The Future Is Multi-Agent

Julia Nimchinski:

Welcome to AI Practice Sessions. Excited you’re here with us for a high-impact day, where we are introducing the cutting-edge GTM architecture called Agents Forms. What is an agent swarm?

Well, we’re officially past the old funnel model, and what we’re seeing today is a new kind of architecture where, one. Multimodal AI agents are connecting sales, marketing, and customer success in real time. And throughout today’s sessions, we’ll be digging deep into multiple use cases, results, and early innings of ROI.

And we’d like to thank all of our sponsors, speakers, and partners for making this event possible.

Now, we’ve got Amos Par Joseph, CEO of Swan AI, our community favorite.

Welcome to the show, Amos, how are you doing?

Amos Bar Joseph:

Thank you very much, Julie. I didn’t know that I was the community favorite. I’m usually the devil’s advocate, the contrarian, so I’m not very favored at all, but I appreciate the vote of confidence.

Julia Nimchinski:

It’s a very special community. So, let’s get this started, Amos.

What does Agent Swarm mean to you?

Amos Bar Joseph:

Wow. That is a great question. I think, maybe one of the main reasons why we’re here today.

I think… before I answer that question, I think it’s important to understand that we’re kind of, like, at the height of the hype cycle right now, and, like, you know, 99% of SaaS vendors all of a sudden turn into agentic solutions, right?

And I think this session is super important today, because buyers… and executives should know how to tell the difference between a SaaS in disguise of just wearing the agent hat and a truly agentic system that could transform your business.

And what I’m going to talk about today is I’m going to answer that question, what is, you know, multi-agent system, and the difference between a SaaS, and then we’re going to see an example of it.

That’s going to be also interesting. And… I think maybe to answer your question, Julia, the main difference that I see is between what we call AI automation and multi-agent architecture, and I think just By telling the difference between these two, it will be helpful for us to understand what does it really mean to build and to run a multi-agent architecture.

So, the main difference, first of all, is… who controls the decision-making process within the software, okay?

That’s what we’re actually talking about here when we’re talking about multi-agent systems. In regular SaaS, all the decision-making process, all the features are, you know, determined in the code in advance. Someone wrote it and says, when someone clicks on that, this is what will happen, and all of the time will happen.

And that goes with AI automation, and SaaS solutions that are just adding just AI on top of their solution are not really a gente solution, because the decision-making process has been determined in advance.

If you’re kind of, quote-unquote, building an AI SDR agent, it’s not really an agent if every time it does the same thing and it cannot really adapt.

If the decision-making process is Find leads that finds the ICP.

research them. email them and follow up, and that’s what happens 100% of the time, because it is written in the code. This is not an agentic system.

This is just regular software with some API calls to a large language model. So, it’s just automation.

But an agentic system is actually comprised out of multi-agent… that multi-agent system that can think and decide together in real time.

And so the decision-making process is actually lying within the large language model itself.

It’s not written in the code. And the result of that is that you can truly work with a system that adapts, that learns, that can improvise, and can work with you like a true co-worker.

And… The cool thing is that that might seem technical, but the moment that you work with a truly agentic system, you just know it.

How? It feels like a co-worker. You can ask it to do something that you thought of.

at that specific moment of time, and the system can accommodate that request. When you’re working with an AI automation, with a regular SaaS solution, what you’re all the time interacting with is this phase where you say, can I do that? It cannot do that.

Maybe it can just… I’d love that output if it was just, you know, in Slack and not here. Sorry, I cannot do that. You’re emphasizing that data point, but actually, I want to focus on that specific sorry, that’s how we just present the notifications, right?

These are SaaS solutions, and when you work with a multi-agent architecture, it just feels like a co-worker that you can work with, that you can train, that you can tell it, how do you want to work with, and it adapts in real time. So, I hope that that was a very long answer to your question, Julian, but I hope it settles kind of like the stage for what we’re going to talk about for the rest of the day.

Julia Nimchinski:

Definitely. Before we transition to that, Amos, I would like to actually focus straight off the bat on the ROI. Because it’s been a, you know, an intense year of experimentation, deployments, and lots of futurism when it comes to AI, agent-tech AI, and lots of marketing, obviously.

And so, what are you seeing?

internally, externally with your customers, what are the, I don’t know, early innings in terms of return on investment, if any, or some tangible results?

Amos Bar Joseph:

Yeah. So, a misfortunate reality is that over 9% of AI implementations fail.

Right, and I think that buyers and executives should be aware of that, that they’re entering the casino here, and there’s a big chance that they’re gonna lose, but there is some tactic to deceive the casino, if you’d like, right? To win the… to win the house. And… I think that the biggest mistake that executives are doing these days is that they’re optimizing for perfection, not for adaptation.

What does that mean?

It means that they think that they can imagine that future state where AI is implemented in the business, and that’s how the process looks like, and everything just works like a Swiss clock. The reality is that this is a new technology, and both as, you know, the executive planning the new process, and the end user using this new technology, we don’t really know what is that optimal state. And so, instead of building a project, together with a vendor, or if it’s, like, building in- house, instead of optimizing for that perfection, for that optimal state, what you should build is a system that can adapt and you can iterate on in real time.

And what you should optimize for is feedback loops, basically.

So instead of thinking, how can I build this AI machine that does this incredible thing, how can we build an AI machine that can learn from the mistakes that we’re doing over time?

How can we make sure that we can input feedback into it, and it can adapt to these mistakes, and uncover them, and erase them, and improve itself, right? This sounded like you know, science fiction 10 years ago, but today, to build an AI system that could actually understands the mistakes, understand the intricacies, and the different nuances in this process over time, it’s not that complicated. Sometimes it’s just a change in the prompt, sometimes it’s just a change in the decision-making process, etc.

But what you need to ensure is that you have a system in place With the humans that can operate that system.

that can grow with your understanding of how the project unfolds. And when you actually follow that rule, when you build a system that is optimized for adaptation, not for perfection, then you see amazing ROI.

And our customers, for example, that are using Swan, Swan is designed, actually, for adaptation, not for perfection.

It has this core concept of feedback loops, where it’s It’s so easy to just tell the AI, look, next time that you’re performing this process, I want it to be done differently, right? And it just takes that message, incorporate it into that process, evolves itself.

And our customers can see significant enhancement in pipeline generation efforts, and in closing deals, and preparing to meetings.

And for every customer, the ROI is different.

And I think How do you measure ROI is also an interesting question, Julia.

If you think it’s interesting, I’d love to answer it, but if you want to take it somewhere else, happy to jump on that.

Julia Nimchinski:

Let’s definitely dive into that.

Yeah, I mean, obviously, Manny Medina is a huge community advocate as well, and our recurring speaker, so would love to hear more, maybe, you know, your collaborations, and how do you measure ROI internally and externally, and what are the metrics?

Amos Bar Joseph:

Yeah, so many Medina is the community favorite, that’s true. That, I can believe. And… I think what happened… with AI is that people forgot everything they know about technology, okay?

They think that because it’s… It’s called artificial intelligence, and it feels like a human, which is, you know, confusing. It means that we need to forget all the, you know, the tech rules that we had in our pocket, basically. And I think ROI is actually simple.

And I think how we treat ROI at S1 internally and for our customers is how people should look at AI implementations and how to measure ROI by large.

And we have a ground rule that would say that AI implementations always revolve around a specific person, or a specific function.

even if you’re replacing something, there’s still someone at the wheel. There’s always someone at the wheel. The AI always revolves around a person within the team.

And if you look at it from that perspective, then two things happen.

One, it’s super easy to measure ROI, because a person has metrics already associated with it, has goals. It’s super easy to understand if we’re measuring productivity, are we measuring more meetings booked? What is that person trying to achieve?

And is the AI you know, helping that person achieve that goal in a better way.

So that turns the entire ROI discussion into a super simple discussion.

The second thing is that what we’re trying to do is to reimagine human-AI collaboration in that sense, instead of just trying to think that AI can run our business autonomously without no one touching it. And that’s not true.

Maybe in 10 more years, that would be true, but the truth is that your business evolves all the time. Your understanding of how your processes should be run, together with technology, evolves all the time, so that your AI systems… and that means that you need someone behind the wheel, and if you implement AI with someone at the center, then you can make sure that you have that operator that can drive that system forward. And so, ROI should not be turned on its head in that sense, okay?

It should be attached to a specific person on the team, and that simplifies everything.

Julia Nimchinski:

Love that, and in terms of the AI nativeness of these role processes and people. I’m just curious your thoughts, Amos, because every time there’s, I don’t know, any conversation about ROI, typically we’re talking about speed. So you could reach X accounts, now you could do XX, triple X, 4X.

But is it really the promise of Agentech AI and, you know, agent phones, per se?

Amos Bar Joseph:

Yeah, so, Short answer is no, okay? Definitely not. Actually, I would say it’s not the promise, it’s the danger.

of AI. And I think that the danger of AI is just buyers not really looking inwards, just trying to think that AI would save their business, and they would just buy this agent, it would just unlock more pipeline.

And if it would buy this agent, it would just unlock more support tickets.

And the truth is that using AI is just not… it’s not about increasing volume.

It’s not about that, actually. And the teams that are using it for that, maybe they get, like, a short-term advantage, but over time, that would just turn into a bidding war where you just spam your buyers, you spam everyone, more activities, less quality.

It’s not kind of like the end game of AI. And basically.

The teams that are using AI the right way, and we’re seeing at Swan, you know, we’re supporting more than 200 businesses that had successful agentic implementation within their business, multi-agent implementations. And… The one common denominator between these teams was that they looked inwards first.

They said. We have a problem with that process. We need to fix this process, actually.

We have a challenge, we have a bottleneck. We think that if we could solve that bottleneck, we could move the needle within the company, right? And that could be maybe, for example, our reps are just, you know, spending too much time on researching accounts, okay, for example.

And if we could actually unblock them, and they would not spend more time on researching accounts, they could spend time on something different.

For example, maybe they would just handle more calls, so we could just increase the number of calls that these AEs are taking, and so more calls for the AEs, just more closed one eventually, right? That is an equation that leads to a really simple and powerful ROI in that sense, but it all started with a bottleneck. that the business had.

And again, people forgot that AI And bottom line is a tool.

It’s a solution, there are vendors who are selling you these solutions, and if you don’t know what your problems are, then no AI agent in the world will tell you what you need to do. And so these 200 customers that had successful multi-agent implementations knew what they wanted to solve.

They knew which process to focus on. They reinvented that process together with AI agents, because they knew where the challenges were. That process was human-AI collaborated in that sense.

It could be that agents can run autonomous tasks on their own, but it revolves around a specific person at the end. Maybe they’re sending that brief to that AE at the end of the process. And so, they knew what the challenge was, they mapped the process, and everything revolved around their humans, and that what led to a successful AI implementation.

Julia Nimchinski:

Fascinating. Well, I want to address our audience a little bit, and the majority of folks watching are mid-market, enterprise, VPs, CXOs. And so, it’s easier to think about those types of implementations when, you know, you’re a smaller company, a disruptor, startup.

But, you know, the classic Lay Christensen dilemma.

what are your thoughts, Amos, in terms of just the easiest ways for teams to start experimenting with those models?

And, both on the reinvention of the company side and, you know, processes and, you know, collaboration, I don’t know, firing, like, the OS of your company, essentially, and the product.

Amos Bar Joseph:

Yeah, so… the fallacy of enterprises, specifically in go-to-market, is, you know, the pressure is so big that everyone is looking for quick wins, right? And… The truth is, to really adopt AI in a meaningful way, you need to change your operating model, right?

You need to change some core, core processes of how your organization is running. And if you’re looking for quick wins, you don’t have the time to change that operating model, and what happens is you’re looking to implement an AI solution to deploy it across your 300 reps tomorrow, and in 2 months, to just see, kind of, like, 150% increase in, you know, pipeline activity, right?

There’s no magic solutions there. Sorry to disappoint you guys. And what you should do, you should start redesigning your organization.

That process takes time. Where does it start? It starts with experimentation.

Where does that experimentation start? With small teams within your company. So, where I’ve seen success with enterprise AI implementations is that they can identify a technical, AI-savvy team within their business.

So, you’re running an SDR, or you have 100 SDRs.

Do you have an SDR manager that manages 8 SDRs, and they could be trusted to actually lead the charge on that transformation? Start… Just with them, okay? They can be your task force.

They could be your experimentation guinea. The more experiments they run, the more they understand what is good for the business and what is not. Okay, and… If you start there, you can understand, kind of like, take it as kind of like, in the lab, try to, you know, isolate them from all the noise, from all the bad processes that you have in your company that you need to change, try to isolate them, and try to give them this island of siloed experience where they can really, you know, redefine how they work with AI.

Start there, have a successful experiment. try to understand how can you design a larger-scale implementation plan across your org with this… without just, you know, disrupting everything and hitting your existing metrics, right? So, I think that this step-by-step approach, starting with a micro-team and just focusing on them as the guinea pigs, enabling them to lead the charge.

I think that’s the best way to do it.

It’s slower, and that’s why a lot of executives don’t like it.

Julia Nimchinski:

I’d like to talk about your internal OS at your company. I know you have a very special story, so let’s dive in into the three founder narrative and debunk some of the You know, not beautiful myths that people tend to think.

Amos Bar Joseph:

Yeah, so.

Julia Nimchinski:

True, is it? Yeah, I just… I want to challenge you, because I, I mean, obviously, we always have some new folks joining, and some old folks, so yeah, let’s just address it.

Amos Bar Joseph:

Yeah, for sure, so… For those of you who are not familiar with Swan, we’re building the first autonomous business. It’s basically a bold experiment to redefine how companies scale from the ground up. We believe that the old playbook is broken, and it’s time to reinvent how companies scale with the different operating system that is designed and architected from the ground up around human AI collaboration rather than human-to-human coordination.

If you look at traditional scaling models, they were built with, you know, hundreds of hundreds of approval processes and meetings and bureaucracies, and what happens over time is that people turn into cogs in the wheel, just small parts in a very huge system.

They’re miniaturized to just, you know, specific steps that they need to follow every day. And what we think is that AI is actually the key to unlock a different model where employees can actually become autonomous.

Employees working with autonomous agents. That’s where the name comes from, autonomous business. It’s not only about AI, it’s about human-AI collaboration in that sense.

And so, at Swan, we’re on a journey to discover, like, the 100x to version of each employee within the company.

That area where we call their zone of genius, where their passions and skills intersect into creating disproportionate value for the company. At the moment, you know, the first step of that experiment, as I advise the enterprises, we started with a small team. We’re just 3 founders, with a bunch of AI agents designed around each person at the team at their zone of genius, enabling us to scale with intelligence, not with headcount.

And just a team of 3 has managed to build an innovative multi-agent solution and go-to-market to serve more than 200 customers in less than 7 months, and scale fast with around 100% month-over- month growth.

The reason why we did that is not because we want to be 3 people at this company forever. What we’re actually aiming for is to get to $10 million ARR per employee as kind of like the north star of building a business designed for AI leverage, designed for human AI collaboration.

And so this is the first step.

And once we nailed that architecture for three people, we can start scaling it and having more autonomous employees joining Swan, in that sense.

And we’re using Swan itself, actually, to build these AI agents internally, and what happens is I’m a single-person department of go-to-market, basically.

And… I’ve built an entire army of agents around my day-to-day, so I can move at enterprise scale with startup speed, basically.

Julia Nimchinski:

Curious to hear more about your thinking about this, Amos. For example, are you… so you’re not starting with a human. But I guess with the problem, or the functions to be solved, and then you’re automating it.

So, the question is, yeah, just the process of thinking, scaling, and what would be your fourth hire?

Amos Bar Joseph:

function. That’s a great question. So… My process is that I try to… look at the areas where I find myself spending a lot of time that are not really generating a lot of ROI for the business, okay?

So, I’m in charge of growth.

And you can look at me as an entire department in that sense.

I, you know, I generate demand, pipeline, meetings booked, I close these meetings, and I make sure that the implementation is successful, so I take it from end to end. But I can look at that, all of these processes, and I can say, you know what, I’m spending a lot of time on LinkedIn, you know, just handling all of these connection requests and seeing if there’s interesting leads there.

Okay, so… can we use an AI agent to help me alleviate some of the, you know, pressure there? So, we didn’t start with saying which cool AI agents could actually, you know, move the business.

We actually said, okay, I’m spending time reviewing these connection requests, so can we have an AI agent that would review them, and would start conversations with these people? And same thing with writing posts on LinkedIn, so I spent a lot of time writing posts on LinkedIn.

and, to, you know, it took me a while to get to 30,000 followers and generating more than 1 million impressions every month.

And today, I do it with the help of AI, basically.

I work alongside an AI agent in a collaborative process that takes these 4 hours into, kind of like, you know, 30 minutes, basically. So, the promise is actually not more, right?

It’s not like I’m generating more posts with AI. It’s not about more activity. It’s about looking at the person at the center, trying to understand these bottlenecks, and then looking to automate the areas that are good for us to automate, and amplify the areas where we think that humans should be involved.

And I think that’s something that we’re missing a lot of the time.

So that’s kind of, like, my point of view, and for our… the next hire, I would say.

So there are actually two potential candidates here.

One, of course, is bring in an amazing technical, you know, founding team member who can help, expedite the development speed, but Second, actually, candidate, which is interesting, is, kind of like an academy person, an educational person, who could actually, write content for SWAN, the AI agent, but for our users as well, for all, you know, humans and AI agents together, to better understand how to work with SWAN, how to build AI agents for go-to- market, what AI agents are actually the best ones that, you know, given the current condition of the technology, etc. And so we think that this educational aspect is super important these days, because there’s so much BS running out there, and we feel like We need a sound, a voice of reason, and all that noise.

Julia Nimchinski:

Love the second option, and curious your hiring process, strategy, like, even… I know it’s too early, but still, what are the criteria?

Amos Bar Joseph:

Yeah, so it’s not early, and we spent actually a lot of time, thinking about the philosophy behind it, because, you know, we were just 3 people, and all of a sudden, maybe we’re, you know, we’re not going to be 3 people, and that’s going to be exciting, so how do we think about that? And I think the main difference between, kind of like the traditional hiring methods, is… lies between our approach that differs from the COG culture, what we call it.

So, we don’t actually create these, JVs, these job descriptions, super elaborate these are the constraints, this is where you’re going to spend your entire time, no matter who you are, okay?

We look at a challenge, or an opportunity area that we have, and then we kind of intentionally define a vague description of a person. Like, what is the absolutely necessary skills that we need for that role? Okay, so for example, taking that educational side of things.

I didn’t tell you that we’re looking for product marketing, or for sales enablement. I just told you about, like, an area of skills that I think that this person should have, and, you know, the output that they will actually focus on. And then what we do, we try to curate talented people, and look at each person.

Talk to them, if they’re super talented, and understand if we can build a system around their zone of genius within the company, and would that complement the other people?

Julia Nimchinski:

Hmm. Amos?

Amos Bar Joseph:

Julia?

Julia Nimchinski:

Yeah, hang the pender, one second. Live TV here.

Amos Bar Joseph:

live TV, yeah.

Julia Nimchinski:

Yeah.

Amos Bar Joseph:

How about now?

Julia Nimchinski:

We can hear you.

Amos Bar Joseph:

Sorry about that. Even AWS can fall sometimes. That happened last week, so apologies over here.

So, just wrapping up with the last note, I think that the main difference that, and I think people should think about AI like that as well, is that our business is designed as a system to amplify our employees.

And not vice versa.

We don’t look at these employees as cogs in the wheel in our business. We’re actually looking at that talent and thinking to ourselves, how can we build a system around that talent that will amplify them, that will put them in their zone of genius, so they can create disproportionate value for the company?

Julia Nimchinski:

Love that. Fender, we are about to transition to your demo, but before we do that, Amos, It’s almost 2026. What’s your prediction?

Amos Bar Joseph:

So… In 2026, we’re going to see… real AI agent solutions for the first time.

So, kind of like cursor for everything.

We just saw maybe one until now. We’re going to see a lot more for the first time. This is going to be the second wave of AI startups.

It’s not going to be the companies that you know, because they’re going to lag behind, and they’re going to be the old school. So you’re going to see new companies that are going to emerge and will change everything you know about almost everything you do.

And your entire understanding of how human-AI collaboration will evolve significantly because of that, because things will actually become clearer.

It will be much easier to work with these types of solutions.

Julia Nimchinski:

Awesome. Hope it will happen. Amos, what’s the best way to support you?

Always a pleasure, you know that. But yeah, where should our people go?

Amos Bar Joseph:

Yeah, so I’m on LinkedIn, and I try to write a lot about, you know, this new model of autonomous business and AI agents, so follow me along. I also have a newsletter. called, The Big Shift, at Swan AI, so, if you’re interested in getting kind of like a backseat into how we’re winning and losing as an autonomous business, you can follow that newsletter.

And finally.

I have a digital clone. So if you have some questions, during the session, you can ask me directly, so you can just, you know, go to ChatGPT, to the GPT store, it’s called Autonomos, so, you can ask me all the questions there, it’s trained on all my knowledge, it knows almost what I know, or even more, I don’t know.

Julia Nimchinski:

Super cool. Thank you so much again.