Coordinated Intelligence: Moving from Disconnected Agents to Goal-driven Swarms

Julia Nimchinski:

And welcome to the show, Depender Dengra and Tajaz Kapoor, founder and CEO and product manager leading all AI initiatives at Rapture.

Welcome to the show, how are you doing?

Deepinder Singh Dhingra:

Thank you, Julia, for having us. We’re doing great.

Tejas:

Sure, Leh, all good.

Julia Nimchinski:

Yeah, great to have you here. Well, you’re always ahead of the curve, but this time particularly. I know that you are actively deploying this concept, agent swarms, so before we dive in into the demo and see it all in action, just curious to hear some early results, ROI, case studies.

Deepinder Singh Dhingra:

Yeah, so we’ve been able to… sorry about that. We’ve been working at, you know, building out our agentic AI platform, and Right from the start, we conceptualize it as not one agent. But as a system of agents, and a team of agents, or the agent swap.

Right? And our key approach has been we wanted to deploy full funnel agents for B2B co-market teams.

So we’ve been working with customers to deploy agents for them.

One of our customers is also another agent AI company. In more of the business professionals, they develop agents for business professionals. And we helped them, they launched last October.

Around, and we help them scale to over a million signups and users by deploying a swarm of agents that goes across, helping them figure out how to prioritize the right leads, how to improve targeting. for the right leads across multiple channels, LinkedIn, Twitter, etc. Agents wants to help them figure out, do better enrichment of their leads, because they’re getting a lot of sign-ups through Gmail.

and then also to help them do personalized email outreach.

And this combination of agent swarms has helped them scale effectively, reduce their cost per sign- up, and we’re very proud of… proud to be a part of their journey.

They’re now greater than 2 million sign- ups and users. Within that. So that’s the early results that we’ve been seeing.

Julia Nimchinski:

Super impressive. Let’s dive into it.

Deepinder Singh Dhingra:

Perfect. So what I thought, is I will start by giving a quick overview of our approach and how it works, and then we will jump into, you know, the actual working session and demo that Tejas will take us through.

So, I was just talking about this case study of ours, where we’ve deployed a team of agents.

What you’ll notice.

is that our approach is to deploy a swarm of agents across the full funnel of B2B go-to-market teams, which means that we’re really focused on optimizing the end-to-end buyer journey. Right, from the anonymous visitor, through to the different funnel stages and the buying stages, through to opportunity, close one, and beyond. right from… the marketing motion, to the SDR video motion, the AE motion, the customer success motion, the partner motion across digital, non-digital channels.

Within that, we solve broadly for three broad use cases. The first is marketing optimization, how do you crush your ROI? Second is pipeline acceleration, how do you drive speed to pipeline?

And third is, how do you drive better revenue productivity?

Now, this takes the form of deploying a team of agents across the GTM ocean.

Now, what’s interesting about this team of agents is that this team of agents are not operating in silos.

What traditionally happens is that if you have an agent for inbound prospecting or outbound prospecting, or an agent for cold calling, or for personalized email outreaches, or you have an SDR agent, those agents have very localized inputs.

They are just looking at A local information, of that particular lead, prospect account. or opportunity, but they don’t see what’s happening before or what’s happening after. So if you’re an agent focused on SDRs, you might not have any context Around what happened with those leads.

And those, contacts prior to the SDR motion, what… did they interact with me on Google?

Did they visit my website? Did they go to a trade show? To a conference?

Did they enter… did they attend a webinar event? Usually, all of that information is not available to that localized SDR agent. Now, different organizations… might try to instrument that context, but our approach has been to intentionally provide full funnel context to every agent, and that’s what happens in our data model and our data graph.

What we do is we integrate data and context across the multiple silos in the enterprise DTM motion, and whatever I’m showing will be made real to you by this, right?

So we integrate all of the data across the… across the marketing motion, the SGI the emotion, the e-motion, so the marketing automation, the chat engagement tools from the website, the… buyer intelligence tool.

from all of data, the advertising platforms, the paid media tools, the ABM tools, etc. And we harmonize that data into a data graph.

This is very different from the MCP approach, which is very valuable in itself, but what it does not do is harmonize data, duty, resolution. Do the deduplication. do the change data captures, do the resolution at the lead level, account level, campaign level, channel level.

Once we are able to do that, what that allows us to do is to put together this B2B GTM data graph. That then serves as full funnel context. That, combined with our AI learning engines, predictive learning engines that learn at scale.

What is the propensity of a lead, account, opportunity.

to convert to pipeline or revenue? What is the next best action?

Our AI learning engines start learning that at scale, and that becomes the context for our various AI agents to take action. Using this approach, we are sure that every agent has full funnel context. not just context about what happened to the lead in the marketing motion, not just context about what happened to a prospect in the AE motion, or not just context about select conversations that the sales team is happening.

It has full funnel context, structured data, unstructured data across all of these data sources. And that’s what allows us to then become the one secure, centralized platform for deploying agentic AI.

for enterprise B2B go-to-market at every stage of the funnel, and each of these agents are learning and are coordinating from the full funnel context, and can be deployed and configured by region, channel, persona, segment for that enterprise B2B GTA motion. So that’s how the approach works.

Now I’m going to switch it over to Tejas, who’s actually going to make all of this real, but as you go through that, please make sure you actually look for what I’ve actually just mentioned in terms of our approach.

So, over to you, Tejas, yeah.

Tejas:

Thanks, Lupinda. I think that was a great intro.

Julia, if I could get screen share access? Definitely.

Julia Nimchinski:

Definitely. Unbelievably hot. Yep.

Tejas:

Yep, got it. Apologies. So, I think, as Dipinder touched upon it, we’re building the most context- aware agentic platform out there, which has all your GTM data ingested into it, in addition to External data that we work with external data providers for, to kind of give you a… to kind of give the agents the complete context around a lead opportunity accounts journey.

So, I think let’s get into the demo.

Just a second. So, what you’re looking at right now is what we call our agent portal.

So, this is where our agent swarms live.

And each agent is purpose-built for a specific task with clear downstream use cases.

So we like to think of these as individual agents serving specific functions that can work together to also, achieve certain goals.

So, our agents… now, when we talk about the full funnel, we mean that our agents operate from your de-anonymized visitors that come onto your website, all the way to your closed one and beyond, into upsells, renewals, and expansions.

So, our agents have context from the top right to beyond closed one.

And if we could now get into how we actually build these agents.

So, we have an agent builder which effectively gives, you know, marketing teams, GTM teams, the power to build their own agents specific to their use case in a simple, no-code, drag-and-drop format.

So, if we could look at this as an example.

So, this is a particular swarm that we’ve developed right now.

So if we go in here, starts pretty simple, you know, you just name it, give it a crisp description.

attack to kind of help people on the portal identify what this agent does. And you can… so we are basically housed on the Google Gemini model, so we… you choose your model.

The next step is actually setting up how the agent gets triggered. So now we have 3 available options at the moment.

Manual is pretty simple, straightforward. You come in here and run it yourself. Or you could run it on a schedule, depending on, you know, the use case.

And what… another interesting one we have is event-based.

This is a real-time trigger, so this When an event is completed, the agent… agentic flow gets triggered in real time. Without any manual intervention. So, in this example, so what we do is we deploy our first- party pixel on our customers’ websites to capture activity up on there.

So, right now, we’ve got our first-party pixel as an event source, and within that event source, we’ve got web visits and form submits as event types.

So, if we go in here.

Right? Now, you can see that now that I’ve switched it to a real-time event, you… and it’s completely configurable, that’s what’s beautiful about it.

You can set up your own custom conditions, you can set up the exact page you want the agent to watch, which houses You know, which brings in your visitors, or which houses your form. You can select lead details, you can select, you know, country, a bunch of things that actually help you narrow down who you want the agent to actually respond to, instead of it being like a… like, you know, a blanket thing which most generic, tools would provide, like HubSpot will give you, you know, a response to a form fill. But the difference is that response will not have any prior context into who I am, what my prior engagements are with your company, you know, what I’ve done before.

Whereas ours has all that full funnel context living in it. So our agent will be able to say, hey.

this person’s engaged on Agentic AI before, so I’m gonna send him an email on Agentic AI, for example. So, getting into the rest of our builder process, so, it starts with an input from a user, so this could be as specific as a single lead, or you could go broad enough to say, okay, I want to run this agent for all MQLs generated from the USA in the last two days, for example.

So we give that range of flexibility, and every single filter within our platform can go into the user to help define the input itself.

Right, so, and how this effectively becomes a swarm is, Now… We have a separate agent that does contact enrichment, that taps into a bunch of data sources that we work with. So you can see here that we’ve got about 4 or 5 providers right now.

You can select the ones most relevant to you, you can select the flow of enrichment also, the sequence. And how this becomes a swarm is basically this is a separate agent, what we call contact enrichment.

And now what’s happening in this particular use case is.

The lead’s information that comes in is going into our own data, getting harmonized in our data model, all the engagements, all the prior touches, journeys, activities, everything is getting identified over here from our existing data. It’s getting enriched, that same lead is now getting further enriched from external sources. In this case, it’s LinkedIn, so we’ll be getting in pain points, an overview of the profile, and key topics to engage with the person based on what they have mentioned on their profile.

And then, you can see all that enriched information is now being passed down into where generative AI comes in to come and say, okay, you know, I want to send this person personalized emails, right?

Now, within that use case, we’ve got guardrails to ensure that you know, the agents perform as expected, and there’s always a human in the loop, because we’ve spoken to a lot of GTM teams, and the biggest thing this is, they want agentic execution, but with human-guided intervention and with, you know, with the human behind the wheel, as Amos mentioned as well.

So, this email preview, essentially.

Tells our AI to stitch out the email. But it waits for a human to come in and review the email before it gets sent out.

And test mode is, again, a testing interface where you can send it to a dummy inbox.

And now, now that we’ve, you know, selected LinkedIn enrichment, we’ve even thought about edge cases where, for sometimes, for a particular person, we might find more than one profile.

It might not always map to the right profile when you’re enriching, so we’ve even set up a match threshold, which the user can decide to kind of ensure that The right profile and the right context is passed every time, instead of blindly pulling in stuff that isn’t relevant. And we could even give it a further prompt here to kind of define the use case downstream.

So we could say, focus on the LinkedIn headline and the personalization.

So this is always evolving.

This can be configured at any time. Again, giving that whole configurability and customization back to the user. So now, if we go back and have a look at the… sorry, and to even touch upon the outputs, right?

So we can have the outputs here.

Or what we can do is we can connect with multiple tools, like Marketo, for example, Outreach, for example, and even push our… the content our AI generates for those emails into these tools, so it can actually sit within your existing deliverability infrastructure, and can follow your existing company policies. So, if we make this real with, so I’m going to take you through a run that we just triggered.

So, we’re looking at a particular lead in this one, and if I go in to have a look at the email that was generated.

So, if you look at this, Right?

You can see that the agent has gone into her LinkedIn profile, seen that she leads the global go-to- market strategy at Genesis. Found that, ensuring predictable pipeline revenue growth is one of her often-talked-about points from her posts and from her profile. And stitched out an email based on that problem statement, and then linking how Revshaw, in this case, because this is our own instance.

solves that problem with our platform. And again, it’s… we’ve, you know, we’ve got, references and things coming in from a knowledge base as well.

And again, this entire email format is completely customizable.

This is just how we do it. So if, you know, a customer were to say, no, I don’t like to do emails like that.

It’s all generated from a prompt, which we have in our config center. To ensure that, you know… see, if you look here, we’ve got our personalized email prompt, so this is… this gives it all the context that we have.

Deepinder Singh Dhingra:

And if I would just come in here, as Tejas is talking about this, and if you go back to the config center, it’s very important… for Agentic AI to be guided by an understanding your brand tone, understanding your business, understanding what is your value proposition, understanding your specific case studies, your product data sheets.

All of this is available to the agents. Each of the agents are being fed through this context. Further, if you go down here, you’ll also be able to see that there are controls on changing prompts, right?

You can define your ICPs, you can define your prompts for your personalized email agent, for your SDR agent, for your social selling agents, for your context generation agents, for your summarization agents. All of these are prompts that can be edited by the company. We are trying to operate in an enterprise context.

where… customers want to be able to configure the context of their own business and feed it to the agents. Further, if you go down, you could do this by the agentic flow. You also have controls on security, privacy, right?

So we support… we work in a very secure way.

We support PII redaction, no prompt logging. no prompt caching. You can go to… you can guide the responses and the privacy if you just click on those tabs, response and privacy tabs, right?

different levels of control, you can look at what brand tone you want, you can do PIR redaction for, let’s say, you’re going to send emails or contact names, so the LLM models that are getting used will never have access to that. So there’s PII redaction before it’s a model, and then comes back, right?

These are very important aspects for deploying full funnel AI. within the enterprise for B2B GTM. Yeah, go ahead, Tejas.

Tejas:

Yep, thanks a lot, I completely agree. So, you know, at the enterprise scale, what we often see is there’s a specific corporate policy on how they send emails, and that must be adhered to all the time.

So, what you saw in our config center is essentially that corporate policy, and now.

For individual users, we can enable more downstream use cases. Now, a prime example is this one here.

So, this is an agent, basically, that de-anonymizes visitors that come to your website that’s powered by a data graph.

And then, you know, pushes all that information on those people.

down to our personalized email agent that understands their prior… say, their page visits, for example. So if I were to touch upon The downstream use case here, if you have a look here, we’ve… I’ve given it a prompt saying, look at the pages visited by these visitors, and send them an email based on those.

So, I’ve given it that specific instruction to focus on the pages visited here, right?

So now, just to make the context stand out, I will take you through a prior run.

So, if we have a look here, so, for example, this lead right here, Giann Rodriguez.

Subject, Streamline AI Agent Development. Now, if you look.

it’s made it very clear that, you know, it’s referenced agents and AI agents, because I’ve given it that particular instruction. And just to show you where this context is coming from.

Just give me a second. So, yeah, this contact is coming from our full funnel data graph, so… which stitches out these journeys and touchpoints of these leads, opportunities, and accounts.

Deepinder Singh Dhingra:

We can’t see it, they just… yeah. Okay, cool.

Tejas:

So, you know, in this example, like, we were talking about agents, right?

This gentleman has engaged with stuff on agents, and you can see it right here. that he’s engaged with building custom agents, he’s engaged with the display ad on agents, which is why the agent has taken that as its context, because all I’ve said is reference the pages visited. So now, if the pages visited were about improving pipeline, it would have spoken about that.

So, this is how we try to bring that context out into our agentic flows to kind of ensure that each email, each interaction.

Is one that actually resonates with your leads, instead of just, you know, the blind mass emailing method with personalizing name and company, for example. So… Now, if we come back here, now, another interesting use case, I think Amos was talking about, you know, saving time for reps on research.

So, this is another agent that we’re seeing great demand for across GDM teams, because… now, what this agent effectively does is it combines our underlying data that we have from your CRM systems and marketing automation platforms, etc.

with external insights that we work with external data vendors for, to give you a complete view of this company, right? So if you look at the insights, now this is all external information.

This is, you know, hiring signals that we can bring in financials, we can bring in company news. So this is how we’re actually, in fact, looking at bringing in a team of agents for… across different functions in the GTM motion that perform these specific tasks, and that can be clubbed together downstream to build complex workflows and use cases that… that, you know, work towards solving a problem for the user.

And, you know, another thing on…

Deepinder Singh Dhingra:

I think what you’ll also see is that every agent is tracked.

Output of every agent and the input to every agent is tracked. Every agent run can be audited, is audit, trailed, and logged. The reason this is important is, just like in the 2009 or 2008 era, where there was algorithmic trading that went wild and caused market disruptions, you don’t want your agents who… that are running in silos to cause disruptions.

Imagine agents, hundreds of agents that are running in your B2B GTA motion, acting on millions of contacts. Hundreds and thousands of opportunities. Tens of thousands of accounts, if not more.

billions and taking billions of actions. That’s the scale at which we operate. We operate at that scale, and even larger scales, right?

You want to be able to have the ability to track all of your agents and what they’re doing. Without that ability. those agents could just run wild, and you don’t want a repeat of what used to… what happened in the algorithmic trading era in the 2009-2008 to happen with what your… the agents that you’re deploying across your B2B duty of motion, because every lead and every contact, every account, every prospect is too precious to… to not protect, right?

And so that’s the inbuilt capability that we’ve driven at… for this full funnel agent AI platform, yeah.

Really?

Tejas:

Yeah, you know, touching upon what you were saying, Deepinda, on transparency and tracking, so we’ve even got… a dashboard that actually helps your team analyze where maximum tokens are being used, which agents are most popular, just to keep costs in check at all times, and avoid any surprises for teams going down the line that, hey, I’ve been running this for so long, and boom, I’ve got such a big bill from OpenAI. And, you know, like, some of the upcoming features on our Agentic side that we’re looking at is, you know, diving into your campaigns to find who’s the right ICP for this campaign.

So you could give the agent context on what kind of campaign it is, and it will give you an output saying, okay, this is the right ICP for this campaign.

Based on all the data we have about campaigns with similar characteristics. So, for example, if it was a Google ad campaign targeted at marketers in the United States, we would analyze all prior campaigns with those characteristics.

and have a look at the ICP, like, the kind of leads that have engaged the best, and would actually then look back and predict out that, okay, this is the right type of person that you should, that, you know, this is the right ICP for this campaign.

And, you know, another thing that we’re going quite deep into is orchestrating agentic flows from natural language conversations.

So you could dive into our upcoming chat agent and say, hey, find me these leads.

And then you could have a follow-up conversation and say, okay, now research the accounts of these leads.

Okay, now write personalized email to these leads from all that research context that’s coming in. So we’re constantly adding more context in workflows, and then pushing all that context down in towards actionability, such as maybe even writing back lists to your HubSpot or to your Salesforce.

So, yeah, that’s how we’re kind of looking at the agents from our end.

Deepinder Singh Dhingra:

Great.

Tejas:

I think to… yeah, I think over to you the brother.

Deepinder Singh Dhingra:

Yeah, I just shared my screen. Getting a good socks here, e.

Tejas:

Yep.

Deepinder Singh Dhingra:

Perfect. So I’ll end this session off.

I just wanted to reiterate what you… what you saw. First is that we’ve built an entirely secure centralized platform for deploying agents for B2B go-to-market with full funnel context. We’re already seeing significant benefits with this swarm of agents approach that can enable on the same platform, with the same context that each of the agents shared, that then essentially transform your GTM motion, right?

This is not really about deploying one or two agents. And, you know, being happy with that.

This is about transforming your GTM motion. to boost conversions, improve efficiencies, improve speed of responses, and ultimately improve predictability across your revenue. This is about having agents coordinating and being able to deploy multiple agents configured to your GTM motion to your account segments, regions, channels, personas, because in enterprise B2B go-to-market, you’re operating in a complex world.

Right? And so, one could try to be the lovable for agents to deploy an agent, but you also have to make it real for the enterprise.

And it has to operate in a secure way, right, with different compliances, making sure these agents don’t run away, you know, doing billions of actions on your millions of contacts and prospects without having the right guardrails, and making sure compliance around these agents are well taken care of.

So that’s our overall approach to agent swarms, and hopefully that was useful.

Let’s stop here, yeah.

Julia Nimchinski:

So cutting edge and impressive, Dipinder and Jess.

Thank you so much. I’d like to address and build on the security point. We have a lot of enterprise and mid-market executives here.

That’s the core of the community, and so, just curious, based on your experience, Defender, for teams that are just early on their AI journey. And when they see this concept, agent sworn, specifically, it might be frightening and uncomfortable, and the war transform, especially.

So, what would be your advice how to even start transitioning into early experimentation and, you know, some early deploying… deployment of this new architecture?

Deepinder Singh Dhingra:

Yeah, no, absolutely, we can talk about some examples. I won’t name the customer, but I’ll tell you how different customers have gotten started with us. So, in the enterprise, what’s important is to identify a few use cases.

Right? You always start with defining what decisions you want to enable in a more intelligent fashion, in a more personal life. For example, we’re working with a customer.

I showed you a previous example in our case study, but we are working with a customer whose CEO has said, no more mass emails. Right? We want to only send personalized emails with the best context of both first-party, you know, what are those leads, contacts, prospects doing with us, but also understanding those leads and prospects better, based on understanding their LinkedIn interests or other interests.

So that is one very good example of a use case that brings together a practical application, a better way of doing things.

Right? Because you want to personalize, and you want to get as much intelligence into the emails that are being sent.

Has full funnel context is not some agency that is running, you know, independent emails, not knowing what’s happening in your CRM, your marketing automation systems, your paid ad tools, your ABM tools, whether those have engaged or not. It’s not like an outbound dialer agent that is just trying to do inbound prospecting, but getting the same results, because, you know, if you’re going to prospect on a lead without knowing any context, you won’t get the same results, even though you might get some automation with the dialing capability, right? So that’s a very practical use case, and then that’s what we’re doing for them, right?

So build on two or three very practical use cases that are… The need of the hour, the need of the business.

transforms the way you’ve been operating, and that’s how you get started. Within that context, make sure you have a certain policy. So, some of the customers that we work with make us go through an AI review.

Right? They ask us a whole set of questions, so make sure your vendors and your partners follow the policies that you’ve set up, and then get started. And that’s how we’ve seen more success.

Julia Nimchinski:

Thanks, Deepinder. And for, you know, the executives planning their 2026 investments, what would be your advice and generally the prediction in terms of the silo GTM, Agentech GTM?

Where… where is the industry going?

Deepinder Singh Dhingra:

I think the industry is going from siloed to more connected, coordinated agents.

It is going from siloed individual marketing, SDR, BDR, AE motions, to full funnel motion. So I think the lowest hanging fruit are at the intersections of the marketing, the SDR motion.

The SDR motion to the AE motion, the AE motion to the partner motion, the partner motion to the SDR, address those intersections, because that’s where most of the breakage and most of the leakage is happening, and that’s a great place to start. Right? Yeah.

That would be my advice.

Julia Nimchinski:

Love that. And what would be the best way to connect with you after this event?

Deepinder Singh Dhingra:

No, absolutely, you can reach out to us at RefShore. Our website is RefShore.ai, R-E-V-S-U-R-A dot AI. You can reach out to me directly, or to Tejas.

Just our first name at RevShore.ai is our emails, and or you could just book a demo from our… from our website.

Julia Nimchinski:

Awesome. Thank you so much.