---
title: "Building World-class Agentic GTM: Learnings from Monaco for Monaco"
speakers:
  - name: Mihail Eric
    affiliation: Monaco
room: "Continental 1–3 (Workshop Room 1)"
starts_at: 2026-11-10T15:30:00-08:00
ends_at: 2026-11-10T15:55:00-08:00
format: talk
recording: 2026-11-10_1529_Mihail-Eric
duration: 1775.7
source: qwen3-asr-1.7b
---

> Transcribed locally with Qwen3-ASR 1.7B (timings per ~5 s segment), prompted with the
> speaker names of overlapping sessions. "Sonar" in the MC's introduction (00:01:20) and
> "Netflix" at 00:19:44 come from that prompt, not the audio. The Voice Memos transcript
> is kept beside it as `transcript.txt`.

### 00:00:00 — Introduction

Monocle. I'm here for today. My name is Rochelle. Thanks for sitting through the whole day. It's been a whole day of really informative talks. So I think I believe after this.

Like there will be another break, so we can go grab some water and such a time. Yeah, but for Mahel, he is the head of AI at Monaco. It is a startup building. AI-powered revenue platform for GTMT. It's backed by Kiva Capital, Benchmark, and.

Found us, and. He's also professor at Stanford. Where he created and taught. The modern software developer, which is the university's first course on AI software development. He trained hundreds of engineers on AI coding principles and have advised.

A dozens of global software here teams on AI development. Before that, he was also the technical lead at Amazon Alexa, where he built the first LLM. For the organization. And then before that, he was the founder of a white sea-backed AI coding company. Which is called Sonar.

And it is a machine learning education startup that trains thousands of students. Locally, and he led it until its acquisition in 2022. Nehan loves teaching, and his resources and talks have been viewed. For more than five million times on various various channels. And his talk for today will be focusing on building world-class agentic G T M, and his learnings from.

Monaco. Let's welcome Mahel. Very cozy room. I haven't been in this one. This is fun. Great to be here, everyone.

As as I mentioned, I'm the Heil. I Leading AI, a company called Monaco. We are we are the company, not the country. So I hope maybe one day also AI has a country. You know, little steps, little steps.

So today we you know we are an AI native. Sales company and into the market. What I want to speak to you guys all about is how we are building world class into the market. Into our agentic platform for our agents. Hopefully, you guys some principles and strategies to do that and apply it to your own companies and.

Especially if you're building one in virtual or in places. So here is a quick overview of the talk. I'll give a brief overview of kind of the design of the model we built out that is optimized to go at market use cases. I'll talk through some case studies of where this. Very go-to-market focused harness has served us well.

I'll step back a little bit and give some high-level principles for how to build. Products, converted spaces. This will be many. Less pure tech and more just general cultural organizational principles that have worked for us, and I think apply at any company. Well, and then I'll include some party thoughts.

So a bit of discussion on on Monaco.

### 00:03:35 — Monaco and the future of sales

This is a question that I usually get. Whenever I walk into a room, are you the the plane company? And you know we we do. We do fly some planes, some banners around around SF, some which you guys might have seen. But we also do cryptic billboards around the set with dollar signs.

Big numbers, and then we host some pretty popular poker tournaments. So yeah, this is this is the end of my talk. Thanks for being here. No, of course. So who are we actually?

Monaco, at its core, is a all-in-one revenue platform. About two and a half years ago, we took this bet. That the future of sales was not more point solutions, more. Frankensteinian collections of of tools you stitch together. But rather an all-in-one platform that should be guided, that covers the full landscape of tools and functionalities from researching.

Targeting outreach pipeline analytics, pretty much anything that you want in sales. We do, and we built in house. So that's the bet that we took. We're very fortunate to be back. By some of the best investors in in the world.

And then you know, I think I think we're most proud of is just. Haven't had the opportunity to work with some of the best companies in the world, and this includes a few names. That I listed here, but but there are many more more accounts. Just thought I would share some of those here. So very fortunate, critical, happy.

These folks in our corner, building for them every single day. So before I get into maybe how we built this, this. Harness this agentic system. I want to take a step back and and. Share a question that this something that has haunted me, and something it's been on my mind for for a good while, basically since I joined Monaco.

The question is, what is world-class to our market, right? If we're trying to capture this in an Asian platform, what does that even mean? And for me personally, before I was at Monaco, I was actually founding a company. And as a technical founder, you never have problems building technical solutions. You have no problem with products.

But what always seemed to be a struggle of mine was how do you get go to market to work? How do you find the right people to reach out to? How do you build scalable sales processes?

### 00:05:43 — Knowledge and procedure in go-to-market

How do you do that over and over and over? And as the companies grow rapidly, right? So what does that even mean? Over the last two years of being at Monaco, it's sort of been like this mission: like we understand the market at core. And we're fortunate as a company that over the last few years we've built out internally.

We have more customers in our Team, we brought together a lot of the best salespeople. From companies around the world, and they are helping us, of course, sell Monaco. But they're also helping our customers learn how to do that market. And so what I've distilled through all this is that there are a few principles that really separate. Decent go-to market, or I think the best go-to market.

And those principles are knowledge. Okay, so knowledge here in the sales context means things like which companies. You sell to who do you sell to at those companies? What matters to them, right? Sort sort of from a value proposition perspective.

And the best salespeople have an almost encyclopedic knowledge of every company and every company in different industries, right? Beyond that, there's there's a huge juggling piece to to really go to market right. So, in the context of sales, this is who you reach out to. What is the angle to reach out to them? What is the buyer on the receiving side of a sequence or a meeting invite?

What do they care about? And then, when should you be reaching out to them, right? Because it's not the case that people always want to. Receive your email. In fact, very rarely do they actually receive your email.

But it needs to find that perfect point in time where they actually want to buy something. So this is a huge judgment question. There is a huge procedural aspect to world-class markets. This is workflows that you bake in. To do sales tasks, right?

So in the context of sales, this includes things like defining and building. A TAM. If you have an ICP that's been defined. Writing of sequences actually converts right. So outbound, they will actually get meetings booked. That'll get you sales.

Right? So it's a huge set of really good procedures and workflows that the best salespeople know. And then the best of those people are adaptive, right? So they're able to. Dynamically evolve their decisions and their strategies based on things that a company cares about, right? And so this is.

This depends both at the org level how a company is evolving in terms of what it wants to buy at a given point in time. Also, at the user level, your receiving buyer, what do they care about, and why do they care about this right now, right? So that adaptivity is something that really good salespeople are fantastic at doing. And I have a little bit of a hot take that I want to share with you all, which is that in this, we're here at AI, the AI Engineer Code Summit. So obviously, we have strong beliefs about future genetic coding.

I would argue that given how the world is evolving, that today in this world of increasingly commoditized software. Go to market is one of the few remaining sources of alpha, so we are in the business of. You know, sharing our knowledge of the best to do our market, sharing our customers. And that's what we've been trying to do in our platform since day one. All right. What does this mean in practice, right?

### 00:08:33 — The vertical AI harness

So now we know what go-to-market world-class go-to-market is, but what does it mean when we actually put it into a harness? And what I really want to underscore in this part of the talk is that there are. You know, as I'll share kind of the architecture of how we built some of this, there are principles and primitives that are true to any vertical company, vertical AI company. And that these are just fundamental building blocks. I think can apply to anyone that is building a space.

But I think beyond that, there's nuances that are specific to your space. And so, building a foundation, and then really layering all the expertise in your domain in those kind of pockets and nuance. So at its core, this is what I think most vertical systems build in one way, shape, or form. When you kind of look from bottom to top, you have this concept of like the world model and the data access. These are all the data systems, the data platform that we build.

It might be proprietary. Sometimes we're maybe not proprietary data sources. Or at least the vertical specific data sources that you're trying to build out. There is a workspace which is effectively like an agentic execution environment. This is where your agents do their tasks.

This is where they. Or assist their outputs, you know. As they're doing their task, they also evolve and they learn things over time. Right. So there's this huge Hugely important piece around the memory and the persistence of agent work. Right, this is where you start capturing things like judgment.

Like strategies, things that are important to your vertical. And then layer on top of this, of course, are the workflows—the things that bespoke to your vertical. These are the things you have to really. Have be guided by subject matter experts in the. Later end, and then the kind of through line through all of this.

Is is the evaluation? It's like how do you know that your system is good? How your system is improving over time, right? So evaluations I'll come back to, but it's something that. You have to make a fundamental build.

So, so these are all this like generic, right? This this could. I could have pulled this and said this was a talk about insurance. For all we know. And the same would apply, but where do we go one level deeper on the things that matter?

### 00:10:23 — Data, memory and workspace

For the sales funnels for go to market, so let's unpack some of these building blocks. We spend a lot of time building out our custom vertical-specific data models or data systems. And so, in context of Monaco, what does this mean for us? We have an internal data system that has some 30 million company records, around 500 million people records. The information that's captured here includes things like employment history, title, right in case of people.

Company people relationships things that we know about companies, of course. But this so far is just static information, right? This is largely something you can either get from data vendors. Or just by stacking them from the web, where they say it's more interesting for the economics of what we're building for our customers is are these. Very dynamic, very scalable data merging pipelines we built for our customers.

For any given customer, go-to-market is an evolving concept. It's an evolving motion, right? So we are able to enrich our customers' go-to-market motions based on what we know about how they're actually trying to sell. We do this iteratively all the time, and what this means in practice is that at any given point in time, the state of The way we build our system is that the state of a customer's sales motion is the most up-to-date it can be in Monaco. We have the most updated version of it, basically.

To give another sense for where we collect this data, right—the data you can't just get from the web. In a typical org, we might have thousands of meetings, sales meetings that are conducted, that are transcribed, that are kept within the system. And perhaps hundreds of thousands, in some cases, of emails in the organization. These are all exchanges about. Potential deal prospects, potential customers, converted customers, and the kind of the gap in the middle, which is oftentimes very wide.

There is a kind of the workspace piece that I showed in the diagram a few slides back, which is effectively just an agent execution environment. At its core, you have an agent sandbox. This includes a workspace with, you know, in our case. Short-lived file systems that are attached to micro VMs, and there are a slew of providers you can use to do this. Of course, as an Asian does its work, this information persisted, and we have long-term persistent stores of Asian work.

And then the piece that has made more interesting is the stuff kind of in this the custom NLP the tool access. To the world model that we described in the last slide. So all the data that we're collecting over time needs to be. Interoperable, something that's made available to the agents as they're doing their work, right? So we have, of course, we roll our own custom MCP.

This allows us to not just query from this world data model that we've created, but also integrate third-party connectors if. Someone comes in and says, "Hey, we use this very bespoke piece of." Ticketing software, or this very bespoke piece of documentation software that is in Notion or something like that. So our customer expertise is able to dynamically do that and evolve as customers' needs also evolve. I think this is one of the most strategic pieces that any AAA company needs. To build out, and in some sense, it's the most.

Where there's more of an art than a science in a lot of situations. So in the case of Monaco, what we built out is a memory system, right? It's a system that evolves and captures information about. How an organization evolves over time and how user needs evolve over time at the very top. All his memory is seated in the case of Monaco.

With things like email transcripts and user preferences, when an organization first gets onboarded. This is at user level, and then the organizational level. Of course, we provide things like business context. Onboarding, you know, onboarding calls that happen with customers, things like their ICP definition. So all this is just the seed of this memory system, right?

That is at the center of the diagram. Two different varieties of memory, and this memory system informs the operations within the modular platform. Right. So we're trying to leverage these things we learned at the onboarding stage to inform. The things that we do from a workhorse perspective. But the really important piece is this kind of thing to the left here, which is yes.

We see this memory. The memory really informs how the system is working with Monaco. What happens as things change over time, as they inevitably do? This is where we developed a concept called reflections, though. You may have heard similar things, and other people might call these things dreaming or sleeping or something like that.

We call it reflections. These are this effectively system we built to as actions and interactions happen in the Monaco platform. We pull this information. We create candidates for what is something that we want the system. Either that we learn about the user, whether we learn about the organization, and then how do we?

Promote those things in the memory system of the Monaco platform to then inform future interactions with our system. And so it's this constantly dynamic bit of infrastructure that evolves.

### 00:15:00 — Workflows from experts

Pretty much on a daily basis. So, so the workflows piece is where we really start to get into this like subject matter expert. What is specific? What is a specific go-to-market workflow that we want to capture? And I'll share a few examples just to contextualize what this means for sales.

What is a world-class developer workflow? We have two different modes of. Workflows. So one class is what we call it the Monaco agents. These are systems that are, agentic systems. That are very opinionated.

They are optimized to our harness. They are optimized consistently built. They are where our world bests the market. Encodes our world-class global market knowledge and makes that all available to customers. So this might include things like the best way to do meeting preparation, the best way to coach someone on sales tactics.

The best way to do pipeline reviews, demand generation. So these are, of course, things that we believe are generic pieces of information that you can do well, or maybe not so well. And we have been very opinionated about those things in the context of our platform. As I mentioned, they are all based on monoliths. To do these workflows in TensorFlow, it would be slower.

It would be more costly. So we agreed that we wanted to make this. Important critical piece. This is mode one. The second mode, though, is hey, we don't know.

We can't possibly, with a team of other people, know all the ways that. Every vertical you know that we're trying to sell into does the best sales. So we do allow kind of bring our own workflow option as well. These are custom agents, right? So this is where we've effectively released an agent builder to our our customers.

And we say, hey, you know better than we do how to sell to dentists in the Midwest. Let's say, and so you bring your workflow. And examples of what falls out of this are particular systems like, hey. We want to find warm intros to reach out to, based on. Deal with prospect research, and then.

Surface some of those requests via LinkedIn. Again, this is very specific, potentially to to a vertical. Or we want to do sales scoring. I'm doing the sales scoring here. Are things like hey if call transcripts became available because when our customers had a call with a potential prospect.

Afterwards, they had built their own agent that would score the quality of that transcript and said, "Hey." These are the skills that you, as a sales rep, should probably work on here next calls. And so this coaching is both done, ranked, and then provided as information. Valuations near and dear to my heart. As a core researcher, you need to know how your system is at the core, right? Within Monaco, we have.

Over 100 agent tasks across the full sales life cycle. Right, and this includes things like TAM scoring, demand generation. Full pipeline management, and then a slew of other things.

### 00:17:41 — Tracing and evaluation

At the core of getting emails right, you have to. Trace things, and so a huge shout out to to folks like Brain Trust and General Labs that we work with. Traces are the oil, right? If you're in the business of early AI companies, you should be tracing today if you aren't already. And you should be treating those things literally like gold because they are gold.

Beyond that, our evaluation system is informed by experts like our internal experts, you know, our blockbuster marketing team. Inform how we determine what a good output is for all these tasks. So. The the blue line, the principle I want you guys to walk with. Zain correctness is always contextual. In the case of virtual AI, there is no.

Absolute correctness is only correctness informed by subject matter experts. So we are very deliberate about making that something that we bake into the system. Let's go a little bit deeper into some of the case studies on on what this looks like in practice. So again, I'll share a few examples of. How we built on top of our Google Market Harness.

And build very specific lower market tasks, and what is the outcome of doing this sort of optimized agentic development?

### 00:18:47 — Case studies

So a few agents in the wild, and maybe you guys have some ideas around things you should should build your own. Good market workloads, if you can. One very successful one is reengagement agent. To contextualize the problem, something I'm sure many folks in the room that you've done kind of sales in the past can resonate with. You reach out to someone you thought was a new customer.

They ghost you. What do we do now? Right. It turns out that you should absolutely reach out to them to them again, and there is a strategy with which to do that very well. Times when we should be doing that. There's ways to frame the messaging to them to get them to actually respond.

And so we built a lot of those those pieces of information into these agents that are just really good at reengaging existing. Prospects, right? So you have this warm lead that would hold. High engagement again. So we've done that, and again, outcome of this is that we're able to get really, really high reply rates. So on these warm warm leads, we can get up towards a 30 plus percent reply rate.

Which, compared to a typical cold star, is around 10 s the average. There's another category of agents that are also very high converting. Again. Strategies that Netflix may use in the future. I'm just confident the problem is that to summarize. Is you have just.

One a certain customer. This is your close one, and now based on this close one, you want to find other promising accounts. They should then also reach out to the highly targeted campaigns. Imagine you sell Uber. You just close Uber.

Yay, wins. And you say now we should go target D. Or I should go target D. Or you have something like that. So again, this is an incredibly effective strategy. And we get a lot of the expertise on how to do this effectively.

Into a local agent, as we call it. Compared to the two to three percent in best case scenario for really cold outbound. And you anecdotally, this is such an incredibly powerful strategy. We've had customers close literally $15 million deals based on just. The kind of reaching out is something new.

So very highly effective strategy. One that we picked in, among other systems, that we picked in, based on our Jeff Fusion harness.

### 00:20:57 — Culture: everyone builds

All right. I'll step back from the tech pieces a little bit to share some. More high-level strategies around product building give a little bit of assistance. So these are a few principles that have worked for us, and I want to split them into two categories. One is. Ten things you can do to set yourself up for success. And second, more cultural things that you do to say result in success.

So in the tech category, you know, we are in the business of trying to bridge the gap between subject matter experts. In our case, those they're not always the most technical people with highly technical product and engineering concerns, right? So how do we bridge that gap? Make the translation of of good market expertise into product scenes. So on the technical side, we do really strongly support the parallel processing.

There are never been places where else can. We have a passion to our commitment to our technology, which includes. That's a piece of system that they use, which our agents also use, of course. So there's no distributional shift between agent. MCP versus MCP that they use.

We have a Slack bot that is again shared by all team members. So the Slackbot not only operates as an SRE software engineering team. But also operates as a customer success manager, as a user marketing team, personalized events who are trying to build out the product. In the case of that feature, skills that are shared. And then you know we've made product observability the traces actually also personalized since.

What I mean here is that everyone in the Mark team can go into Bright Trust and see exactly what's happening in the system. So it gives them a really deep understanding of what exactly our agents doing, how they can actually improve these agents over time. On the strategic piece, there's a few structural things that I think are really important for. For companies to get right, if they're going to try and bridge that gap between technical experts and non-technical experts. So we have done this, and we you know, have only done this a system that we call Monaco for Monaco.

Monaco consumes Monaco. Monaco uses Monaco, right? So we are the first. Customers of our own system, and because of that, we're able to. Dog food. Everything that goes right, everything that goes wrong.

Right. So we have made this pretty much from day one. We've made this a part of how we operate. We've gone one step even beyond that, which is we embed domain experts into the product board. Okay, so we really have a head of Asian deployment. This head of agent deployment is not a product person.

This is not an engineer. This is literally the best salesperson at Monaco. That became a part of product war. So this is something where now we have literally a direct line of sight into exactly what best. Salespeople think, and what we need to build to the product.

We also supplement this with bi-directional enablement sessions. What I mean here is. Every time we are designing our new product, the product we need to build. We will allow. We will actually have.

Salespeople in the room with us, and they have equal votes with any junior partners in the room on what should be on their ballots. So this is the one direction. The other direction is that once we actually are sharing something with the team members. They are the first to know. The first to hear on Slack.

They're the first to actually use the system. So it goes in both directions. And finally, you know, I think part of bridging the gap. And this is a bit of a soft cultural point: is you really want to make. The process of building a vertical AI company or a vertical AI product.

Fun thing, right? So especially when you're when you're bridging together folks that come from different ends of the spectrum in terms of their their expertise. We try to make this fun. It's something we should be culturally all aligned with. And so, here are just some pictures of.

We have periodically we'll have hackathons where we really bring together. The salespeople and the tech folks. And we'll just say, hey, let's let's build some stuff in in the product. This is a picture from one of our last ones. Super super competitive.

As you can see, people hiding on screens. I was telling them I was just taking them on pictures for presentations. They didn't even like that. I thought I was cheating. You might see in the upper left corner here.

That was the Manhattan Project. They won a bunch of prizes. Hackathon. They take it very very seriously. You know, jokes aside, there's there's a huge value in bringing together the folks in the same room, having them share their pain points, share what they want to build. Really, making a translation as as quickly as possible, and we try to make that a part of how we operate.

To conclude, I'll I'll end these thoughts.

### 00:25:21 — Vertical AI as an opportunity

I believe strongly that vertical AI is one of the largest economic opportunities in history. And I think there's lots of questions around the defensibility of frontier labs. Especially when you're building these general intelligence layers, and I'm so strongly in the belief that the one thing is defenseless vertical AI. So, if you are wondering whether you say horizontal or vertical, I mean. Live your life how you want to live it, but I'm I'm strong in the opinion that there's so much opportunity and actually.

Capturing workflows that are specific to each domain. What we've done to make this work for us is just a lot of manual customization, both within the harness. Models and really systems and tools we built. To make workflows that are industry specific a part of our our platform. And finally, just you know, you know, do everything you can from a cultural perspective in your company, just foster.

Team level knowledge sharing, and I think beyond that, not just knowledge sharing but empathy. Right, it's not just about. Putting people in a room and saying build a thing, but actually. Understand the person across the table, and get them to really. You empathize with what they're trying to go through, what they're trying to build, and how to do that most effectively.

I think that's. The rest of the things pass. Thanks very much. I think I have time for thirty seconds for a question or two.

### 00:26:46 — Questions

So, I don't know how fast the question is. Okay, I'm losing time. Questions, so many. Oh, oh! I see. I see. Yeah.

Yeah, not not as much space. I think it's something that we've talked about over the couple months, months of quarters. I would say we're a lot more focused on sales than we were. Yes, that could be a whole half-hour conversation at least. The question around reflections: How do we do this effectively?

There is different. There's insertion points on the platform based on activity patterns on the platform. A few examples could be demand-related activities. A just converted a meeting. A few of them could be.

And you just close the deal. At this point, these are very trigger events that indicate, hey, something happened in the platform that we should learn from. At this point, we effectively have like. Agent systems as well, and we'll go and read all the context around the thing that happened. Write that. Essentially, to a store, and then these become like a candidate reflections.

That we then, on a periodic basis, will then promote into the main, the main nervous system that becomes. Now available to everyone on the platform. So it's it's kind of two-step process of insertion points. Happens, happens. Reflect on things that happened, and then. Rank the candidates of things that have happened, and then promote them in memory.

Yes. How do you organize the organization to build these specialized work? Like, are you doing like squats? Specializing one single workflow, or. A single person, one single. Yeah, great question.

So, how do we organize the squads to be suited to these workloads? On the on the EPD side, it is based. On on catalyst workloads. So there's on the engineering side, there's like people that are focused on demand, focused people that are focused on pipeline. Etc. On the on the sales side, it's actually more general.

So we have you know experience. So I'd say that's kind of the separation on the sales. Kind of the sales role side, but functionally they all can contribute to everything they're responsible for. And are they difficult? Do they have to be technical to be able to work?

Even salespeople? No, not really. Yeah, I mean. Yeah, no, we—they're not. The extent to which they're technical is they—they like using. Or ChatGPT. So, so, but even that's not a strict requirement.

Yeah. Thank you.