What AI Agents Mean for How You Build and Sell Software with Brian McCarthy

0:10Welcome to the Revenue Builders podcast hosted by John Kaplan and John McMahon. Today's clip features Brian McCarthy, president of global revenue and field operations at Cursor and former CRO at Rubric, where he helped scale the business from 118 million to$1.5 billion in ARR. In this segment, Brian breaks down the rapid evolution of AI and software development from simple code autocomplete to fully autonomous agent-driven environments and what that shift means for competition, product strategy, and go-to-market execution. He also explains how Cursor's multi-model approach creates a unique advantage in a fast-moving market and why aligning that innovation with focused sales execution is critical to winning.

0:51Let's dive in.

0:54you know the ai space is like this so uh first of all not the ai space let me be very uh specific about this ai space a massive space it does a lot of things our world is very in particular to the sdlc it's a software development life cycle it is about helping companies deliver and ship software or applications, whether that's tech companies or big banks or whatever, build applications faster and deliver them higher quality, faster, and with less cost. So improve the productivity of an engineer to be able to do more with less. And that space is not a binary space. So if you think about it, what's required to in this space, you need to, it started off, this space started off with what we would call engineering assistance.

1:56Originally, it was tab, you know, so it was like, you know, think about like auto complete on your phone, right? You're texting and a word comes up and it just auto completes. It was that for engineers. It essentially is, hey, you're in there writing code and it knows what you're going to write and it completes the code. So that's, that was the start of code generation. It was like, and the companies that did that best were the ones that had deepest understanding of the code base and most engineers using it. And because the engineers used it, they had the best ability to anticipate what they were going to write and complete the code.

2:36That's why cursor blasted onto the environment. We had 3 million engineers using it. We were the best at anticipating code completion. It wasn't a comparison. And so it exploded. And you didn't need sales to do that. It was just get engineers using it. It's self-serve. And they did that, Brian, at the AI level, at the task level, right? So they're basically, the reason why there's very little competition, I think, right now is because it was done at the task level. The AI was built right there. Yeah, totally. So essentially, you know, as people were building and building code in what was referred to as an IDE, which is the, you know, individual development environment, right?

3:25And so their environment is in there. They're in there writing code and they're going out, getting code, like searching for code in the code base, bringing that back and then auto completing that. And that was happening in Cursor. And what changed, what pivoted from that, that was, I would say, you know, these lifespans in technology in our world used to last five years, six years, get disrupted, something comes along. Now it gets disrupted in a year or months or weeks. And so this was the most innovative thing in the world. Every engineer in the world wanted it. It was like nutty. And then it went and got disrupted in months.

4:07And so then what happens is Claude comes out and the next frontier is, all right, do you even, you know, you don't need to go ahead and complete the engineering test because you don't even need to be a linguist anymore. The agent can build the code on its own. So and so that pivot took place. And now you see companies like Cursor that are just leapfrogging in that space. And what they're doing is they're creating, instead of an IDE, an ADE, an agent development environment where agents are building code on its own. And so this then is the current model that we're in. And this becomes like a little bit of a competitive market currently.

4:59But the way the market is competitive is this. Those agents are going against the model. And that model is fine-tuned to be able to generate the code. So it goes in through the model, pulls it out. So if you think about models like Opus and Anthropics model, Claude going against Opus to go build the application or generate the code for the application. Cursor came out with its version, Composer. And this is probably more than everybody's super interesting, but it's just it's really interesting how quick the place moves. And so, you know, in Cursor, it has its own agent build and its own model that comes.

5:43And what happens is every... Which has 3 million users. Yeah, exactly. And so every few weeks, models jump each other. Opus 4.6 was the best model in the world. And then OpenAI came out with the best model in the world. And by the way, Cursor just came out, Composer 2.0, and that proved to be the best. Like objectively measured on performance, ability to generate the code, completeness of code and quality of code, amount of code kept. It was inarguably better. So every time the models keep jumping, then you have Devin from WindSurf or the Cognition folks. You have DeepSeek and the open source models.

6:23You have Google and you have Microsoft models. So all these things are being built. And what Cursor did that's very different, and here's where I'm saying this space is like super wild, is Cursor is agnostic to what model you use. Yes. So if you're Claude and you're a Claude user, you're going to consume an Anthropic model. When they reduce the model, you're only locked into that model. However, you can use Claude and you can use the Anthropic models in the cursor harness. You could use Codex in the cursor harness. You can use Codex against their model in there. You can use our model. You can use our agent build.

7:07You can go directly against the terminal. It doesn't matter. You can switch based upon capability and pricing. Exactly. So we provide essentially a very differentiated solution. And this is where we're almost like more like frenemies right now with a lot of the competition. More Anthropic is consumed through Cursor than anywhere else in the world. In fact, when OpenAI announced, I think they said like five exabytes were already consumed. It's like a week ago or two weeks ago of their new model. 40 % of that was consumed through Cursor. 40 % of all their model usage. So what that means is that Cursor is giving the engineer the unique ability to be able to pick the right model for the right use case, because not everything might need an Opus 6.

7:58And there's a cost trade off between speed, quality, and the dollars it takes to go build that and agent build. And so Cursor is giving people the ability to automate that process, to auto-select the right model for the right job. And in that way, we see this space slightly different. This is where I was getting to, John, around the competitive thing. We look at Anthropic and an open AI is going to be our best partners in the world for a long time. That's why I talk with Paul and Brad regularly. I'm engaged with these guys, you know, and I look at them as the hyperscalers. and we're the snowflake databricks of the world in that we're building the vertical expertise for SDLC and they are building models and agents to solve maybe the largest TAM in the history of time, human work, which is a$50 trillion TAM.

8:52They're trying to build models for poets and actors and voiceover and nurses and lawyers and everything else. We're just building agents for SDLC and and do that better and faster than anybody in the world. And in doing so, we're going to consume and burn down more anthropic usage than I believe Claude will burn down in due time. We'll burn down more open AI usage than Codex. I'm selling the picks and the shovels for the gold rush, right? Exactly. So that gives everybody a little bit like a... That's really good. Amazing. Talking about that, though, and you have all these deals, where does account management fit?

9:34Your reps are trying to figure out if they're doing the best deals, the$1 million deals. Otherwise, they can't get out of bed, let's say. Where does the account management fit into this picture at all? Totally. So one of the things we did right after I heard that comment about getting out of bed, I segmented the business. So I segmented the business. It took me seven days on the job. I segmented the business like this. I created an enterprise strategic vertical that no rep had more than four accounts. And I opened up 80 reps to go hire 80 reps across five verticals just for North America. They can have no more than one existing customer and three prospects.

10:16And now that becomes a forcing function to go execute and run, like both give our customers and our prospects an incredible experience, a buying experience to understand their needs and be able to help them through their AI journey properly and give them the right attention and focus. And it also lets a seller exercise what they love to do better than anything, right? Like they actually have to go do PG and go build business value cases and understand the business value justification and meet the engineer and the top down together in the middle where you're having, you know, value-based conversations as well as the bottoms up approach.

11:00So we did this. And a lot of people you all know, we've now ended up hiring. I then segmented the enterprise business to go from 50 accounts down to no more than 18. And they could have no more than three existing customers. So they couldn't chase. Then I created another segment, what we call GEO. They can have no more than 30 accounts and they can have no more than five existing. Then I created a commercial and they are basically everything under 250 engineers. And John, it's the only time in my entire life that I've ever segmented business and taken accounts away and changed the entire structure and nobody complained.

11:40In fact, every single person in seats said, thank God, like now I can do my best work. You're helping me. Like nobody explained that their territory got smaller. They all said, this is, I've been begging for a cavalry to come so that we could do better work.

12:21Thank you.