything that the best developer can do with full access to a computer. They can use any API, any service, any website they can interact with. Extremely powerful, but There are challenges with how do you run them in a way that is safe. And the way that they were run for developers is having developers sitting in front of their computer and babysitting them and approving each action that they do. The concept of Claw was saying, let's actually connect it to not just the 5 or 10 tools that most developers are connecting it to. Let's connect it to all the tools, hundreds, thousands of different tools. Just connect it to everything. Don't limit what it can do. So don't have a developer or anybody sitting there approving each action, but just say approve all, YOLO mode, can do whatever it wants. And then have it run in this sort of loop where in ChatGPT, let's say where you're starting a new thread all the time, starting a new conversation, but one endless conversation that just goes on and on. And then because it's a coding agent running in a computer, It can save files. So you say this computer is your computer, right? To the agent, this is your computer. It's your environment. It's yours to install new tools, download new libraries and new capabilities from the internet, and to store information, to add files and folders and create your own file system, organize it how you want, start building up your memories. When you start adding all those different capabilities together, ability to add new tools from the internet. So it's kind of self-evolving. It can grow its capabilities, extend itself. You add that endless conversation where it doesn't reset. So then it's learning, it's gaining memories. And you add connected to, and the key piece, connected to a messaging app. All right, so don't just have it where I have to go to the computer to a specific app, but put it in Slack or Discord or Telegram or WhatsApp, iMessage, whatever messaging app you actually use day to day. So you start to talk to it not as a coding agent, you start to talk to it as an assistant, the way you would talk to a person. It's conversational, just messages back and forth. You add all those capabilities together and you get something that's— and one more capability of being able to schedule something. So not just I send it a message, it responds, but I can send it a message and the agent can schedule for itself a task to do tomorrow or the day after. Or every day at a certain time. You add all that together, it becomes proactive. It starts to learn, it evolves, it extends itself. What you get, the sum total of all that, is something that is not just a chatbot you go to, you ask a question, it gives you an answer, gives you information, but more an assistant that you can give a task and say, do this for me, and it can go off and do work for you and figure out all these things on its own and add capabilities together. And do things for you and then come back to you and say, hey, I figured this new thing out. Hey, you might want to take a look at this. So it's proactive. And that's a totally different way of working with AI. That's a totally new kind of thing. OpenCLAW was the first one. And NanoCLAW, I think, is the first one that takes that and makes it secure enough, safe enough that you can actually bring it into a workplace.

It seems like we've shifted away from babysitting agents from constantly approving access to this tool, or yes, you can go search the web, to building an agent that's fully autonomous, but maybe needs a little bit more management structure. And so with NanoClaw, we're building the sandbox in which your virtual employee or virtual person that you're interacting with. can actually go crazy, but do so in a safe way. So with that, what surprised you the most about how people actually use NanoClaw or how teams actually use NanoClaw once it went live?

One of the most surprising things was about 3 months ago, NanoClaw had taken off, the open source project, tens of thousands of people using it, teams using it, Andrew Karpathy, Tweeted about it, and we had been thinking, "Do we build Nano Claw? Do we build this AI native agency?" And then Andrew Goprati tweeted about it, and it just went to another level. And decision was made for us. We notified our AI native marketing agency customers. Hey, we gotta—we're shutting this down. We gotta let you guys go. Went all in, and then the first weekend that I took some time off and went on vacation with my wife, we went to a nice hotel and. Saturday morning scrolling on Twitter, and suddenly I see this post, and it's somebody talking about their NanoClaw setup and how they use it and how they built it and the memory system. And I start reading this post and looking through it, and it turns out this was the Minister of Foreign Affairs of Singapore, and he is a major NanoClaw user personally. And he did this whole post publicly about how he's using it and how he's building it, and he says This is my second brain, my diplomat second brain, and I wouldn't dare switch it off. And I use it for everything, for speeches, for research, planning trips, helping prepare for policy sessions. At that point, it was before we had gone fully into this direction of starting to bring this to businesses. We were building it. There was a community. There were people taking it to businesses, and we still had these different ideas of which direction do we go with this. And then that made me realize this is actually highly valuable. There are still people, and there were people back then questioning online, is it all hype? Are people spending all their time just debugging their agents and building their agents and the whole craft is just building your agent, but do you actually get value from it in the end? And this was somebody who they're not trying to post For notoriety or for likes. They were going out and saying, you need to get on this. This is my edge. This is my advantage. And that made it really clear to me that we need to build, and this is the use case we need to build for. Not agent factories, not agents that automate processes for you, but your assistant, your main agent that you talk to, that's in the loop on everything, that helps you do your job. And do it effectively, that makes you 2, 3 times more effective. That's what we needed to build. And we needed to build it specifically for the government agencies or government offices, the large enterprises, the banks, the businesses that are highly regulated, and organizations that have to think about compliance and security and privacy. If we weren't going to build it for them, there was nobody who was coming to go and give them this. tool to unlock the value for them.

You clearly found a product market fit and had an incredible example. Honestly, so cool to see too online how people were just so willing to share how they were leveraging your tools and then have it be someone with such prestige is just an amazing feat. It makes sense why you went on to create a formalized enterprise behind the NanoClaw movement into NanoCo and what became the startup that you raised for. And just to put this in context for our listeners, NanoClaw was an open source movement that you all had built and created, which today has over 30,000 Git stars. For those who are less familiar, that is an incredible number in such a short amount of time and just in general. So huge congratulations on the success of the open source project. And tell us a bit about the decision to go on to raise that initial $12 million that you did this year with great investors like Slow Ventures, for example, and kind of who was the first person to say yes to investing in you and this company.

One thing we constantly hear on The Room is that grid operators don't just need more data, they need a faster way to turn that data into decisions. As a founder, I know how painful it is to wait days or even weeks for a report while a real business problem is unfolding. That's what makes Rippling AI different. You can ask Rippling AI, is our sales team at risk of missing quota? And Rippling AI instantly pulls real-time data from Rippling and your CRM into presentation-ready dashboard data showing quota attainment, headcount trajectory, and monthly revenue to quota by region. But it doesn't stop at the insight. Rippling AI can act on what it sees, recommending and building an automated recruiting workflow when open roles and slow time to fill are putting quota at risk, all while following your existing permissions and approval processes. Question answered, action taken, problem solved. Ready to rule your business? Head to rippling.ai/theroom and sign up for exclusive access.

And now a word from our new sponsor, Gradiol. Whether you're building a company, launching a product, or growing a brand, you know the idea is just the beginning. And marketing's bottleneck has never been creativity. It was always the work it takes to get an idea out the door. Gradial handles the operations so that marketers can stay on craft. An agent will learn your process, work across the systems you already use, and delete point solutions you don't need. Designers will ship on-brand pages without counting prompts or tokens. Product marketers will launch across markets in days, not weeks. And digital experience teams Test customer journeys, catch what's not working, and fix it all before launch. Gradio learns that your brand, business, and process and brings that continuous context to every workflow. Every task gets done your way with brand, quality, and approvals built in. No more token maxing, slop, or runaway AI bills. More time to think bigger and create the impact people remember. Gradio, the system of work for marketing. Ashurst Perkins Kiwi supports the most innovative entrepreneurs and investors in fast-moving and high-growth sectors, addressing their myriad of legal needs. But the firm doesn't just provide end-to-end legal and business counseling to its startup clients. It also facilitates introductions to key advisors and sources of capital. Ashurst Perkins Coie's interactive website, Startup Percolator, offers access to programs, resources, and rich dynamic content designed to assist entrepreneurs in their startup journey. To learn more, go to startuppercolator.com.

Doing public relations for startups. We worked with incredible companies and we got to know a lot of great investors. Valley Capital Partners, Mitchell and Steven, they invested in great companies based in Tel Aviv. And we did the launch for the initial kind of launch, announcing the funding, announcing their exit from stealth for some great companies that Valley Capital Partners invested in. So we got to know them before we started NanoClaw and NanoCo. We already had a relationship and we became a go-to agency that they were starting to recommend all their portfolio companies to. And then in February, they sent this email introducing us to another company and saying, hey, great company, can you guys do their launch? And my co-founder and brother Laser, he responds saying, I don't know if you guys saw, but Actually, we've got this thing going on, Nano Claw, that's us. And Steven responded instantly to that email thread saying, we want to lead your round, just immediately. And I think that level of conviction was really important for us in this space with how quickly it's moving. Things are moving so quick. There's so much uncertainty in this space. I think the only way an investor could have that level of conviction about a company came to knowing us and just believing that, yeah, there's all kinds of challenges. We're going to have to make all these kind of twists and turns and little pivots on the way, but we're going to figure it out. And actually meeting with Sam Lessin from Slow Ventures, also great conversation. He was in the airport, missed each other a whole bunch of times. Finally, he was in the airport at his gate getting on the plane and we were having this conversation over FaceTime. He had really good questions, really sharp questions about the space and about things And I had ideas then. In retrospect, some of them, those ideas were right. Some of them were wrong. But he said, yeah, okay, you know what? You got to get on the plane. Look, you've got all these questions, all these challenges, but you'll figure it out. You'll figure them out. I think that's for all of them. You know, Docker invested in us as a strategic investor of the company. Vercel invested in us, full disclosure, and lots of other great investors. And I think that's what it came down to for all of them. It wasn't that we came with a perfect pitch. We raised without a deck, right? And we raised within a week, got a term sheet, and we closed our funding round. It was about 6 weeks after I wrote the first line of code sitting on this couch over here. It was a collaborative process, I think, like all the best work we do these days. I launched it on Hacker News and people were like, We could see you're using AI. And I was like, look at the commits, right? If you look at, uh, on GitHub, you can see each piece of work that you do. And I said, every single thing that I pushed to this project, every single batch of code that I pushed, every single one says in collaboration with Claude Code. So I'm not hiding this. The project was built as an AI-native software project. I was deep in this mindset, went way down the rabbit hole building an AI-native agency, marketing agency. And when I sat down to write it, I said, this is going to be an AI- Native open source project. It's going to be any software. We're doing it in a different way. So everything we do in our business as well, someone shows up first day, I sit with them 20-30 minutes, try to give them some quick overview of what we're trying to do and what we're trying to build, and then say, okay, here's my agent. My agent will help you with the onboarding. Ask it anything you need. It knows everything that I've been through in the last 5 months and knows it better than me because I barely remember any of it. It's all a blur. But my agent's got all the call notes, it's got all the email threads, so just ask it anything you need. So yeah, everything that we do with NanoClaw and at NanoCo is agents with people. So it's, none of it is fully autonomous agents and none of it is fully sort of manual people.

At a luncheon recently, someone pushed me on what is the definition of AI native? I said that term so casually. I think we use that word quite flippantly and it actually took me aback because I realized I had to defend this thing that felt like I understood what it was, but in that moment I realized I didn't have a codified definition. And I think what you just described for your workflow is to me what it means to be AI native, that both by default you leveraged the tools at your disposal, of which many are agents themselves, and then also you're not ashamed or there's no hiding it. And that feels like the best way to be for your business.

Absolutely. I think with NanoCo and our mission, It's about not just using AI as much as possible, token maxing, whatever you want to call it. And it's not just using AI for everything, it's using AI effectively and responsibly and doing it the right way. So we strive to be an example of that, of not hiding it. We're using AI throughout everything we do, but trying to do it the right way, in a thoughtful way, where it's guided by humans, by the people on the team, and humans are making the key decisions. It changes how we do everything in the company. So when we're sitting down to work on a development task, or people on the team are sitting down to develop a feature, the first thing they do is do a big research session with some agents, right? Brainstorming and attacking it from all kinds of angles and debating and discussing. And then they share with other people on the team, here's the research that I did, usually a nice HTML, nice visual page that their agents help them create. But here's what I researched. Here's what I discovered in this research process. And not 12 pages, but a 1-page thing that consolidates and condenses the most important pieces of information that you need to know from this rabbit hole that I went down. With all of my agents, and we came back and can present to you distilled information about what are the key decisions here that you need to understand and what are the key open questions. And let's discuss it together, agree. And then once we've agreed on those, everything is laid out for us in terms of what direction we're going to take with this project or how we're going to build this. And then at the later stage, the later reviews become a lot less important once there's Lots of things to review, we know that we're aligned on the key decisions, on the key direction.

I love 2 concepts you brought up there. One is like token maxing is not always the right thing. Traditional software has always hinged on, are we increasing daily active users? Are we increasing sessions? Are we increasing duration? But what I'm hearing is with agents, it's really around, are we getting closer to the task we sought out to achieve? Are we getting closer to an outcome? Which is really interesting. And I also think to your point, now that it's easier than ever to build and we have agents that can execute, really where the strategic sort of layer is on what are we building, that being the critical juncture where discretion and like human sort of judgment needs to be, is more up in the funnel of like, what are we building? And then let's just go crazy and build it. Which is really cool.

And I would say restraint, right? Like the agents, you let them go, they'll build all kinds of things. They'll write all kinds of things. They'll go in all kinds of crazy directions. And I think we need to rein them in a little bit and exercise that restraint saying, let's scope this down. We don't need all those extra toggles and buttons and charts and modes. Like there's information overload, context overload that we're all suffering from because there's just Things are moving so quickly, being bombarded with so much information. And I think it's the responsibility of the builders, product people and designers and developers and everybody else to give our users some clarity and some simplicity and help make sense of all the noise and present to them, not just throw everything at the screen and see what sticks, but come also with an opinion. Say we went out there, we did our research and we came back. This is the right way to do it. Earn the trust of our users where they trust us that we're bringing a good opinion and a good perspective. They can accept that. They can, they don't need to try all the things because they trust us that we're bringing a good way of doing things.

And this is why in many ways, I know there's been a big debate around, are product managers going to be dead in the future? Maybe not, right? Like we still probably need that taste layer and that like editor-in-chief. Prior to getting into building.

Talk your book, Claudia.

Yeah, don't fire me. We're still useful people, I swear.

Well, look, the way that we're building things at NanoCole and our whole approach is the people, the humans have to be at the center. Agents are really good at amplifying capability of people, but in the end of the day, it's about people in the organization. They have knowledge, they have experience, they have good judgment and good taste. And as agents that we're providing our assistance that make people more effective, but we're not in the business of replacing anybody with agents. I don't recommend it. I don't think that's leading businesses, individuals, society at large anywhere good if we're trying to replace people with agents. I think in the end of the day, The people need to become managers of agents, need to become really effective managers. People who are really effective managers will be able to manage. You start with one, your first agent, your main agent that you go to for everything that's plugged into your email, your calendar, knows everything that you're doing, and that can help you make you more effective, 2, 3, 4, 5 times more effective. People who are competent and capable of that will be able to manage a second, a third, 5, 10, and some people will start to manage a whole team and a whole department of agents. But in the end of the day, people need to take responsibility, be accountable. Ultimately, I don't want to have a quarterly review with an agent, and you can't put an agent on an improvement plan. And especially when you look at larger businesses, large corporations, Fortune 500 companies, those industries, the culture, they're not going to just be Replacing people with agents. At the end of the day, agents, the way we see it, are amplifiers of people. It's not a replacement for people.

You've spoken a bit about the NanoClaw technology. We've spoken a bit about where the puck's going in terms of how the future of work looks with agents and how people are interacting with agents. Let's spend a little bit of time on company building. We chatted a bit about fundraising and how that went for you, but you also reportedly received significant acquisition interest. I think especially today, there's been many incidences of companies existing for 6 months, a year, and having life-changing outcomes via M&A. How did you decide to keep building independently?

I had a friend who very early on, he's a friend before NanoClaw. And as soon as I launched NanoClaw, I think it was the first day that it launched, he sent me this message going, holy shit, I can't believe that's you. And, and he became very involved. He became an early maintainer of the project. His name is Tom Gernot, played a really important role in contributing and helping manage community in the early days. We were getting on kind of calls every day to discuss the Discord and the open source and the imposter sites and all the other things we were dealing with. Every day getting on these calls, he'd be like, I launched a new product on NanoClaw. So he was building these kind of products using NanoClaw, and one was a WhatsApp moderation thing, and another one was a shopping agent, and every day another product. And I was like, okay, something crazy is going on here. Things that used to be a product that you would pay maybe $20 or $30 a month for have now just become a prompt that you take this agent framework, you give it a bit of a prompt, and it can go and create this thing for you. The potential was huge. It was massive. I wanted to see this thing that I had created and it was taking on this shape and how this whole community building around it, I wanted to see it realize its potential. So more so than me saying, I think I can do better, I think it was like this project and this idea that had come into the world saying it, I saw how big it could be and I said, I don't want to have a company acquire it and then just turn it into another vertical agent or another little piece of their platform. I wanted this thing to live up to its full potential. So, that for me was that decision. And I think still, we're not in it for an early acquisition. This is a huge, huge opportunity. There's a wave like we've never seen before, and we want to just keep riding it.

So much of what you just described reminds me of classical change management motions that I've experienced inside of larger enterprises. And it's so interesting to hear You articulate this concept of an agent or a claw working on your behalf inside of an enterprise to be effectively a second or third employee on your team. Really interesting and very clear how that picture is painted for someone who might be starting their own agent productivity user journey. You are truly building on the bleeding edge of what is possible with agentic infrastructure, and I just want to highlight how you've thought about architecting NanoClaw and NanoCo and the product suite itself. Specifically with some of the features you mentioned earlier about scheduling or cron jobs and how you've actually been pushing these primitives to the edge. What have been some of the architecture decisions you've made? And then honestly, like, where are you finding it's not even ready yet for primetime?

We're out in the front on human-in-the-loop approvals. If you look at Claude Tag, which, you know, integrates with Slack, they don't have human approvals yet. It's been, I think, 3 or 4 months since we released it and I don't know if there's another business-ready agent that has that. And that's a key thing. Like, you can't put agents into a company if you don't have the ability for the agents to stop at certain points and get approval from a person. We've been out on the front on having— you mentioned scheduling and cron jobs. The person who's using the agent doesn't write a cron job, right? They tell the agent, About some work, something that they need done. For example, my wife, I put her in a group with her agent, Andy. All my agents I call Andy after Andy Bernard from The Office. He's, yeah, he's friendly, he's competent, but he's not your best buddy, right? You're not going to fall in love with him. So that's, I think, the right dynamic to have with an agent. We're not building boyfriends or girlfriends. They get work done. I put her in a group with Andy, with her agent. We're expecting a baby in A few weeks, she said, hey, Andy, can you let me know when this baby carriage goes on sale? And he's like, sure, I'm going to let you know. And she's like, came to me and said, hey, Andy's going to let me know. He's monitoring this thing. I was like, is he really though? Or did he just hallucinate that? So we went and said, hey, Andy, like, how are you going to let Ilya know when this baby carriage went on sale? And he said, yeah, I've got this list that I put together of websites that sell this baby carriage and I've got the product pages. And I'm tracking, I have a file where I've listed the prices of them today. And I've set this recurring job every day, 9:00 AM. I visit using the browser, Vercel Agent Browser. I visit each of these websites every day and check the price and record it and compare it to the previous price. And as soon as one of them goes on sale, I'm going to send you a message in WhatsApp to let you know. And then it did a few days later, I let her know. She didn't schedule a cron job. She told an agent about a problem she had or something she wanted done. And the agent is scheduling cron jobs. The agent's using tools, installing tools, writing bits of code and scripts. And then one of the things we did is if you tell your agent, instead of update me when this product goes on sale, you tell it, notify me when my company is mentioned on social media, right? Instead of scheduling it once a day, It might schedule it every 10 minutes, and that's a lot of tokens, and you'll be on your leaderboard in your company for token maxing. But if you're on a cloud subscription, they may actually cut you off because you're just pinging them too often. The agents just keeps on doing the same thing again and again, and it just uses up a lot of your subscription usage, a lot of your token usage if you're on API. So one of the things that we did was implementing This capability where the agent, besides writing for itself this job that it does every X amount of time and setting for itself the prompt of what it should do when that job fires, it can also write a bit of code that runs before the job fires. And first the code runs, no agent involved. The code can check something, do a search and get a result and check if a certain word shows up there. And then only when that script identifies that something has changed or something needs to be done, that's when the agent kind of wakes up and kicks in and starts to do its work. So that's something that we implemented, I think it was late February, and then about a month later it showed up in Anthropic and Claude Code as a monitor. So we've been ahead on a lot of things. I think one of the things that I wouldn't say it's not quite ready yet, we've got it working, but it's only just starting to work, is agent-to-agent, having agents talking to each other. And a key part of that is the human-in-the-loop approvals. If my agent is talking to someone, I, at certain points, want to know and want to approve what it's going to say, its answer. And then some other capabilities that we've added recently to allow the agents to find who the agents can talk to and when. And that is just at the edge of really starting to become useful right now.

Thank you for letting us peek in the future a little bit. And this is definitely a reminder that I need agents to go make sure that items I'm tracking are letting me know when they're on sale. Um, so thank you for sharing that. And also congrats on the upcoming baby and new member of your family. Exciting things happening in agent world and exciting things happening in your personal life. So with that, as we come up on time for today's recording, we'd love to ask you our hero question of the podcast. We ask this question of all of our guests, and that is, who's a woman that has had a profound impact on your life and your career?

I guess it might be obvious considering, you know, anecdote I just shared, but, uh, definitely, uh, Yuli, my wife. When we just started dating, so this is 10 years ago, we were both students at Tel Aviv University, and we went together on early date, one of these places that do like the VR headsets and you can do a game with it, and spontaneously went in and it was this zombie fighting game where you have this gun and zombies coming and attacking you. You have to shoot them as they're attacking. I don't know if you Either you've done a game like that or used much of VR headsets, but once you put it on, you're in it. Like, that's real. That's your world. That's your life. You are really in this place and zombies are about to get you. And, you know, I did it and it was scary, but I was like, all right. And then she went and did it and she was terrified. Like, she was screaming, but she was fighting them. Like, she didn't pull it off and walk away from it. She was fighting these zombies and as terrified as she was, she was shooting them. And that approach, which I've seen again and again when things get really tough, just leaning in. And especially when they're really tough, like the toughest situations, that's when she really shows up. When I saw that, I was like, okay, she could deal with anything. Knowing that I can rely on her when things are tough, that gives me confidence. So thank you, Julia.

Definitely true qualities in a life partner. And thank you so much for sharing a little bit about the founding journey behind NanoClaw, NanoCo, and All of the things we should be thinking about as we operate and get things done in sort of 2026 in the future. Gabriel, it was a pleasure having you on today. Thank you so much for joining us.

Thank you both.

Thank you so much for joining us at The Room Podcast. If you want more from The Room every week, subscribe to our newsletter at theroompodcast.com/newsletter. We'll be back next week. We'll be back next week with a new episode and inspirational guest. Tuesday, 10:00 AM Eastern, 7:00 AM Pacific. See you in the room.

One thing we constantly hear on The Room is that grid operators don't just need more data, they need a faster way to turn that data into decisions. As a founder, I know how painful it is to wait days or even weeks for a report while a real business problem is unfolding. That's what makes Rippling AI different. You can ask Rippling AI, is our sales team at risk of missing quota. And Rippling AI instantly pulls real-time data from Rippling and your CRM into presentation-ready dashboard data showing quota attainment, headcount trajectory, and monthly revenue to quota by region. But it doesn't stop at the insight. Rippling AI can act on what it sees, recommending and building an automated recruiting workflow when open roles and slow time to fill are putting quota at risk, all while following your existing permissions and approval processes. Question answered, action taken, problem solved. Ready to rule your business? Head to rippling.ai/theroom and sign up for exclusive access.

And now a word from our new sponsor, Gradiel. Whether you're building a company, launching a product, or growing a brand, you know the idea is just the beginning. And marketing's bottleneck has never been creativity. It was always the work it takes to get an idea out the door. Gradiel handles the operations so that marketers can stay on craft. An agent will learn your process, work across the systems you already use, and delete point solutions you don't need. Designers will ship on-brand pages without counting prompts or tokens. Product marketers will launch across markets in days, not weeks. And digital experience teams test customer journeys, catch what's not working, and fix it all before launch. Gradio learns about your brand, business, and process and brings that continuous context to every workflow. Every task gets done your way with brand, quality, and approvals built in. No more token maxing, slop, or runaway AI bills. More time to think bigger and create the impact people remember. Gradiol, the system of work for marketing. Ashurst Perkins Kewey is a leading full-service international law firm known for providing high-value strategic solutions for startups and investors. Lawyers at Ashurst Perkins Kewey support the most innovative entrepreneurs and investors in fast-moving and high-growth sectors across the country while addressing the myriad of legal needs that arise when fundraising and scaling their businesses. The firm's tailored legal services are informed by understanding of each of the client's discrete needs. To learn more about Ashurst Perkins Coie, visit their website, ashurstperkinscoie.com.