One Brain Across the GTM Loop

Julia Nimchinski:

And next up… We welcome OneMind’s co-founder and CTO, Sachin Bhat. Very excited to have you here. How have you been?

Sachin Bhat:

Been good. Thanks, Uriya. Excited to be here as well, and excited to take a part of the Undercursor GTM series. Awesome.

Julia Nimchinski:

Awesome. Let’s get into it.

Sachin Bhat:

Perfect, let me just quickly… Sure, let me start with a quick slide run of what OneMind does, and would then love to give a quick product demo, as well as, take questions thereafter. To give some context to what does one mind… what is one mind? We build GTM superhumans. Superhumans is what we call our GTM agents, and we are full lifecycle agents.

Before diving in, let me tell what, our core belief has been of how GTM has evolved, what we think is true going forward, and how we are positioning ourselves for it. First, GTM started off as a systems of record, CRMs, close to two decades ago, started on the systems of record in the cloud, and then it started with systems of engagement, outreach intent tools, where the engagement in GTM happened.

That’s the one that we started looking into the last, decade. But now, we are in the systems of outcome world, wherein AI is not just built on top of the record or doing the engagement, but it’s actually owning the outcome itself. An entire autonomous agent can manage end-to-end an account from pre-sales all the way to close and likely renewal as well.

And that’s basically the core change that we are seeing in the market. And… The core beliefs that we have that’s driving this change is one.

Human bandwidth is not the limit for GTM anymore. there used to be a time wherein we built a lot of optimization on the engagement layer, on the system of record layer, to figure out who is a good lead, who should we spend the sales GTM demo or bandwidth on, who should get a demo, who should not get a demo, what is a qualification funnel, pipeline, ease, etc.

We don’t need that anymore. we can basically spin up an agent for every single buyer out there. Everyone can get their own personal agent. The agent can demo them on demand, anytime, any hour, any time zone, any language. It can pitch, qualify, and close for every single interested buyer. We don’t need the qualification pipeline.

The second major… Belief that we have is that the rule boundaries was a very human artifact. A salesperson could not do support. A marketing person cannot do account management. Yes, all of that is true, but agents don’t have that limit. An agent that is selling. Can also handle the support question, and can also manage the customer onboarding, adoption, and support. And even… Carry the entire account from.

The pre-sale all the way till renewal. So these are the core fundamental beliefs that we are building our company on. And that comes down to comes to one mind superhumans what we are building. GTM agents built on a single GTM brain that manages an account from first touch to close, including taking the job of the… doing the job of an SDR for scholars.

Sales engineers, CSMs, as well as in-product guides and implementation engineers. Our agents deliver the post, queue demo, handle objections, build value models, manage ROA calculators. Track the revenue, track the usage. Upsell, cross-sell, and manage… provide support to the customers as well. How do we do this? So fundamentally, we have a single GTM brain.

That is participating with the buyer across all the channels, website, meetings, email. He’s doing live demo, tracking the solutions with the buyer, he’s managing contextual conversations across multiple meetings and across emails. And this way, our agents can basically, you know, cover from SDR awareness all the way to AM expansion.

We today have already been deployed with some of the most, weakest enterprise brands out there, right, from LinkedIn, Altruism Info, Outreach, Nutanix, and so on. 100 plus customers, and we are selling 4 major use cases. We have a website inbound, which tracks the pre-sale on the website, converting leads who are visiting the website. We have a solutions engineer, a sales engineer, which attends the meetings.

Manages context across meetings, does the email back and forth, carries forward any demo requests, any… answers the hard technical questions, and is a co-pilot assistant for the account executives in those meetings. We also have an in-product PLG guide agent that scales one is to one and provides personalized onboarding and adoption to our customers.

Primarily does the job of an onboarding guide and a product demo guide. And then we have a customer success agent that is useful for the renewal conversations, for the implementation support, in-app support. So, I think this… between these four agents, we are able to capture the entire funnel, right, from acquisition all the way to retention and expansion.

Covering engagement, education, pitch, demo, qualification, deep technical conversations, and consultation, onboarding education, as well as activation across the board. Now, this is the underlying architecture, though, how a single brain covers the entire GTM funnel in a Recursive fashion. As you can see, we have a harness at the top that covers the buyer to the agent interaction.

And then we have the four surface areas, website, meeting, Zoom, we are support… we are already support on ZoomMate Teams. We have an in-product agent as well as an email agent. And then the brain itself. The brain has skills which understands the role of the job the agent needs to do. A sales engineer is a skill, a customer success is a skill, website SDR, BDR is a skill, implementation engineer is a skill.

Every agent is armed with a bunch of sub-agent capabilities. Live demo is a capability, screen understanding, seeing what is on the screen and speaking for it, co-browsing with the user on their product. Creating slides instantly, in different languages, based on the conversation. ROA calculators, ROA deck creators, meeting chat Q&A, these are all capabilities that all the agents have.

And different skills can reuse these capabilities on demand as and when needed. We had 20 plus integrations that, on the knowledge side, that understand what the agent is supposed to sell, or support, or help the customers adopt on.

And then there’s a self-learning loop itself, which Every agent has a golden set of evals and custom evals that the customers write, simulations based on those evals, and the self-learning loop on top of that, that automatically improves the brain of the agent over time. And then there’s the context that carries forward per account over time.

The context is derived from CRM, from email, from the meeting chat itself, and the workflows that all the users can set up to bring context from their own warehouse, BI tools, and so on. The core thesis that enables this, the Recursive GTM, the knowledge loop for us, is that every single agent, every seat, fundamentally has these five slots. The knowledge about the product.

The surface where the agent is interacting with the buyer. It can be email, website, or meetings. The context per account from the CRM, the context is the one that keeps on getting updated per account on a regular basis. It can come from CRM, it can come from email, it can come from different workflows that are set up.

Skills, the… what the role of the agent is, what it is supposed to achieve, what is the outcome, and then the sub-bases, the different capabilities that each agent has. Just as an example, a sales engineer would basically have the surface of ZoomMe website, email, their context comes from Salesforce, HubSpot, CRM workflows, the knowledge could be from GDR, SharePoint, Notion, iSpot.

The sales engineer skills could be technical deep dive, security review, objection handling, and the capabilities that they need is a live demo of this case generator, ROA calculator, and so on. And if you look at the customer success agent. Pretty similar set of surfaces. The skills change. It becomes onboarding, pause, adoption, place, renewal prep. The knowledge changes to be runbooks, Notion, Zendesk, Confluence.

The context changes via analytics, dashboards, product workflows. The capabilities tend to be pretty similar. Instead of a ROI generator, you now have a renewal case generator. Live demos continue to be there, and so on. So that’s basically the underlying belief in how we have built all of our agents to manage the entire account lifecycle.

Now, coming to the context itself, because that’s the one that keeps on changing, that’s where the actual underlying Recursive layer comes into play. For each account, our agents are participating at the website, at the different surfaces where the buyer is present, and are continuously learning about the buyer.

Each interaction creates context for us that we learn from, and our agents continue to use that in the further interactions with the particular buyer. The way we ingest context is basically we capture north of, you know, 60, 70 different major signals per account. We capture the source of these signals, where it’s coming from. We also have bi-temporal validity.

Which tells us the signals. valid from and valid until timelines, and the sensitivity, the permissioning of who should be able to here, what context, and when it should be used. Also, these signals have different types. There’s an ephemeral signal, which, you know, can change very regularly, the more volatile ones and the stable ones, such as compliance, circumstance, and scale indicators.

We extract signals across multiple different layers. The conversion signals and the qualification outcome are some of the examples of the signal catalog that we have. And we also have authority for the signals.

The actual CRM would be the highest authority, a human correction would be followed through, and then the actual human providing that in the live meeting itself, and then the agent inferring based on the conversation, as well as from other third-party integrations that we maintain ourselves. And finally, all of the signals improve over time.

Signals have a decay layer wherein the agent can forget what is not relevant, and a memory layer that keeps on appending new context so that the agent becomes better with each single interaction with the user. And we have also built all of this through a self-serve platform, wherein the users can set up their own integrations, their own knowledge of their own skills and brain templates, wherein they can update.

And the self-learning loop, wherein they can track how the agents are performing through the evals, see what updates need to be made, and approve the updates themselves, and the tool for builder. All of that is now in our platform. I would love to quickly show everyone. The platform itself, and take the, I’ll pick your demo. So this is our, one-man platform.

You basically can see that different teams can come in and set up their own different agents. For example, Nigel is one of the agents that we have that can attend meetings and is a sales engineer agent. We start with the brain of it, wherein there’s a lot of skill information about the agent.

Now, all of this is not straightforward for a user to manage themselves, so we provide an assistant, wherein we can see, for example. Ask email before sharing. I can basically, if I say this, the assistant will work on how to So, I add this… skill, or at this prompt update, to the brain of the agent. You can see that there’s a lot of different, agents that you can basically, attach.

You can provide, start your initial chat, the agent will work on it, we’ll come back to this in some time when it will actually make the brain changes for this.

Separately, we do also have a scorecard wherein we can track how the agents are performing across different sessions, and then there is an anti-implement hub itself, wherein it’s saying where there was a miss. in an agent, and it’s suggesting what other improvements that we should make, and I can basically use Assistant to draft these updates to the brain directly. And then mark complete.

Or we can automatically make these changes directly and go to the brain every single time. For example, during a walkthrough in a live demo, there was a next step question that was being missed, and the implementer flagged it, suggested this is a change that we should make. It said that we can basically add this particular change to the assistant.

And then we have Signal Studio, wherein we can capture all the different signals. Different agents have different signals over here. For example, Mindia website BDR will have a bunch of different signals that we have set up on how to capture this. And then we have the superhuman knowledge itself. This is the actual knowledge that the agent leverages during the conversation. It can be presentable and non-presentable.

Presentable once the agent will present in the conversation, and the non-presentable once it will use to speak about it. And then we have the different deployments, wherein you can come and pick and choose the face, the voice, the languages about each particular, agent. You basically can choose all the different languages the agent needs to, use.

This is a website BDR agent, and this is the kind of deployment that it will look like. We also have an email where you can set up the campaigns, you have an inbox where the agent’s full inbox back and forth can be seen over here as well.

At the workspace level, we track a lot of analytics that, you know, at each individual account level, we can see how it’s performing, what are the different conversations that are present about the company, can manage all the different Agents as well, at the main superhuman level. quickly, quickly going back to the assistant, the live demo agent that we had. I had started an assistant thread over here.

We can see how it performed. It suggested… so the email capture before Argo Edeker had asked. It suggested, you know, this is the current conversation flow, you can ask once, you can suggest a required email compulsorily, you can keep current behavior. I can suggest this, it will then tell me what are the changes to make. And which I can use thereafter.

We have also different integrations that you can use, or an entire workflow layer that you can set up, as well as a scheduled meeting there, as well, that you can use with the agents itself. Now, let me quickly show a live example of how this plays out. But before that, let’s see. It’s still doing the working.

For… if I have to call, Nigel, our sales engineer agent, into the conversation, all I have to do is go to the meetings. take the particular meeting link, I can also set up directly an integration with our calendar itself, or in this case, because I want to just go one of them, or hard skills, exchange. Demo.

I have different modes in which the agent can join the conversation, and then we have… if I want to generate some context for this particular account, I can pick and choose the context I want, and then I can basically start the agent. And we can see that, the agent will then be able to join the call once Julia, allows the agent into the meeting.

Yeah, meanwhile, while this is still working, or… Let me just quickly… we can wait on this one for a while. Julia, I think, to invite the agent in the meeting, you might have to let the agent attend this one. Might be in the waiting room.

Julia Nimchinski:

Yeah, we’re still waiting, no one in the waiting yet, but While we are…

Sachin Bhat:

waiting.

Julia Nimchinski:

Yeah.

Sachin Bhat:

Yeah, while we are waiting, we can also… there is the other agent that I also wanted to demo quickly, which is our live console Superhuman. I think the webinar one, it might not… Be able to directly join. This is our other Superhuman. Can you hear the audio here, Julia, or should I be sharing the sound?

Julia Nimchinski:

No.

Sachin Bhat:

You can’t hear it.

Julia Nimchinski:

–

Sachin Bhat:

Simon, right? Okay, let me just quickly update the share settings to make sure that I’m able to share with the audio on. Dated song… Alright.

Audio shared by Sachin Bhat:

All good. I can hear you through the conversation interface, and I’m here when you’re…

Sachin Bhat:

Hi, Simon. So, Simon is an in-product guide, and he’s an agent that is present in the product, and can suggest or help the user use the product live during the conversation. Hi, Simon. I want to give a quick demo of how you can be used to help a helper users in the product onboarding and adoption guide. Can you tell me how to add a new document to my knowledge base?

Audio shared by Sachin Bhat:

Sure thing. I’ll show the actual upload path in the console rather than just describe it. Knowledge Hub is the workspace wide library, so this is where you add documents that can later be assigned to one or more superhumans. The Add Document flow is where you choose the source type, like uploading a file, adding a URL, or pulling from a connected content system.

Upload file is the path you’d use for PDFs, decks, docs, or other source material you want the superhuman to learn from. This is the drop zone, where you can click Choose File or drag a document in, then optionally assign tags before uploading. That’s the basic upload path.

One key distinction Knowledge Hub is the shared source library, while Superhuman Knowledge is where you see what a specific superhuman has access to. Do you want me to show how to assign uploaded knowledge to a superhuman?

Sachin Bhat:

No, I think that’s fine. Let’s actually go into the assistant itself and the improvement hub. Can you explain to me how the improvement hub can be used? To recursively improve the agent’s brain. Overtime.

Audio shared by Sachin Bhat:

Absolutely. This is the agent gets sharper every cycle part of the platform. I’m closing the upload flow first so we can get back to the Superhuman Improvement Tools cleanly. Improvement Hub is where conversation, feedback and performance findings become a queue of things to review, prioritize and turn into training improvements.

Triaged is the clean working view where raw issues get organized into actionable tickets instead of staying as messy feedback. Each ticket represents a concrete improvement opportunity, like a missed instruction, weak response pattern, knowledge gap, or behavior that needs tightening. Have Assistant Draft is the bridge from finding the issue to proposing a concrete change that can improve the superhuman’s brain.

That’s the Recursive loop. Conversations create evidence. Evidence becomes tickets. Tickets become proposed brain changes. And those changes can be tested before release. Want me to click into the assistant drafting flow without submitting anything?

Sachin Bhat:

No, Simon. Julia just shared with me the updated link for agents to join this webinar. Can you actually go to meetings and show to me where I can invite The ride-along agent into the meeting.

Audio shared by Sachin Bhat:

Got it. You’re talking about the ride along Superhuman joining a live meeting or webinar. I’ll take us to meetings and point out the invite path without changing anything. Meetings is where Ridealong Superhumans are managed for conferencing, so this is the right place to bring them into Zoom, Teams or Google Meet workflows.

Add to Meeting is the main action for inviting Nigel into a call, and you can also use the Superhuman email address in a calendar invite when that workflow fits better. That’s the spot. You can click Add to Meeting to get the invite flow or add Nigel’s Superhuman email address directly to the calendar invite so he shows up on the scheduled call. Do you want me to show the how scheduling works explainer too?

Sachin Bhat:

Let me try this once. And, you know, if Nigel can join the meeting, then I think we’ll wait on that. Perfect. Nigel is joining the meeting. Simon, you can take a back seat. So, I think Nigel is now in the meeting. We have different meeting controls for Nigel as well. In the active mode, Nigel will jump in a lot more. In standby, only when he’s called for. In silent, he will take a back seat. So let’s see.

I think we’ll have Nigel present for the rest of the meeting. Hey, Nigel! We are in a hard skills exchange webinar, wherein I wanted to demo your favorites, and primarily highlight how you can act as a sales engineer in a meeting. Can you give a quick product demo of… or rather, can you quickly tell us what you are capable of, and how you can participate in a meeting and help an account executive in a meeting?

Superhuman Nigel AI:

Here’s the ride-along view. I join the meeting alongside the seller, listen to the conversation, and step in when I can add product or technical depth. That can mean answering architecture and integration questions, pulling up the right visual, handling objections, summarizing what was discussed, or responding in chat while the human presenter keeps going. What you’re seeing now is the core pattern.

The human runs the meeting and I provide an always available product expert layer without needing to staff a human SE on every call.

Sachin Bhat:

Love it. That is great. Can you tell me, do you have lightning of favorites, Nigel?

Superhuman Nigel AI:

I may have missed the phrase there. If you’re asking whether I can stay lighter touch in a meeting, yes, that’s exactly how Ride Along works.

Sachin Bhat:

Now, Nigel, I meant, can you do live demo? And if so, can you quickly showcase Signal Studio for us? In Alberta.

Superhuman Nigel AI:

Let me pull the console up and walk it through. This is the console where I was created. So let me show Signal Studio where teams define the conversation signals they want to track. Signal Studio is where you define the topics, pain points and outcomes the platform should identify across conversations. Signal Studio turns conversation data into structured intelligence.

So instead of reviewing every transcript manually, your team can define what matters and filter on it later. What signal would you want to track in a sales engineering meeting?

Sachin Bhat:

I love it. I think, can you show a quick slide deck about our architecture?

Superhuman Nigel AI:

Here’s the architecture at a glance. The platform connects customer surfaces like web, apps. Zoom, Teams, Meet, CRM and content sources into the OneMind services layer. The important part is that the Superhuman is not a single static chat surface. It has interaction, streaming content, memory, voice and rendering services working together so it can answer, present and participate in real time.

Sachin Bhat:

I love it. This is great, Nigel. And what other cavities do you have? Can you see the screen? Can you… answer chat Q&A.

Superhuman Nigel AI:

Yes, in Ridealong I can answer spoken questions, respond to meeting chat when participants mention me, and support the seller with slides or live product walkthroughs when that capability is configured. Screen awareness is the ability to use what someone shares on screen as context, and we’ll have that for beta testing soon.

For this call, I can use the screen context available to me, and the meeting chat Q and A uses the same knowledge and guardrails as my spoken answers.

Sachin Bhat:

This is lovely. This is great. Thanks, Nigel. I think you can take a back seat. I wanted to quickly. So that’s primarily the different, capabilities that we have. We do have, like, close to another 3-4 minutes into the meeting. I want to make sure that if there are any questions, I would love to answer that, and take this time. To… to handle any Q&A about the… from the audience.

Julia Nimchinski:

Thank you so much, Sachin. Great session. I’m curious, your, you know, CTO lens, how much context do your agents need in order to operate this efficiently?

Sachin Bhat:

Third. Oh. I think we can work with pretty low context, as well as a lot of context. It’s mostly about the efficacy of the agent, to be honest, but I think for… light touch agents, like website BDR, that context can be thin.

We mostly… there’s a lot… not a lot of personalization that happens over there. we are catering to a lot of en masse visitors, but for something like a sales engineer or a customer success engineer agent, they thrive on the context that is present in the system of record, like a CRM or a BI tool, as well as the email back and forth and the prior meeting context.

We have seen, agents perform a lot better once, after the first touch, second meeting, third meeting. At that time, they can personalize, they can track the previous ask, and actually respond with that context a lot more effectively, and the wow moment becomes a lot more obvious at that point.

Julia Nimchinski:

And in terms of recursion, Amanda always , you know, the handoff problem. I’m curious if recursion is happening only within the context of one agent, or agent-to-agent recursion.

Sachin Bhat:

The way we have built currently is an agent-to-agent recursion. It’s a… interesting play, this form of agent versus one big agent is always a question that the market will have to answer. Right now, we are going the small agent approach. There’s one primary agent that does trigger a lot of other agents as well, etc.

But to answer this particular question, the way the handoff happens is from the pre-sale, all the way to onboarding, and then to the post-sale and the renewal conversations. These are different agents, because they need different contexts, different knowledge, different integrations, different skills, and the context handoff happens between them, and the same context can be used by any of these agents as well.

They will pick and choose what context to use. And so there is a continuous learning loop. Every single interaction with an account across all these agents acts through the same underlying context. And that regression will further fuel in how the different agents can. Improved themselves with each individual interaction.

Julia Nimchinski:

You’re persistently innovating on, you know, the GTM front, and obviously all things Agentic. Really amazed by the recent capability, you know, of agents doing the demo within the customer’s environment. Curious, what’s on the roadmap?

Sachin Bhat:

There are a few things that I can say, there are a few things that I cannot say, but I think Nigel has the answer wherein he can exactly say what he’s allowed to say. Nigel, do you want to take the question, though? What is on our roadmap?

Superhuman Nigel AI:

The biggest roadmap areas are live demo, screen awareness, deeper cross-meeting memory, calendar-based context resolution, and more native scheduling and follow-up. Live Demo is being configured for customers and launched as a beta. Screen Awareness will be in beta testing soon.

Cross meeting memory is in development with work starting in September, while CRM based continuity already works today through pre processing and post processing. Native calendar and routing is already in external beta and automated calendar context resolution is targeting mid August.

Sachin Bhat:

Pretty good. I think he has the approved one that is much more easier to share, probably.

Julia Nimchinski:

We’ll soon have just agents demoing, you know, the webinars. This is great. Where should our community go, Sachin? What’s What’s the best next step?

Sachin Bhat:

I think this is… in a world wherein the product and the market are both changing simultaneously, I fundamentally believe everyone’s job will become one of managing and evaluating agents. Can they improve the agents to do a better job? When every single company, every single GTM team have their own agents, which agent is performed better? How do you look at the traces? How do you evaluate? How do you improve?

How do you… make your agent better than the competitors. Agent is the fundamental skill set, so this community will have to become You know, from being hands-on themselves to being managers of agents, understanding how to build great agents, and how to continuously improve them, and stay ahead of the market.

And I think It’s a very exciting time to be here, and I think the competition will always fuel more and more innovation. There’s a ton of work, and we are still scratching the surface, and how they… how to manage agents. And I think, you know, that’s what I’m most excited about. As a community.

GTM, we are just understanding the impact, and very soon, I think we will all Learn the art and the skill of managing these agents and scaling them.

Julia Nimchinski:

Amazing. Thank you so much.

Sachin Bhat:

Same. Thanks, Olivia. Thanks, team. Or I will take Nigel along with me back as well. Oh.