Latest / Growth After Headcount / 100x The Human, Skip The Headcount — Amos Bar-Joseph (Swan AI) #002
Transcript
- Haris Odobasic: Today we are talking with Amos, the co-founder of Swanai, They are building an AI that automates the go-to-market engineer. We are gonna talk how they try to make any human 10 to 100x more efficient, why he believes in personal brand, and why they're betting on analog and traveling the world to meet customers. Amos Bar Joseph: It's not my first rodeo of actually built Haris: Mm-hmm. Amos Bar Joseph: and scale two B2B startups based on the old playbook, you know, the unicorn playbook growth at all cost model where you raise a ton of money before you know who you're selling to, and you get to thirty, forty people before you get to your first million dollars in revenue. and then kind of sooner or later you realize that there's like So much bloat at the company and you need to maneuver and so hard to do Haris: Mm. Amos Bar Joseph: it with you know with all that head count. and in Swan, I felt like, you know, with this rise of agentic workflows and AI agents in the workforce, I felt like it's time to reinvent the playbook of how do you scale a business, you know, from zero to one, from one to ten. and so Swan is really kind of like this journey about discovering this new model. we call it the autonomous business. It's a business that is kind of like architected from the ground up towards scaling with intelligence, not with headcount. you know, human intelligence and artificial intelligence working together. it's about scaling ARR per employee rather than just, you know, valuation. so we we think that that's the you know the most important metric to measure in an AI native world that really can become a good proxy for how much leverage. you're creating for your talent with using AI. in 2025, we just just the three founders, we scaled, you know, from zero to over 200 customers, seven figures in ARR, without any kind of additional headcount, just us and a lot of AI agents helping us along the way. And Our goal eventually is to get to ten million dollars ARR per employee. That's what this company's about, and kind of show the world that there's a new model to scale. Haris: you have agents and I think you started with many agents and now most recently I think was like seven if you can talk me through this journey of iteration and where you are now and where you maybe hoped even to get Amos Bar Joseph: Yeah. So the initial intuition for the tech community, we know with agents was there's gonna be a lot of agents and we need a specialized agent for everything, right? So similar to kind of like how humans operate. You think you need an agent for writing emails and then you need an agent for outreach and maybe you need an AISDR and then you need an AI CMO or whatever, you know, and we try kind of like hacking different ways into it. And what we did is we were too focused on kind of the process of the agent rather than the environment that the agent actually lives in. And what happened over time is that at Swan, we just came down to just one agent for our entire go-to-market operations. and The difference is that that agent is basically a coding agent with a file system, with a knowledge base. You can you can think of it that it can just, you know, use that knowledge base and iterate Haris: Mm-hmm. Amos Bar Joseph: on it in runtime, et cetera. What does that allow you and give you? Is instead of humans working on every and agent and separately and trying to fix those and and improve them, you have an agent that has a knowledge base. And then, if you want to specialize in a task, the knowledge allows the agent to specialize in that task, but it's the same agent and it can access different areas of the knowledge base in runtime. And then it can even iterate on that knowledge base by itself. And it can kind of like improve itself over time and change the way that it works. And so what it gives you is flexibility and scalability to have just one agent in a file system unified all of the agentic workflows under one umbrella. And what eventually we realize is that we need the agents to specialize at just one thing, which is move Haris: Mm-hmm. Amos Bar Joseph: moving inside a digital environment. Okay. That is what the agents specialized in. That's why a coding agent that can communicate with his environment using code is actually a much powerful implementation of an agent rather than having the agent write in language trying to abstract away. All the digital environment into language as well, and just having to do one specific task like you know, like a very small cog in the wheel, right? So instead of building like that, what you give is you create general purpose agent that can do anything inside a digital environment because it's built natively to move around that digital environment freely without any obstacles, and you give it a knowledge base, a file system. That allows it to specialize and understand processes within the business and mimic those and kind of perform at the best way possible and improve itself over time. Haris: Yeah, I guess you cannot just use Claude, right? So you cannot just tell Claude to run your business. You need to have some solid fine tuning, I guess. And you mentioned a lot of knowledge. I guess data, clean data and accurate data becomes even more important than ever before. Amos Bar Joseph: Yeah, so I would s I would say two things. One, Data and integrity and accuracy is you know becoming more important than ever. Yes, that's true. But I would say that there's a misunderstanding about AI. People think that data equals context. Okay. But that's not true. Okay. And context for AI is to understand how does the business operate and what is good and what is bad. What types of best practices do we have? What bad practices do we have, etc.? And data doesn't tell that story, right? And if you kind of like throw an agent on your CRM, it can look at deals and can learn a lot of bad practices in those deals. There's, you know, no one fills up notes and explains, you know, the band about this deal because no one really updates the CRM. So the agent might think, that's what we do at the company. We don't update the band. about about a record because we don't care about it probably. Or yeah you know deals are jumping from you know step one to step five. that's how we usually do. We take it from step one to five. But what what happens like the rep didn't have time to actually move it to two, three, and four, right? Haris: Yeah, yeah, yeah. Amos Bar Joseph: So data doesn't equal context and that's like a very false assumption of a lot of people. And what you actually need to have for AI to work as Powerful as possible within your business environment is to teach it what's good and what wrong, how does the processes Haris: Mm-hmm. Amos Bar Joseph: look like, and without it, it it's worth nothing. And that process, we call it context engineering. And it's a skill that anyone, especially if we're talking about RevOps, that's the that will be the number one skill that RevOps will be measured by is how well could they articulate to the AI. What's the process looks like, where's the data lives, what's good, what's bad, etc. And the way that you do it with a coding agent, that's why having just one is very powerful, is you just have a knowledge base, you have docs that explain what's your ICP, what's your assignment logic, what's your funnel looks like, etc. And then you just need to make sure that the agent knows when to access these docs. And if you solve that, that's context engineering basically. Haris: That's quite interesting. And you mentioned a few times this one agent. Like how do you maintain that agent or does it maintain themselves? Amos Bar Joseph: Yeah. No, no. Agent agent can't maintain themselves. That's the job of humans. Haris: Hmm. Amos Bar Joseph: so you know, we're moving to a world where humans were doing a lot of execution, but then we have agents that can do the execution itself. Now it's our job, it's our you the humans are in charge of reviewing their work, giving them feedback and improving them. So, first of all, it's the job of the humans to do it, and that's actually context engineering. It's actually Haris: Mm-hmm. Amos Bar Joseph: reviewing it's so. The process of context engineering is is is looks like that. You need to specify well to the agent what to do. So what's the process here? What's good looks like, where the data lives, which systems to interact with, etc. That's their you know the specification. Then you have the reviewal process. When the agent does the work, you want to review it because you know it won't do a good job at the first time. Just like any, you know, any even the smartest person on planet earth, if they will come to your job and will try to do what you're just teaching them. They wouldn't nail all the use cases at the first hit. You'll need to keep on, you know, improving their context about the job. So specification, review, context engineering. That's like the process. That's how you work with agents. And that's that will always be the case, even 20 years from now. That's how we work with artificial intelligence, basically. And that's the job of humans in this new environment, and especially revops. And Haris: Mm-hmm. Amos Bar Joseph: basically a coding agent, the reason. Why it's powerful because it's the easiest way to work with it. You just specify the process, and because it can just move naturally within the digital environment, it can write code, it can communicate with APIs, it can build apps, it can write scripts, it can do all the stuff that you need it to do within the digital world. You don't even need to understand a lot of these technical details. You need to just be well at specifying the process and and what's good looks like, and then reviewing the output and than giving it feedback to improve it and imp impacting the context engineering process as you go. Haris: care. It's a continuous feedback loop in that regard. Never ending. Amos Bar Joseph: Never ending, never ending. It's that's what we do. That's what's left in in an AI native world is just you know to improve the work of AI agents and it will always be the work of humans. And we could automate some of it, etc. But it's always like our responsibility, it's our ownership to make sure that they do a good job. And I always love telling people that you know, if you have AI agents helping you generate pipeline and you missed, you know, your your pipeline goals in Q1, you You're not gonna fire your agents, right? It's not what you're gonna do, right? You're gonna fire the people that build those agents, right? Whether it's a vendor, whether it's an agency, or whether it's your RevOps person that didn't build these agents well enough, right? So it was all it will always be the ownership of a human, not an agent, right? Haris: nice. So what can the agents do like on three levels? What is like where you see you care? Happy autonomous, this is great. Where you still see some samey human interaction with autonomous and where do you maybe hope in the coming few years to also automate further? Amos Bar Joseph: Yeah, so I asked I get asked this question a lot and I think it's the wrong way to look at it actually. So Haris: Okay. I Amos Bar Joseph: Mm-hmm. Yeah. so you need to look at it two different levels. So one You should constantly ask yourself what should be fully automated, what should be semi-automated, and what should be human first. Okay. These are just three questions that you always ask yourself, and it's a matter of resource allocation. Okay. So if you have a ton of volume of inbound leads, okay, and You don't have enough manpower to handle that in a good way, then you might say, okay, let's take the lowest value leads and let's fully automate the interaction with them to try to get them on a you know an as a sales qualified lead. Then you take the medium ones and you do the you you allow the AI to do 90% of the work and then kind of like leave 10% for the human. And then you have like the highest value leads, and you want to say, Yeah, that's where we want our humans to be focused on. And because humans plus AI. Is always better than just AI. Okay. It's just it's just the the that's how intelligence works. If you have an agent that does the job and you have it then another agent that looks at the job, you can get to a better outcome, right? Same thing. Humans in AI will always be just AI, right? And so you have the highest value leads always go to humans plus AI with an emphasis on more creative and more strategic work for the humans. Medium. goes more to the AI, less to the human, but still collaboration, low is like fully automated. Haris: Mm-hmm. Amos Bar Joseph: That's like a mental model. You can think about everything basically. And it's all a matter of resource allocation. And you need to constantly iterate on it and ask yourself, like, I I want my humans to just be focused on the most strategic s things that they are most competent in. And I want the AI to take the most lowest hanging fruits, the low level shit that they you know humans don't want to do, basically. And that's how you create that resource allocation machine. That's One way to look at it. The second way, which is the most important one, that's how you connect them, is that especially in go to market, you know, go to market is always like a zero-sum game in the sense that you're fighting over the attention of the buyer, over like a finite budget most of the time. And it's kind of like there's a lot of competition and you just want to be different, you want to be unique, you want to be better than the other people, right? It's not like in software development where you just want to build and build and build and build and build. In GTM, you know, different is better than better. Okay. And What you need to understand is you need to find your unique value proposition, your unique DNA in the organization that makes you special and double down on it in terms of your humans. What does that mean, Harris? If you have a good cold calling DNA, then You don't need to give cold calling to the AI. You just want to make sure that your reps are spending their entire day on the phone and AI kind of like amplifies a 24-7 from all around, basically. Right. Then you have like LinkedIn and email automation. AI can take care of that, for example. And with Swan, I'm really good at storytelling and I'm really good at kind of like, you know. Playing the LinkedIn game. So I will never let AI automate my post creation and do that because then I will lose all of my edge, right? So what I'm what we're doing is we built the like an entire AI machine around my post creation process so that we can, you know, squeeze the juice out of it, push it even further, but we will never automate that, right? Because that would defeat the purpose. Then my post becomes like everybody else's post, right? Haris: Exactly. Amos Bar Joseph: So that rule will remain true for GTM, you know, forever. Why? Because you know, you just want to be better than your competition, right? And GTM. You just want to be outbeat them. You want to be different. And if you would give everything to the AI, you would just lose your alpha, you'll use your differentiation because everyone can use AI in an AI native word. It's not interesting, right? So you need to always think of it as a resource allocation machine. And the top of the top of the resource allocation goes to humans, and you should double down on what you are specifically good at. Haris: Hmm, no, definitely. Let's bring me to the next question. Adam is also very strong in personal brand marketing. So what do you see the role of personal brand marketing and making that probably to differentiate? Because yeah, everyone has the same tools. Everyone could build something similar Amos Bar Joseph: Yeah. Yeah. So I I I think that in a lot of in a lot of areas in GTM AI is not a revolution, it's an evolution. Okay. It's just a continuous amplification of a specific process that is already happening. Okay. To your regards, and there's like tons of different areas where it's happening, but to your regards, it's about you know the erosion of trust, okay, in GTM. And sales automation started that process. Sales automation started erod erosion the the the trust of buyers because you know suddenly every you know SMB in the world, no matter where you are, could start blasting like 100,000 emails at buyers and people and then we started going through this phase where you know buyers feel bombarded, everyone telling them stuff, and they feel like I can't believe anyone, etc. AI took that like 100x to the exponent, where now Everyone can just you know generate personalized spam at scale, and everyone can kind of like promise you that they can build anything, right? But you don't really know about the quality of that, and you don't know if that is really helpful. and you don't know if you should listen to that person or not, etc. And in this world, trust is becoming like the number one differentiator between companies, right? And then you should ask yourself, how can you build trust at scale? And in a way that is repeatable, right? And kind of like you can engineer that. And, you know, personal brands are the number one way to do that at scale. And what we're starting to see right now is that, you know, the last five years, it was just proof that, yeah, in B2B, personal brands are exploding. And you have like, you know, the CEO of Zoom info going on every like week, posting about like a new feature, stuff like that. you know, there are ways to do it. In a good way or in a bad way, but you you i Haris: I think so, yeah yeah. Amos Bar Joseph: yeah, exactly. But you see that marketing teams are understanding the fact that you know personal brands should be at the front. What is the next phase that already happening right now is taking your entire leadership and your entire company eventually to try to build their own personal brands. What we're doing at Swan right now is that every person on the team Should start building their personal brand. And we're actually turning them into a media operation machine. So we just hired Jeremy, who is our like our VP of autonomous GTM. Okay, who was like a VP sales and he's coming to manage agents. He's not gonna manage, you know, reps, right? But we're building his brand into a media operation. He's gonna have his own newsletter, his own newsletter. It's not the company's newsletter. Like I have a newsletter, I have a personal brand on LinkedIn, I go on podcasts. Jeremy has a personal brand on LinkedIn. Jeremy has his newsletter. Jeremy's gonna go on podcast. And when Jeremy goes on a podcast, he's not going as a VP sales of Swan. He's just going as Jeremy, right? Because people want to listen to Jeremy. So we're seeing kind of like again a continuation of that because of AI for everyone to build their own personal brands. What is also happening there, Harris, which I think is interesting, is to build trust, we're gonna see what I call the rise of the analog. Okay, so. Analog experiences are going to become much more prominent and much more powerful in GTM. That's why we are an AI native company with a small team and we're running an event motion where I'm flying around the world for events to meet people face to face and telling the the story of Swan because you know in an AI native world, nothing beats the personal connection. And if you can do you can create powerful analog experiences, then that's emote. Right. No one can do that at scale using AI. Right. So trust is the number one game that we're playing in GTM right now. Haris: Nice, nice. Sounds amazing. And I agree basically in everything you are saying. What is your plan for the future? Increasing your Claude bill or higher more? Amos Bar Joseph: Yeah, so I would say both. Okay. but eventually what we are realizing that today it is more efficient to focus on hiring the best talent in the world and then instead of hiring more, try to scale them to their one hundred X version of themselves. So right now we have more demand than Jeremy could handle. And a lot of people tell me, look, you're just wasting a lot of revenue right now. You can just hire more sellers and they could just hail, you know, handle more demand right than right now than Jeremy could. But what people don't understand is that every day that Jeremy works and we're building agentic workflows around Jeremy, Jeremy's capacity's growing. Okay. And it's not fixed. It's constantly growing and growing and growing and growing. And we're actually taking the most talented seller on the planet and kind of like replicating it. To have like a massive scale so he can support like 100x the capacity that he had when he started the job. And that's much more efficient than trying to hire and on board and educate and enable 10 other sellers in the same time, right? And so we are planning on doing it in a different way. We're scaling our headcount in a way that we take one talented person and then we scale them to the maximum that they could scale themselves, which is Right now we're not hitting that wall. There's so much more to harness. And when we're hit Haris: Mm-hmm. Amos Bar Joseph: when when we're gonna hit that wall, we're gonna get okay, let's let's take another one and then let's scale that person to their 100x. And so this the way that companies scale in an AI native world is not the same as before, because before you had like a a ceiling on the capacity of a person. So you brought a person, you knew the capacity, and then you want to bring another person. But right now, the capacity is just Know a question of how many resources do you invest in that person, right? How many intelligence resources, how many tokens are you investing in that person? And so you can deploy more tokens on a specific person to scale them rather than hiring another person. And so that's how the scaling model changes, in my point of view. And so I think that, yeah, eventually an autonomous business could grow into a 500-person team, it gets there. In a very different way, it would take like maybe ten years to get to five hundred people, and five hundred people would operate as five hundred thousand people like Google does today. Haris: Mm-hmm. That's very true. Basically that also means, so many companies had to take a lot of VC founding, to basically pound their headcount. So what do you see there in the future? Like if everyone does that, that would be less investment money needed. So the case of bootstrapping becomes more interesting. So where do you see the link VC versus bootstrapping? Amos Bar Joseph: Yeah. so I actually think that autonomous business that this model is is not only a new model for founders, it's also a new model for investors. And Haris: Mm-hmm. Amos Bar Joseph: I think that what is what it gives investors and what we've actually seen that in the last 15 years, VCs aren't doing well. you get the the one top, top, top, top, top zero point one percentile of VCs are making tons of monies, and then the ninety nine point nine other percentage of VCs not. really, you know, making it better, beating the you know, the S P 500. They're just not managing to do that. And what autonomous businesses promise is actually, you know, a business that requires much less capital. It's not like it doesn't require any capital, but it it requires significantly less capital. And it has much higher chances of succeeding because it's a more it's a more cash efficient business, right? It's not like you're trying To raise $200 million to build a business. And if you can't raise these 200 million, you're gonna fail. You might need 5 million, you might need 10, 15 million. That's it. And you can still build a massive company, right? It's not like we're trying to build a lifestyle business. I want to replace Hubspot and Clay, and I'm I'm fighting against Claude because I want to be the coding agent for GTM. And I think that's the central place of the stack in this new world. So I want to be the biggest player in GTM, I'm not trying to aim low. Haris: Love the ambition. Amos Bar Joseph: But but but what I'm offering to the investors is like, you know, just a more risk-verse, cash efficient way to get there, basically. And what we're going to see, Harris, as I see it, is just a lot more autonomous businesses in the world. And you're you're we're going to see in the next five years enterprises crashing down and being replaced by autonomous businesses. So you have You know, a companies like Salesforce or Hubspot that are going to be replaced by like, you know, ten different types of businesses that are going to be autonomous businesses. And these will be great opportunities for investors and founders. And we're just gonna have a lot more of that. Haris: That sounds like the future actually would also like. one final question looking at employment and compensation. where do you see the talent market is going to go and how the compensation might develop? Amos Bar Joseph: Yeah. yeah, that's true. Right now it seems like AI is also an amplification for inequality. what it will do is that if you're highly talented, your value to the company because of AI just grew like one hundred X, right? We're seeing for for example, Jeremy, you know, VP sales, right? In before Swan, like his value was capped at his capacity. 'Cause he's like how much he can sell, right? But with AI, we can just scale that and scale out and scale out. So his value is getting much higher. And on the contrast, if you are not that talented and you don't have a lot to contribute, you're actually fighting with AI over like your ability to contribute to the business. And we're already seeing, you know, juniors that don't have any experience and they're not that smart. Sorry to say it like that. You can just use cloud for that, right? You can say, like, why do I need someone to just go over these Excel's and and and just organize them if I can just have Flaw do that. So we are going to see this inequality even widening. and I I do think that the positive thing that is going to happen is that we're going to see a redistribution of work from big enterprises where a lot of people were just cogs in the wheels, of like cogs in the machine, and they had little impact, etc., to a much more distributed world where businesses are much smaller. Than they were. And there's a lot more businesses just in terms of the sum of the businesses, because they are now much smaller. And so people have more impact, have more mobility to actually find themselves in a place that they feel like is their zone of genius. What we call Swan, part of the autonomous business is trying to put employees in their zone of genius because that's how you scale them the best way. And so more people will find themselves at better jobs. With that gives them more happiness, basically. Haris: I think it's going to be a super exciting future, like the coming years. It's definitely a societal shift almost. Amos Bar Joseph: No, yeah, it is. It is. It's a first and foremost a societal shift, and then it's a technological shift. Haris: Mm-hmm. Exactly. Amos Bar Joseph: yeah. I see it as a software-enabled society societal shift. I think that AI in enterprise and B2B, that's what we're talking about. AI and B2B is a software revolution. It's not a labor revolution, it's a software revolution. And What we're seeing, you talked about headless, etc. It's about the rebuild, the rebuilding of the software stack from the ground up around coding agents, basically. And so coding agents are becoming the most fundamental unit in the technological stack, and everything just revolves around them. And the reason why anthropic is beating the shit out of open AI is because anthropic stumbled upon the fact that coding agents are the most powerful things in this AI revolution. Haris: Okay, wow. Exciting. Amos, thank you very much. This was amazing.