Organizational Design for the AI-Native Era

Julia Nimchinski:

And next up, Mary Shea joins us to lead the CEO Roundtable. Mary is the co-founder and CGO at Meerkat, formerly Mediafly, Outreach and Forester, and we’ll be talking about organizational design for the AI native era. Super excited for this one. Mary, take it away.

Mary Shea:

Thanks, Julia. It’s great to be here again, and thank you everyone for joining us. I can’t imagine a better topic that’s a fit for what I’m thinking about right now, as well as what Meerkat and Meerkat OS does. So, let me kick it off, and then we’ll get to know, all of the panelists here and dive into great conversation.

So, you know, as you think about organizational design for the native AI era, you know, AI is reshaping everything, from coordination to execution to. org charts, and so we’re here today to talk about how the org chart is changing, and what the implications are for workflow.

You’ve got companies like Block, which are replacing management layers, with AI native models built around builders and operators. You’ve got, DRIs. Does anyone know what that means? I had never actually heard that acronym before, so Directly Responsible Individuals, so… I like to say the AI boss, like, somebody’s got to be accountable, right?

You’ve got player coaches, and then you’ve got this thing that we’re now referring to as the world model. And this is where Meerkat OS sits. It’s the intelligent layer that replaces the information routing with context, visibility, and machine-assisted coordination. And so.

What we’re gonna talk about today is how CEOs can start to think about this redesign of the structure, what leadership looks like, and, how we can leverage these new tools that are going to dramatically increase our ability to scale, at speed. And so, you know, kind of with that.

I would love to start, with an introduction of our panelists, and… have them talk a little bit about, their companies and, really the angle that, ties into, into the topic. So, let’s, let’s go sort of clockwise, and David, Teggurit.

David Knight:

Hello everybody, good morning and good afternoon, wherever you are. I’m David Knight, I’m co-founder and CEO at Avara, and our view of the world is that lots of folks are using AI to automate the tasks that people do, particularly in go-to-market, which is where we’re focused, whether that be the research, or the follow-up, or the outreach.

But our view is that companies’ unique competitive advantage is actually not in the what, but the how, their unique playbook. How do you get through this first 10 seconds of a cold call, or do good. second-level discovery in a first meeting, or defuse an unhappy customer, or hold value in a price negotiation.

All of that sort of, unique behavioral expertise is really what separates the winners from the losers in your ability to scale a company quickly. And that’s not written down anywhere, it’s not in your systems anywhere, it’s in the heads of a small number of leaders and product experts.

And so, Avara is a platform that captures and scales that expertise, because it’s very hard to scale that, it’s not written down anywhere, you only learn it through practice. And it’s constantly evolving. And that’s a problem that I faced when I scaled first WebEx and then Proofpoint from about 50 million to a billion plus.

And it’s a decades-old problem, but to the topic of this conversation, it’s about to get a whole lot harder, because that expertise actually now needs to be scaled across a blended human and avatar AI workforce. And so that’s really the platform we’re building here at Avar.

Mary Shea:

Yeah, love that. Thank you so much. Great introduction, David. Christopher, off to you, please.

Christopher O’Donnell:

Hey everybody, welcome. Excited to be here with these wonderful colleagues here on the panel. My name is Christopher O’Donnell. I’ve led product at HubSpot for about 10 years, building the marketing hub, then the sales hub, and then later the service and data hubs, of that product. That we know and love today.

I think the best way to think about day AI, which dovetails really, really well with what we’re all, I think, talking about today, is… Everything is now recorded. This meeting, is being recorded, all the content today. Many of the internal conversations that we have.

There is this new magical data asset available to all of us, which is the actual transcript. How do we turn that transcript into action? If a customer mentions a bug over Slack, or if a prospect has an objection on a sales call that the rep doesn’t handle well?

How do we turn that into a bug fix in production, training for that sales rep, or enablement for the rest of the team? That is what Day.ai is doing. It’s, in many ways, what comes after CRM. I think, you know, the nature of this data is different enough that the world will change, not overnight.

First, it plays well with CRM and our friends across the river here in Boston and Cambridge, and eventually, you know, turns into something pretty different over time. So, excited for the conversation today.

Mary Shea:

I like that, and I can’t hold myself back. I know this is an introduction section, so I’ll try to be short, but can we all agree, then, there is life after CRM? It’s amazing, yeah, it’s incredible to see what’s happening right now.

And one of the things that I’ve been finding, you know, with the transcripts that I use from Meerkat, is I… I’m selling a lot as a founder, because that’s what we do, right? We all have to sell it. But I have never been as strategic and targeted in my life as a salesperson, as I am right now. And my hit rate is off the charts.

It’s like nothing I’ve ever experienced, so I’m looking forward to digging in a little bit more to some of those topics, which are really exciting to us all. But, it’s not about me today, so Amos, take over, please.

Amos Bar Joseph:

Thank you. Yeah, so Amos, co-founder, CEO of Swan AI. We’re building the first AI go-to-market engineer, so something between a developer and RevOps that can turn any go-to-market process into an agentic workflow in seconds from prompt. To pipeline, so you could scale that process with intelligence, not headcount.

I wasn’t invited for this panel because of what we do. I wasn’t invited because of the way that we’re building our company, so we’re trying to build the company very different than most of the companies out there. We’re trying to really figure out what does it mean to scale with intelligence, not headcount, as a company from day one.

In 2025, just the three founders, without hiring anyone on the team, we grew from… Over 0 to over 200 customers, over 5 continents, 7 figures ARR, just us and a bunch of AI agents. Today, we doubled the team to 6.

So we’re 6 people, with the goal to get to, eventually to $10 million ARR per employee, and redefine what does it mean to build an autonomous business at the age of AI.

Mary Shea:

Yeah, it’s fantastic. Going back to one of my favorite movies of the 90s, I’ll Have What He’s Having. So we’re doing the same thing, so I look forward to digging in, as, as well. And then, let’s talk to, to Tuba for a minute.

in reading your background, I’m gonna go out on a limb here and say, we’ve got some very, very successful entrepreneurs here, people who’ve built businesses and are really smart people, but in looking at your background, I’m just gonna go out on a limb and say you’re the smartest guy in the room here, so, Tell us a little bit about your company and what you do.

It’s a bit esoteric, but I look forward to learning a lot from you today.

Tooba Durraze:

Yeah, nice, nice to be here, nice to see everyone here. I’m not the smartest person in the room, I just really love math, is what I like to.

Mary Shea:

I don’t know.

Tooba Durraze:

But, yeah, I’m Tuba, founder and CEO of Amoeba AI. By background, I am an AI PhD data scientist, who has been kind of enamored with the world of AI and all these things prior to, kind of, the craze that we’re in right now, essentially. So, Amoeba is essentially looking… is looking at, decision intelligence.

So, all of us were talking about, kind of, you know, creating a better understanding for your system so you can leverage intelligence. And context comes as a part of that, like, obviously, system of records come as a part of that.

We are a layer that’s above that, which is your system of reasoning on top of your system of understanding, essentially. So, the problem statement is very simple at any level, whether you’re kind of, like, at the C-suite, mill management, IC level.

Your day-to-day is about making decisions, and what decisions you focus on, what decisions you make, how you make them, all of that is dependent, obviously, first of all, in how much of an understanding there is of your system, but then there’s, like, a higher-grade compute problem on, like, okay, from all the understanding, what is the stuff that matters?

And what happened is, because human nature is, like, kind of, lowest hanging fruit, it’s like, we also exasperated the problem by then, like, going in our holes and saying, okay, well, now I can do all this, so I’m gonna do so much more.

So we’re the opposite side of the problem, where, like, we are trying to focus people towards, like, what are the core things that you need to do, and when, in order to get to your objective.

And that’s what neurosymolic AI systems are really good at, because they, are not based on things like pattern matching, so not kind of next best action, but basically building a understanding using a liquid neural net of your business against the objective that you’re trying to drive.

And then hopefully your world becomes as simple as, these are the three things I need to do today, and kind of moving on from that. So that’s kind of the problem space we’re trying to solve, and very apropos, because we are extremely plugged into organizational design in general, because if you think of those layers, it’s almost like an org chart.

So, for us, also, we’re… interestingly observing how organizational design now is shifting based on, kind of, some of the stuff that’s happening in the market as well. So, excited to dig into it today.

Mary Shea:

Yeah, me too, and thank you so much for that background. Look forward to digging in as we progress our conversation here, but I want to go back, I want to go to Amos for a minute, because you really got my attention, when you, I read that you’re looking to build, you know, sort of 10 million ARR per employee, like, that is not trivial.

I’m just going back on my career as a… when I was a CRO many years ago, and, you know, if we could extract a million, 2 million per salesperson, like, you were killing it.

And so, I really want the audience, and myself personally, selfishly, to learn a little bit more about how you’re thinking about getting to 10 million ARR per employee and agents, and, you know, maybe start by describing, not at the theoretical level, but in reality. What does your org look today, both from a human and agent perspective?

Amos Bar Joseph:

Yeah, would love to give the audience a peek. First of all, about the metric, I feel like… 15 years ago, when people set out to build a unicorn, it was, like, impossible. Like, what? You’re going to build a unicorn company? A $1 billion valuation? Now, companies hit that after, like, 8 months since inception, right? It’s not really, inspiring.

And I feel like the playbook should shift from optimizing for valuation for investors to actually, getting to this North Star, this unbelievable metric, which is $10 million ZRR per employee. And the reason why I think it creates a healthy culture today is because it optimizes us for leverage, rather than just inflation of the stock price. Right?

Mary Shea:

Right.

Amos Bar Joseph:

And, when you focus on that, another thing that happens, you start focusing on the human, not on the AI. The human starts being at the center, and that’s a good segue to our philosophy of how we’re building our autonomous business.

We call it autonomous business not because we have autonomous AI that runs in the background, it’s because our employees are autonomous, because we try to scale each employee on the team so they can get to that $10 million ARR per employee threshold, which sounds impossible, so how do you do that?

You try to build a system around each employee at the team. Each person should be, an IC, an individual contributor, and an executive, both at the same time. They should be able to lead a goal, and should build a plan to get there, but then should be able to execute, working with AI agents hand-in-hand to actually do the job.

And the reason why there’s this DRI, or all these acronyms floating around, is because every human being on the planet should be able to work with agents. It’s like working with Excel, 20 years ago, and working with agents is the equivalent of going back to being an IC right now.

Mary Shea:

Yeah.

Amos Bar Joseph:

And that’s the job, basically. And if you’re sitting at the ivory tower, you know, still waiting to just give the people their job to do, and not knowing how to interact with that technology, then you’re not really a fit for the next, evolution, basically. Yeah, go ahead.

Mary Shea:

Can I jump in for a sec? Yeah, I just love what you said, and I think that’s why, like, I’ve found so much joy in the work that I’m doing right now.

You know, 30 years into it, I’m going back to being an IC, but it’s like an IC on… steroids, like, you know, I’m working 3 days a week, I go to bed, I set my agent up before I go to bed, the first thing I want to do is now look at the work my agent produced when I wake up in the morning. And it’s just really fun.

So, I think that’s really an interesting piece of it, because when you think about these topics, I think there’s a lot of anxiety and fear on the employee side. What are your thoughts?

Amos Bar Joseph:

Yeah, so I do think that… the AI’s worst side of this revolution is the marketing side of AI. That’s the, like, the technology is amazing, but the marketing narrative is so terrible. Everyone is trying to, sell more by injecting fear into the buyer, injecting fear into the market.

Anthropic is doing it, everyone is trying to say, we’re gonna build an AI SDR, we’re gonna replace your SDRs, we’re gonna build an AI model that will replace your entire workforce, and that creates this wrong impression of this amazing technology. And if you ask the people that are really embedded in there, they’re working 10x harder.

They’re not managing to replace themselves out of it at all.

Mary Shea:

Right.

Amos Bar Joseph:

much harder, because what they can achieve with this technology, all of a sudden, it’s unbelievable. Yeah, exactly.

Mary Shea:

Yeah, I think that that’s right, which is… I’m also finding that I have, like, immense flexibility in my work life, but, like, I can’t wait to get back to work, because the results are just… astronomical, so I think that’s interesting.

But I do want to go back to, and pin you down a little bit, because I don’t think you totally answered my question, although I loved your, your talk track there.

Tell us exactly what your org chart looks like, like, you know, virtually walk us through that, and then let us know if there were any traditional roles that you’re consciously deciding to, like, never hire for.

Amos Bar Joseph:

Yeah, basically, we only have… R&D and go-to-market, and everyone is kind of doing everything, in the sense, we don’t have any different roles within different departments, and it actually tells a lot about the future org, because you don’t have product and then developers, because every developer needs to develop Product capabilities, and every product person needs to develop engineering capabilities, so we don’t call them that way, because it creates a confusion, which this person needs to know the product, and this person needs to understand the architecture.

No, they’re… basically, they need to know the same things, and should be able to work together. So that’s like a different perspective of looking at actually how R&D operates in an autonomous business. And then on the go-to-market side. We don’t really have CS, and we don’t really have support.

We don’t think that these roles live in an AI-native world, basically. The reason why is that, if you look at AI-native companies that are selling AI solutions. A lot of the value could accrue after the sale, actually.

And so you can start small, and then, like, expand and expand and expand and expand and expand to infinite, basically, because AI could just continue to do more and more work for the company. And so the majority of the revenue creation is actually post-sale, not pre.

And what happens is that creating this division between sales should, you know, look for how to sell the product, and then success should make sure that there’s a success here, is totally wrong.

And it actually should be one person who’s in charge of the entire revenue life cycle, who can be with the customer and understand the change management required, the politics, the buyers, the dynamics. and kind of, like, steward it end-to-end. What’s more interesting, we don’t even have solution engineers at the GTM. Why?

We think that the seller should be able to do that work today with AI. You don’t need a technical mediator right now to actually tell the buyer how to implement it. The seller should possess that skill as well.

So you kill the CS, you kill the sales engineer, you kill the support, because AI could answer knowledge-based questions, and if it’s technical, you can just route it to engineering, and then you get, like, this 100X seller persona.

which can be in charge of the entire revenue life cycle, and really contribute to the ARR growth at the company at unprecedented manners. So I think if I try to, like, condense it into, like, 90 seconds, how we’re looking at the org chart differently, I would say that, you know, roles are turning into generalists.

doing the IC work, but also the executive work, and are expected to kind of, look at the life cycle, whether it’s in R&D and go-to-market from end to end.

Mary Shea:

Yeah, so, thank you, and that is a 10 out of 10 for, answering the question, so you nailed the, you nailed the landing on that. Thank you so much. So I think what you’re saying, you know, to kind of summarize. and we’ve been… I’ve been talking about this for a while, which is there’s really no need for sales specialization anymore.

You know, you’ve got to handle everything. With agents for different areas of the specialization over the cycle. So, yeah, thank you for that.

Tooba Durraze:

No, no, interesting.

Mary Shea:

Yeah, please, please jump in.

Tooba Durraze:

The scientist in me is like, we… when we talk about these things, sometimes someone’s going to pick up a generalization that we’re making and say, okay, I need to get rid of my entire CS team, as an example.

Mary Shea:

Right.

Tooba Durraze:

There’s a lot of nuance in… like, unless you’re at an advantage for having started your organization in a very AI-native way, but for existing organizations. there’s a lot before kind of getting to that point.

So I… my TLDR is, like, everything, the caveat is basically dependent on, kind of, how you do your business, how you run your business, etc, because.

David Knight:

Very important to your customers.

Tooba Durraze:

side.

David Knight:

Your customer may want to talk to a human who can provide accountability.

Tooba Durraze:

I agree. On the enterprise side, we’re seeing a heavier, like, focus on very, like, not generalists, very specialized ICs.

Mary Shea:

Interesting.

Tooba Durraze:

Organization even becoming… even more… even more narrow. Now, you know, again, like, most has an advantage for starting in this era, and, like, starting with, obviously. advantages, both the mix of market and your brain, in how you design your organization.

So, not discounting that, but, I think people should think about it in the context of, kind of, their own, organization.

Mary Shea:

their own businesses, yeah. I think that’s a very good asterisk to put on that, so thank you for that, and I’ve been thinking a lot about it as well, because we don’t have CS at our organization yet, and I’m starting to think there is going to be a moment in time when we will need that, right? But it’s… Maybe not 15 of them, as an example, so…

David Knight:

So, as a practical example, we have a… our AI avatar customer service manager embedded in our application, so our customers can talk to a fully trained AI CSM, so they don’t have to ever call a support desk, because, well, why would you do that if you can deploy a fully expert resource 24×7 on site.

And so, that’s what we do, and… but what we’re finding is that They’re talking to that individual much more than they would have talked to a human, because it’s always there.

Mary Shea:

And it doesn’.

David Knight:

replace the need for a CSM, because they still want that human connection, the person who’s accountable, the person who they’ve built a relationship with, who they can double-check what is happening, etc. So, so it’s going to be a mix, and it also varies dramatically by customer.

Some of our customers are very comfortable talking to that AICSM, and others have no real interest in doing that. And we need to meet the customer where the customer is, and what we want is to ensure that we run the play. We give the same best practice advice, whether that’s a human giving that advice or an AI giving that advice.

Mary Shea:

Yeah, great point. I mean, I do think it is going to be a hybrid type of scenario, and you’ll lean in or leaning out, more or less, based on what your customers want, what your business model is. So, I completely agree. So, well, let’s move on to Christopher, if you don’t mind.

And Christopher, do you mind if we kind of go back to this concept of world model? I got into it a little bit at the outset of the call, but in, you know, plain terms and how you think about it, so that we can all really grasp what we’re talking about here.

Why don’t you just describe that for us a little bit, and then, I’ve got some specific questions about AI relative to that topic.

Christopher O’Donnell:

Absolutely, and on the world model, I’m very curious to hear Tuba’s take as well.

Mary Shea:

Yeah?

Christopher O’Donnell:

Because there are, you know, maybe a couple of different, context in which this is used.

One is the sort of traditional research, you know, science definition of a world model, which, and again, you know, forgive me, hopefully I’ll get close enough to the pin, but some virtual environment That represents reality, so that experiments can be done, and the outcome of those actions can be evaluated.

You know, an environment, so if you are creating a robot, you can figure out if we tweak it this way, or if we change the physics in this way, we can get an idea of what happens.

Very useful, obviously, for a world of frontier model AI and loops, where we are evaluating things and can, in fact, create massive value simply by brute-forcing some of this training, again, to oversimplify. And you mentioned Block and Jack Dorsey talking about the world model of your business. Well, it’s pretty similar.

And in that context, we can even get to some specifics of what this is and how one would attain a world model of the business. And this echoes, really, the nature of the conversation across everybody here, which I’m listening to with delight. This is really interesting stuff, and the CS conversation, everything, I’m totally with you.

So I’m having my popcorn over here and enjoying it. But some environment That has a representation of what’s happening at your company with your customers. It’s kind of hard to imagine a company without the customers.

And to evaluate it, and to understand what are the forces within this company, you know, how are the wires connected inside this dynamic system? And if I put a little more voltage over here, or if I pull this lever, what happens to these inputs and outputs and so forth? So what does that look like?

I think all of us on this call, certainly, and the world at large, the last 6 months has been this coming online of the idea of, okay, if I somehow had captured Every conversation, every… piece of data that matters, every transaction, every, you know, block goes as far as to say all the engineering activity, and had that as some substrate, some raw material to be… read and shaped, you know, and experimented with.

Well, in that context, then a world model would be a lot of fun. Because we’d be able to say, what would happen if we hired, you know, 4 reps a month instead of 3 reps a month? Where would things break? Well, you know, an interesting thing would happen with facilities in EMEA.

A very interesting thing would happen with, you know, training as a bottleneck. Or, what if we could reduce ramp time for go-to-market people, you know, in this way? Oh, that would have this amazing impact on the model. Well, that’s easy, we just need a spreadsheet to figure that out. But how… What keeps us from doing that? Right?

And how would we develop a real plan as a company to do something like that? And that’s why I think the, you know, the foundation models that are out there, and for the few days we had Fable, oh, boy, oh, boy.

Mary Shea:

Oh my god, I know. I got shut down, it was terrible.

Christopher O’Donnell:

Yeah, yeah, this is the altitude that this class of model wants to be working at. You know, if you show it a deal pipeline, just as it’s been traditionally represented, you show that to Fable, Fable will, you know, maybe even say something rude to you about, you know. you bring me this? You know, did you want me to add these numbers together?

Did you want me to make you a bar chart? Like, what are we talking about here? I mean, this new mythos class of models, and there will be more. This is a one-way street, baby. You know, we’re not going back in time. Those models want existential company-level issues. Like…

Mary Shea:

You know?

Christopher O’Donnell:

You know, where are… engineering and sales not connected in their priorities. This is a huge problem for McKinsey, for the greatest CEOs in the world.

And that is something that, if you had the right world model, you could expose to these, you know, new classes of models, and they would do… they would come to some very interesting conclusions, and maybe even give you a path forward. The way we fit into that is, you know, maybe pretty obvious from how I’m setting it up here.

You know, you need to store that and have that set of data somewhere. It needs to be stored, you know, not in a way that is limited by what humans can enter, but limited only by the tokens you spend and the raw material that you’re starting with.

And to be stored in a way that is accessible to AI agents that are not… limited by doing a SQL query and, you know, not bothered by any of the things that we’re bothered by, learning curves and so forth. They have their own learning curve, they have instructions, right? So you need to store it, and then you need to be able to interact with it.

Ask questions, human in the loop, and so forth. Get all of the right kind of data out of it. And, you know, we have a particular kind of target user, and so, you know, email drafting and seeing charts and that kind of thing is particularly smooth and optimized for off-the-shelf.

And then asynchronously, you know, what behaviors need to be done automatically in the background on a standard set of instructions? And what is the, you know, what Amos is pushing at here, I think, is really very critical.

to expect more of our employees and our team members, what is the form factor of how we deploy this agentic AI such that each of these humans is responsible now for more? And the DRI concept, and so forth.

You know, if you have a bunch of humans, and then you just deploy some AI cloud, or some… app that you give them from operations and they chat with. You know, it’s very hard to, raise expectations unless there’s real ownership. of that AI deployment on an individual level.

And the paradox there is, you know, we are not yet, as individual contributors, ready to stand up a whole AI strategy for our own practice.

Mary Shea:

Right.

Christopher O’Donnell:

So getting both of those things right, but, you know, if you look at, for example, we use linear to track engineering work. And Linear has a really beautiful new feature set that they’re pushing hard, rightly so, which allows the association of a particular customer mention or request to a feature or a bug.

And so this is beautiful for product people, because now we can see the work that we’re considering doing, or that’s in progress, and we can see you know, the logos and the revenue and everything associated with this work. This is wonderful. Well, so where are you gonna get this data?

A good illustration of how, you know, the purpose of our product is our linear has everything that anybody has ever mentioned associated with all of the work that’s in Linear. If there’s a bug that we don’t know about, it gets added.

If there’s a roadmap item that, you know, is in there and we’re considering, and somebody mentions it on a sales call, that moment and that company and that deal are all associated with that feature request. And, you know, I think this is representative of where we’re going.

We will still be doing this work, and our editorial and taste responsibilities will be magnified, but the responsibilities, in terms of work output, will be magnified to, you know, sort of the overture of this conversation so far, which I think is exactly right. If we’re not able to do that, then kind of what’s the point?

You know, we will still have a huge role in this, and each of us is… ought to be able to do a lot more. And to a point that was mentioned very briefly in passing, but very important. you know, in coding, we were able to do 5 times as much in maybe 150% of the time.

And I’d really like to see us be able to do 3 times as much in, you know, maybe 80% of the time, and that’s kind of where we’re headed.

Mary Shea:

Got it. Thank you for that, and tuba, I… Chris, Christopher, I didn’t ask for you to weigh in on the world model. Do you want to share, you know, your perspective?

Tooba Durraze:

Yeah, Amos, I don’t know if you wanted to go first, your hand was up.

Mary Shea:

Oh, I’m sorry, I didn’t see it, Namos. Go jump in, please.

Amos Bar Joseph:

Cuba, you’re such a sport, I appreciate that.

Yeah, so I just, you know, every time there’s a new model, new, terminology that gets tossed around, I love just breaking it down, because sometimes it creates this, feeling for, the folks out there that they’re kind of, like, passive, they’re waiting for it to happen for their company, and they’re just like, oh, I can’t wait for a world model to come to my company and solve all my problems, and then they’re just like, okay, but what do I do today, right?

And I feel like There’s two approaches. two different approaches to a world model, and when I read Jack Dorsey’s, materials, and when I look at my own company, I actually look at a world model a little bit different. And the best analogy for us here at GTM is actually to think of enablement. And if you look at… enablement documentation.

That’s the closest thing that you can get to a world model, basically. What does that mean? If you’re trying to get your reps not to hallucinate And to align them around processes, what’s your tool to do that?

You create enablement docs, you describe your ICP, you describe how do you run a cold call, you understand objection handling, battle cards, all of that documentation. And then it just sits in a Google Drive that no one ever reads, and the hallucination happens right after that, right?

So, what happens today in a world model that you can create for your companies, for the first time in history, you can take that enablement docs and just let AI read from it. And then someone can just ask the AI a question. what’s our ICP? And the AI could tell it, and then someone can just ask a question.

Someone told me that, you know, the competitive is doing better than us. What should I tell them? The world model can go into the enablement docs and just answer you the question. And I think a better way to look at world models for company building is actually to think about alignment.

Not to think about how AI could solve my biggest challenges, but how can, you know, me as a CEO, how do I create alignment downstream? If I am… the sales leader, and I’m seeing things, and no one else sees, how do I create alignment around the stuff that I see?

If I am, you know, the product, and I want to create alignment around the product roadmap, how do I create that?

We have a world model for it, and everybody could just ask the world model, or even more so, that world model could even proactively tell someone at that moment, this is what you should do, this is what you should focus, this is what you should look at.

But putting the weight on the AI to think for us, and to just say, yeah, this is where we should go, this is how we should solve X, I don’t think is the right approach for world models, and I think the better approach as leaders of our companies is just to think, how do we use this knowledge base that we can create so an AI could communicate on our behalf across the team, so that we can strengthen the alignment moving forward?

Tooba Durraze:

I have gone.

Mary Shea:

it. Okay, okay, good.

Tooba Durraze:

Okay, two things. One, for, like, the broader, kind of, population is listening to this. To Chris’s point, yes, there are, like, technical world models. We’re talking about the concept of world models, like, applied a certain way here. I think, Chris, your definition, some version of Mosa’s definition, is correct.

Context of the world that you’re in, in reference to answering questions. But we’re basically talking about really deep, like, systems of understanding, built on top of, like, data, context, all of that, essentially.

this idea of, you know, like, outsourcing think… we can’t outsource our thinking, et cetera, to AI, and all of that also gets conflated quite a bit in organizations these days. Because, if you look at where we’re trying to go, as far as, like, using computation for what computation is good at. you’re basically… I use the analogy of, like.

being data-driven, like, sitting in a Weibo. It’s like, you’re telling me where to go, right? As a human, I’m setting the objective of where to go, but I am not sitting there figuring out every tiny little turn that the car needs to take to get there. It’s a computation problem.

Mary Shea:

Right.

Tooba Durraze:

Let it figure out, under certain definitions, of, like, I want to go this route versus that. I want to go the scenic route on Highway 1 versus, like, I want to get there as fast as possible, along with, like, kind of where you’re trying to land. those are the two definitions you’re setting, and you’re letting the machine kind of do that.

The problems that we’ve created as humans out of our, like, intellectual curiosity, so we love to tinker, so then we create a bunch of different things to do all the time to tinker. The promise of AI was, like, hopefully you would end up in a beach. We’ve ended up creating an extra 10-hour-plus workdays, essentially, at the moment.

Hopefully in service of the beach at some point, but the problem that we have created is we’re creating more information constantly, more data. Because we’re creating more data, then we are needing more systems to manage understanding of that data.

So then we’ve created, like, this whole system of, like, oh, constantly, like, is this making its way into the context that the model needs to understand? Who’s responsible for updating that context?

It’s like the problem that AI SDRs had sitting on websites where you discovered answers were wrong because you had all these outdated web pages that were… Yeah.

Mary Shea:

Yep.

Tooba Durraze:

The world that we need to be kind of heading towards, and hopefully everything all the participants in this are kind of working towards, is… Like, you’re not offloading cognition, but you’re reducing the cognitive load by offloading computation.

Mary Shea:

Yeah. I love… can you say that again? Because I think that’s a really great phrase for us to remember.

Tooba Durraze:

Yeah, because I think, I think we conflate, like. strategy, taste, etc, which cognitively, like, kind of, I think, belong to humans, with, like, what is a compute problem. So machines are better at finding patterns continuously. Machines are better at looking for, whatever, silver bullets, etc.

Machines are better at, like, distilling through noise. all of that, like, falls flat if you have not set an orientation for the system, which is kind of, what are you trying to achieve? And for most people, the answer should be as simple as you’re trying to make, or in most of this case, like, a $100 million company.

You’re trying to grow your company in the context of, like, what we’re doing right now. So, I think the ideal world is hopefully all of these systems contribute towards kind of getting to that place where, you know. What to do? Like, when to do it.

kind of has the human in the loop, but what to do when to do it is, like, kind of our orientation of, like, what the systems are telling us, instead of, like. us directing the system to say, you have to do all these different things. Tell the system what you’re trying to achieve.

have the system take on the load, the cognitive load off of you, to be like, I’m going to be a living organism that kind of helps you achieve that, and then you’re in the loop as, like, you guiding the system. Whatever the models are in the middle, like, the concept of world models will, like.

that theory will never go away, because in order for you to get there the best way possible, it needs to understand the entire thing, right? So, I think that… that people should think of these things as more kind of, like, less rigid stacks, but more kind of malleable, like, organisms that evolve.

Mary Shea:

Yeah.

Tooba Durraze:

As we’re going.

Mary Shea:

alone. I love that. Really… fabulous, fabulous riffing on the topic. I’m gonna bring us back down to Earth a little bit, or be a little bit more grounded, and Chris, I’d love to go back to you, because this is just something that’s been gnawing away at me.

So, you know, obviously, you were CPO at HubSpot, and you scaled or had a big role in scaling the SaaS go-to-market org. two things. What’s the single biggest assumption from that playbook that is now just dead weight in a native AI company? And then second.

what can we bring forward from that time and continue to keep in an AI-native world where we’re driving In scaling go-to-market.

Christopher O’Donnell:

Yeah, I think, I’m gonna be preaching to the choir, I know, already, given the prior conversation here. the single biggest, to try to answer the question literally, I think it’s productivity per salesperson.

You know, if there’s one thing about traditional SaaS go-to-market that is too, granted in the consensus, compared to how much it’s gonna change in reality, I think it’s that. you know, the productivity per rep, OTE, quota, you know, how much we pay reps and what they can produce. And this will be true, of course.

For other roles, for marketers, it’s already true for engineers. You know, engineers that are doing two and a half pull requests the way that, you know, you would get a promotion every couple of years for doing that for, you know, maybe decades. You know, two and a half pull requests a week right now in engineering is not… employable.

Mary Shea:

Right.

Christopher O’Donnell:

You know, and so, I think for go-to-market folks, this is pretty wonderful for two reasons. One, you know, they get to feed their families and have, you know, maybe some abundance here. And number two, this is done by spending more of their time doing the parts of the job that they.

Mary Shea:

That they love.

Christopher O’Donnell:

That they really love, yeah. We had some reps join just about a month ago, and one of the reps, in her first week, said, looked up, we all sit together in the middle, and then we break out to do focus work. We’re sort of an inverse of a typical office in tech.

And she looked up from her laptop and said, it’s just amazing to not have these two-hour CRM data entry blocks on my calendar. My whole career, I’ve had these admin work blocks throughout the week, and I’m just… I don’t have…

Mary Shea:

Don’t need them.

Christopher O’Donnell:

I’m never gonna have any, you know? Would she like to do twice as many calls, being consultative and using her business acumen to understand somebody’s problems and whether she has something up her sleeve that could help? Of course! Of course!

You know, any sales rep who’s doing, you know, 2 calls a day would be happy to do 4 if they could trade that time.

I don’t know if there’s something to hold onto other than that, really, which is the… and again, this has been echoed here, and so I’ll say it in the context of the kind of thread of conversation that we’re on, because it really is one thread of conversation, which I love. You know, Let’s play out this enablement idea as a rep.

to what Tuba’s saying and what Amos is saying, you know, these loops have a human in them. The… I’m guessing Amos would agree with me, afford me this, and then, you know, hop in and yell at me if I’m wrong, but, you know, the enablement materials, really, in the best case, would come from the facts.

And, you know, a battle card against competitor X would have who knows what’s on it, but it, let’s say, ought to have… a killer insight from maybe the founder, you know, who said something in a meeting about, like, we should never lose to them because of this one thing. Okay.

And maybe that’s a clue, maybe that’s not on the battle card, but maybe that’s an interesting clue that is pursued. Maybe there are half a dozen instances of somebody saying… a version of a particular thing that really took Competitor X definitively off the table. Okay, well, that certainly should be on there.

And then maybe next week, something happens that is input to this battle card, and so the battle card ought to update, and it does, because it is a point in time, it is some sort of… Intermediary, you know, document.

David Knight:

Let me just interrupt.

Christopher O’Donnell:

Go ahead.

David Knight:

Because it’s exactly what Avara does, is help with this. And the problem with the old static battle cards is they get outdated, but they also lose all the nuance. Like, all of these playbooks, they’re incredibly nuanced. Who’s the customer? What’s the situation? What’s the problem they’re trying to solve?

And when you try and write it and distill it down into text, you lose all that nuance, and the why is never written down. The why’s in the head. And the beauty of large language models is they’re really good at dealing with messiness. Like, you can just feed them all of this, stream of consciousness from the people who know.

And they’ll make sense of it. And they’ll coach appropriately, and let people practice. And then the other thing, is that is that the AI can directly have these conversations. But I want to talk about the practice for just a minute.

The other thing that’s important about running these plays is these aren’t plays that you learn by listening, they’re only learns that you… only things that you learn with practice and coaching, and it turns out that the AI is really good at helping you role-play and coach and give you crisp language so you’re better prepared for that meeting.

And so that’s… that’s another area that the tools that… that update the CRM give you the two hours back, AI and coaching and training and game planning actually make those four calls more productive as well, and so it helps you on both sides.

Christopher O’Donnell:

Can I… Just very…

Mary Shea:

I’m going to jump in, I’m just gonna jump in for a second, Chris, because I want to stay on this topic a little bit with David. You know, so, you know, since we are talking about org in this conversation here.

So, the player coaches were placing the manager, and the manager’s job originally was routing, you know, information and, you know, building and developing others, and you know… where do we… where do we go from… from here?

Is… are we looking at teams with… player coaches that have massive amounts of, direct reports, you know, if avatars and AI is delivering the coaching onboarding and product expertise, like, what does this human leader… How many reports does this human leader have, and what do they do?

David Knight:

They lead, actually.

Mary Shea:

Hey!

David Knight:

That’s the thing, is, like, the coach is a better coach than most line managers, but they don’t inspire, they don’t motivate, they don’t build culture. So the attaboy that I get from my real boss is fundamentally different than the attaboy that I get from the coach. And then, take the SDR, for instance.

Like, that’s a really hard job, and I can do everything right, and I can get hung up on 10 times in a row. I need somebody who’s gonna motivate and lead me through that, paint a picture, and that’s gonna be the role of the future leader is to lead, and be a good human, and to inspire, they’re not going to do much management.

And if you’re… if you’ve been managing and tracking and following up, like, you’re not going to be very successful in this new world, and span of control will absolutely increase.

Mary Shea:

Yeah, a great, great, great point.

Amos Bar Joseph:

Chime in for a second.

Mary Shea:

Yeah, I, I do, I do want to ask… One other question, we’ll come back to it, it’s on accountability, but go ahead, Amos, jump in.

Amos Bar Joseph:

Yeah, I think, I’d like to kind of refer to what David said and what Chris said earlier, just to wrap that subject, because… there’s something about the word facts that were thrown here, okay? And I feel like… or also, David said that, you know, AI could be a great coach.

Chris believes that, you know, we could create an amazing enablement from the… from just looking at the facts. I have a different point of view here. It’s not contrasting, necessarily, but it’s a little bit different. I would argue that, you know, GTM is science and art together. Yeah, yeah. I actually believe that it’s more art than science.

If you want to be at number 1 to 10, if you want to be at the, you know, 50th or 100 in your market, then you could lean into more just the science, but to get to the being at the top of the top of the top. You gotta have more art than actual science, and I explained what does that mean? Large language models are trained to predict the next world.

Basically, they are prediction machines. They are generating the most obvious thing that we are expecting from them. That’s actually what they are trained to do by nature, and I think GTM, a lot of the times, is about doing something that is unexpected. And a great.

Mary Shea:

Yup.

Amos Bar Joseph:

coach. could actually coach someone to do the unexpected thing here. And a great coach could understand something that… understands the specific person behind the scenes here, what could, you know, move them towards the right action here.

And so, I’m not saying AI couldn’t be good at coaching, I think that if you look at human-AI collaboration, we could create amazing coaching. And I think when we look at enablement, there’s no such thing as facts in some cases. It’s subjective, and we need to have, like, a human-AI collaboration process to determine these things.

And so, I’m not an AI maximalist. I don’t think that almost nothing should be… we should kind of, like, offload the computational process to the AI in that sense, because I feel like the moment we do that.

we kind of… we lose the most important thing in GTM, which is the human factor, and it’s not We’re not trying to build software here, which you could argue maybe it’s fundamentally different, but GTM is eventually people selling to people, and as long as that’s the case, I do believe that we should lean into the human aspect in that process and use AI as a tool to amplify ourselves, not to, you know, extend ourselves outside.

Mary Shea:

I completely agree, and I think… I think I’ll… I also think.

Tooba Durraze:

One second, sorry.

Mary Shea:

No!

Tooba Durraze:

I feel like we talk about things as, like, oh, like, it’s humans versus machines, or then it’s only machines if humans can lead the machines.

Mary Shea:

Right.

Tooba Durraze:

There’s a genuine computation problem in go-to-market. You’re sitting on top of so much data, you cannot possibly compute it.

Your intuition and all of that stuff kind of coming into the picture can, yeah, help you in terms of, like, your, like, guiding of the system, so yes, the objective, I think, plays a big part, but It’s not a, like, a… human in the loop eventually, like, evolves to, like, just human as, like, a Sherpa.

We got to a point with planes where, you know, you still have pilots in the cockpit, but the co… like, it’s the plane is, like, you know, on autoflight, so it’s, like, on autopilot. I feel like we’ll get to that space. The second really quick point, sorry, Mary, is Okay.

Mary Shea:

Good.

Tooba Durraze:

Talk about AI, like, well, you brought up a good point about, you know, pattern matching, for large language models, but there are other kinds of AI that people are using as well, so we use neurosymbolic AI, which is, like, not on pattern matching, etc.

So, just general for, like, the public listening to this, it’s, like, all AI is not LLMs, there’s, like, LLMs, and then there’s, like, other stuff as well. Yeah.

Mary Shea:

Got it. No, it’s great. I mean, clearly we could go on for a couple hours here, because we’re all very passionate about the topic. I just wanted to circle back to David for a moment, and then get back to you, Tuba.

But, David, let’s talk about accountability, because that was one of the things, you know, I’ve been a sales manager, many of us maybe have come up through the ranks. And that was a key element.

Of that role, and as we get into more of these hybrid, relationships and go-to-market and using multiple agents, do we get a fuzziness around accountability, or does that tighten it up, or how does that look like in your vision?

David Knight:

I think accountability is kind of uniquely human. Humans want commitments from other humans. One example is we’ve got an avatar that helps us run our forecast meetings, and it does a good job looking at all the data and helping us understand and forecast, but we still whole forecast call. Why?

Because you want a rep to commit to their teammates that they’re gonna get a deal done, right? They’re making a human commitment to their teammates. And so, I think that empathy and accountability and those sorts of things are super important and uniquely human.

And as long as we’re operating in these collaborative things and selling to other people. we need to not lose sight of that, and let the humans do things that humans are good at, and have the AI help scale their ability to do that productively.

Mary Shea:

Gotcha. Thank you, thank you for that. So… so… I… I… I want to go back to YouTube, you know, if you start to think about, This middle management layer that’s used to develop people, it’s gonna start to thin out. We’re seeing that in, you know, restructures across the board with companies across multiple industries.

how… I worry about, sort of, the next generations coming up. Like, I feel like some of us are really, really great.

at using the agents, because we’ve been there and done that for many, in my case, many, many years, and so I can call BS really quickly, and I can get it back on track really quickly, because I’ve gone through the hard knocks of… like, doing everything manually, like, how… How do we help build judgment and skills, you know, for this… these next generations that maybe won’t have the level of experience that some of us in this panel have?

Tooba Durraze:

Well, the experiences, in a lot of ways, work to our disadvantage, right? Like, why are we all… why do we all have custom ACV fields in… our CRMs is a great example of that. Like, we went, like, hyper-personalization, so then we created this problem where now we had to create semantics and, like, context errors, etc.

So I think, I think if you think of them as, like, basically people who are growing up in this world where agents were at their disposal, like.

Mary Shea:

Sure.

Tooba Durraze:

always had Siri on their phones, there was not even a thought that occurred to them that there was a phone before Siri. The world looks quite a bit different than how we’re trying to solve for it right now. We’re in the messy middle of.

Mary Shea:

Yep.

Tooba Durraze:

There’s a lot of legacy baggage, there’s some tooling, we’re trying to figure out the best way to utilize tooling to get ourselves out of the messy middle.

But, like, the third S-curve is, like, a completely different thing, and will require like, new paradigms will require new ways of working, will require new ways of coaching, training, all of that. Which I think… I think that… those playbooks don’t exist, are not written. Yeah.

And it’s not, like, to a certain degree, yeah, we’re kind of… you know, helping shepherd it there a little bit, but, the exponential, like, acceleration that they’ll see, will be, like, very different versus, like, what we see. We think our world is moving fast, their world is moving quite a bit faster.

Mary Shea:

Yeah. That is very true. Thank you for that perspective, because I’ve been thinking about it a little bit differently, to be frank. So, I see my friend Scott Brinker here, and a couple of other people. We’ve got 4 minutes, and, we all like to talk, including myself.

So, I’m gonna ask you all to give a very quick You know, sort of… one sentence, rapid fire, and Chris, we’ll start with you. You know, what’s the one thing every CEO sitting in the session should do on Monday morning, when they start working to more aggressively and deeply move their organization towards native AI?

Christopher O’Donnell:

I think identify the gaps. What are the real questions? To which the missing answers would change the behavior of the org. And even unrealistic questions, you know, the real wake-you-up at 2 AM, you know, write those down, because they are… they are answerable now.

Mary Shea:

They absolutely are. Thank you for that. Amos.

Amos Bar Joseph:

Yeah, so, Chris, you nailed it. I feel like stop… stop looking externally for answers. Nothing has changed with how you run a business in an AI-native world. It’s the same thing, you still need to find your biggest bottlenecks, identify the process. map where the friction is, and then go outside, whether to build or buy.

It’s the same process, the technology has changed, but the way to identify your challenges and what’s the future state still looks the same. So, lean into that.

Mary Shea:

I like that a lot, Amos. Thank you. David?

David Knight:

I think people are focused too much on what the AI can or can’t do in its skills versus humans. I think we need to understand that, that AI and humans are different, even if the AI’s human-like, and there’s certain things that are uniquely human.

The, you know, the empathy and the trust and the leadership, and those kinds of things, and so you need to decide where you want to apply the tech, and where your customers want you to apply the tech, and then, design your organization appropriately.

Mary Shea:

Thank you. Tuba, do you want to close the deal here?

Tooba Durraze:

Yeah, stop talking about AI transformation, number one. Number two, the questions that you’re writing down, according to Chris, in the middle of the night, that keep you up at night, give those to your folks and say. solve for this. That’s what’s keeping you up at night.

And the things that they use to solve for that, methodologies, tools, all of that, is the thing that is going to allow your oak to be like, AI trans… AI transformed, right, in that way. But stop talking about, we need to lead an AI transformation, like, that doesn’t…

Mary Shea:

You know, and thank you for saying that, and I was just talking with my business partner this morning, and we went back and forth, and I said, I quoted the Yoda, there’s no try, only do, which I think is what you’ve just very elegantly said, Tuba, so thank you, thank you for that. What an amazing panel.

I feel so honored and happy to be connected to each of you. I would love to continue the conversation one-to-one, in a different forum, and learn more about each of your companies, because, this was just fantastic. So, a huge thank you to Julia for bringing us all together, and… Allowing me to, moderate this event. It’s, it’s been terrific.

Thank you so much, everyone.

David Knight:

Thank you, Mary.

Julia Nimchinski:

Thank you so much. What a phenomenal panel, that hour really flew by, and let’s just do one last minute of shameless plugs, all of you. Amos.

Amos Bar Joseph:

Follow me on LinkedIn. I like to talk about AI stuff and how to build an autonomous business. I also have a newsletter called The Autonomous Age, so you can follow me there. And if you want to learn about Swan AI, go visit GetSwan.com. Thank you, Julia, for having me.

Mary Shea:

I love it.

Julia Nimchinski:

Suba?

Tooba Durraze:

The biggest way to support me is, like, if you are thinking about decision-making in the future, we are co-designing a new program around that, so if you’re a C-suite leader who’s interested in what that future needs to look like, reach out to me, and then you can find us on amoeb.ai.

Julia Nimchinski:

Christopher.

Christopher O’Donnell:

Go to day.ai to see what comes after CRM, and start using it, even in the CRM age. Come, look us up, and get on the horn with us, and let us show you this thing.

Julia Nimchinski:

Thank you, David.

David Knight:

Yeah, if you’re a go-to-market leader and you’re struggling to get your team to run your play, why won’t they just run my play? We can help you scale running your play and come to Avara with two hours.ai and learn how.

Julia Nimchinski:

Awesome, and Mary.

Mary Shea:

Yeah, thank you for that. So, follow me on LinkedIn. I’ve been a little quiet lately because I’ve been in major build mode, but we have, built, and I can’t wait to get all of the word out to the world. Also check us out at, TryMeerkat.

ai, and just as a quick snippet, MeerkatOS, it’s a single intelligent layer that connects people, relationships, and context into shared machine-readable memory. So, really excited about, the possibilities, and look forward to hearing from everyone and continuing the discourse.

Julia Nimchinski:

Thank you so much again.