Speaker 1What is the future of sales in the post AI era? That's a key question on Sony, people's mind. Have I got a doozy for you? I'm Brennan in my buddy, Christopher O'Donnell. In 2013, Brian and Darmash, the two co-founders of HubSpot sat he and I down and said, "You guys are going to bring the HubSpot CRM to the market." So for years, we pontificated around how sales should work. Today, Christopher has founded Day AI, a stage two capital portfolio company. And he's been thinking about this all day for years in so-of-eye. Today, we sit down for the first time in exchange notes on the future of CRM, the future of sales, and the future of go-to-market system design. I'll mark her bearish, and this is the Science of Scaling. Alright, so Christopher, I'm so excited, a life moment here to have gone through the battle we went together in the last generation of sales tech, MarTech, go-to-market strategy. Here we are again, your new company, Day AI. You've been thinking about this for years again in a new era. So let's pull that out of your brain and let's start with the roles, the talent, the jobs. Like, there's all this press about. I'll ask you. It begins with like, "SGR is going to lab." So like, where are you at now in all the thinking and research and, you know, customers' cover you've been doing on what these roles look like?
Speaker 2Yeah, every day there's a new news article saying, "Nobody is going to work five years from now, or nothing's going to change. Everybody's going to have the same jobs and just be 15% more productive." And then, "Oh, the research shows they're actually 15% less productive with the AI, and it's kind of all over the place." You know, look, I think some of this is settled in my mind and clear now, that people will work, or we're just all screwed, right? And the AI will take over the world and, you know, there's nothing that we can do about that. Anybody who is still working three years, a year, five years from now, I think it's safe to say that they will be producing two to five to 10X as much work product. And that will have impacts on sales, that will have obviously impacts on engineering where we've seen the most of this progress, I think, so far. It will have impact in wonderful areas like pharmaceuticals, you know. We may actually cure cancer. Okay, fine, but I drive into work and I check email and I have meetings like, "What's going to happen to me?" That's the lens that I would use, is how can you produce more work products? As people do produce more work product using AI, using agents, those roles will start to blend together. I think in really beautiful ways that are great for customers that make work a lot more fun. Everybody's talking about taste right now, which I find delightful. It's super important. We are going to be taste makers in this new world and be able to spend more of our time on the details, the why, and on things like design, things like copy, and getting the fine details just right, because the kind of grunt work of whatever it is, updating a CRM or writing code, that stuff kind of blends away. You still have to have a strategy, you still have to understand what matters, but you can actually get to 100%. You can actually get your product to be perfectly consistent. The customer memory across all of your reps and anybody who's ever talked to anybody, you can hold on to that, right, and grab it instantly.
Speaker 1Yeah, I've been thinking a lot about this. What are the future jobs for humans in a very mature AI world? And Christopher's ideas are giving me a lot of conviction on some of my hypotheses. I've always said since the beginning, politicians, and we do want humans out there that we're voting on that are making decisions on how we run as a society. Number two, I agree with him. He says taste, I say creative. I just think you put Ed Sheeran and Taylor Swift in the room against AI, even 10 years from now, and they'll still create quite a few best selling hits. And the same applies to movies, to books, to art, certainly to sports, et cetera. And then this one's a little trickier, and I've developed a conviction more recently in Christopher saying it in a similar way. And I would say human in a loop. I think technology's been able to fly planes for quite some time, but we still have pilots there just in case. I know when AI is able to perform, if I'm having brain surgery, I'm okay with an AI neurologist, but I would like a human neurologist in the room just in case. And if I buy a million dollars a software from someone, I'd like to shake a human's hand to make sure that they have blessed my needs and the fact that they will actually deliver. So I agree. There's some ideas on the future human work in a post AI era.
Speaker 2So that's kind of where I see it going is. Everybody needs to start thinking about if they could hire a virtual worker of some kind, or even if they could hire somebody and they walk through the door right now, what would that job description be that they gave that person? And to really start investing in those artifacts, even if you don't have headcount, actually, especially if you don't have headcount. And especially if you've never written a job description in your life, you should still take your current role and cut it up into one, three, five job descriptions of work that you don't want to be doing. And focus on your new job description, which is work you do want to be doing. For sellers, that means being on, you know, camera or in person with a prospect or with a customer and listening. That's a great thing about these meeting note takers is instantly we can pay attention in the meeting, we can build rapport, we can talk about paternity leave, maternity leave. Oh, how are your kids doing? You know, what sports teams, whatever. We can get very deep into discovery because we know that if we miss something in discovery, it's going to pop up and remind us. Or if a question has already been answered in discovery through some other channel or some other clue from somewhere, we know we don't have to ask that. We can focus on the rest of what's missing about how this prospect could succeed. So just taking those things and starting to cut them up into this new agent kind of workflow because we all do get these virtual workers. We all get them today, right? There's no worker who can't have some AI companion now helping them with their job, literally.
Speaker 1Okay, so I love that. I think that's very accessible right now is like everything besides seller talking to buyer. Let's make that, you know, let's AI that. Do you see it going beyond that? Do you think that we'll have agent sellers? Do you think we'll have agent buyers? I don't know if you've gone down that thought exercise.
Speaker 2Yeah, and you know, there are folks doing really cool things like customer onboarding for a SaaS app. There are things you can get out there that are like being on the phone with an implementation person who coaches you through how to use the product. I think that stuff is great for people. I think it doesn't eliminate the need to have somebody who owns that customer relationship but maybe it does blend sales into account management. Maybe that's the right way to do this, you know. I think ultimately all of this allows you to solve for what the customer actually wants. Why do you have six touch points with a given vendor? It's not because the customer wants it, the business model requires it because these are specializations. Working a renewal is different from training a customer, is different from selling a customer doing discovery qualification, right? These are different skill sets but maybe some of these we can hand off to the AI and the role of that seller becomes something really new and exciting. One thing we talk about is we wonder if sellers will become masters of their own domain. Remember in the early days we had sellers who would focus on, there's one I'm thinking of in particular, who just sold sign shops. I think she's still there, right? Been there like better part of 20 years and found this thread. Okay, there is like my uncle-in-law Vinny has one of these sign shops in Buffalo, New York. Totally. And she just knew exactly how to help this person get leads and so has sold like every sign shop in America. We think that that's the kind of thing that'll happen where somebody with a broader, more general role. Okay, maybe they're not a marketer but as part of being a seller they know everything about this persona and can get all the help they need to do battle cards, white papers, customer testimonials, marketing site content, you know, solutions page or whatever for that particular slice that they sell.
Speaker 1Yeah, that's interesting. 2025 was definitely year of AI in product and engineering. And Christopher's talking through how there have been so much specialization from just engineers to product-based. front end engineers, back end engineers, designers, data analysts, data scientists, etc. And he sees it going back to a single role. And I think we see the same things in 2026 and go to market. In 1995, we had one role, salesperson, they did it all. And the last 20 years, we went to SDR, AE, AM, CSM, customer support, rev-ops, etc. And AI will probably get us to go back to the athlete. Specialization is great. It allows you to align the talent with the role. And use the hardest to find skills in that most important part of the funnel. But it comes with a cost, a cost of local maximum optimization. And a post AI world, the full cycle salesperson, will be the optimal design.
Speaker 2So that type of expertise becomes more valuable because it's easier to leverage in different ways where it used to be very siloed. You know, the seller isn't going to walk over to customer marketing and share all this stuff necessarily. And so it kind of goes into a black hole.
Speaker 1All right. I'm curious if we've landed on the end architecture. If I could draw parallel to the.com era, I think in the beginning of the internet, everyone thought like the internet was a brochure online. Sure. I was like, oh my gosh, this is like sharing economy. This is like, there's so much cool stuff. That turned out to be 1.0 that all got disrupted. I think we've tasted that with like the initial co-pilot era, which ended up being these wrappers, which are super light, and all those kind of like, went away. Now in this agentic era, have we landed in the longer term architectures? That what this is going to be? Or is there another one? Talk to us through this agents.
Speaker 2Yeah. Red Point has a great piece recently where they talk about these phases. And I think they have those phases pretty much right. And you start with this co-pilot era, and you end up with this very ambitious sort of strategic agentic era where entire parts of your business are being reconsidered and replanned dynamically. And we're kind of in the second or third chapter of this. I think some of it is still very undecided, and some of it is pretty decided. In the language that we use around this stuff, Jerry Tan has this G-stack Claude Rig that went viral and has hundreds of thousands of GitHub stars. And nobody even knows what that thing is called. Like is that a skill pack? Is that a harness? Is that a workflow config? So some parts of it are evolving really quickly. I think there are a couple parts of it that are really decided, and that is the idea of an agent and the idea of skills. Any AI product you go to has skills, whether it's the DIA browser or perplexity, Claude, cursor, Claude code, whatever, DAI, it's a slash command. It's a set of instructions about how to do a particular thing, whether it's your browser, NBA score checker, or it's a GitHub pull request review doer. Whatever it is, it's just a little training manual that can get pulled up in the right moment so that AI can do something perfectly every time. That is settled. Like that's definitely not going anywhere. The idea of an agent is probably also not going anywhere. The idea that an agent has a job description, that idea that I'm going to be responsible for three times the work product, well, who's going to do this work product, agents? Okay, so what is an agent? You need more than one. They have a job description. We give them a view of the world, a set of opinions of what's important and where to look to get your bearings. In the case of OpenClaw, which is always a fun one to talk about, don't use it at work, but boy, is it powerful? It has that same architecture as well, and it goes so far as to have a sole document. That's one of the things that you set up when you do OpenClaw and get that running, is you actually give it a sole, like a name, a whole identity, a purpose for living, a backstory, whatever, that I think is actually settled and is correct. And so there are a couple of layers that are still being decided. One is the automation layer, even way back in the day, which I'm sure you did a long time Salesforce reps. The ones we talked to said, look, at the end of the day, I feed my family selling automation. That's really what it is. It's Salesforce automation, SFA is really what people think of as CRM. How do you automate these skills? We don't know. We don't know yet. Claude code has a loop command. Now Claude desktop and web have this idea of an agent platform that can run asynchronously. DAI lets you take a skill and say, do this every four hours or do it when there's a new opportunity added or whatever, right? That I think will firm up to some degree, but there will be a lot of vendors and a lot of different ways of doing it. There will be a lot of vendors and a lot of ways of doing it, but we'll coalesce on this idea of a job description and using skills. And then there's the fourth component of it, which I think we're really entering into the era where people really need this and see the need for it, which is, what's your data source? So these agents are wildly intelligent. You can have as many of them as you want. They have job descriptions. They're ready and willing to do work products. And what are they working off of? Our take is that they should be working off of the actual customer truth down to the word that was said, the minute it was said, who said it, why it all matters, and to keep all of that wired up to be instantly accessible, regardless of what happens in these other
Speaker 1layers. I want to ask this question, because I think everybody's out there. You go talk to a founder, you go talk to CEO, and they're always like, we are so AI native as a company. We're so AI native. And what I've been pushing the ecosystem is like, how do you know and how do you measure
Speaker 2that? In R&D, I mean, in engineering, Stripe had a great post about this recently. If you look up Stripe minions, they talk about the productivity that they're getting now from agents running in the background, some of this automatic orchestration that I'm talking about. I mean, the industry average for an engineer pull requests per week, you know what it
Speaker 1is? No. Oh, God. So now your test test. I have no idea.
Speaker 2I've never studied it. So four requests is like a unit of work. Yes. We're showing somebody that turns into a deploy.
Speaker 1Five. I don't know. I'm probably way off. Two and a half. Oh, okay. Two and a half. Not bad. Because I could have said 50 and had some thoughts that that would be the case.
Speaker 2Okay. Two and a half call request. Yeah. And it really could be anything because you could have a culture where they're smaller or larger or larger. But two and a half pull requests a week for a good solid engineer. And this was true at HubSpot. This was true at Stripe. They talked about this. And I think that you are seeing 80% of these engineers still at two and a half. It's a problem.
Speaker 1Huh. That's a problem. And they're all bragging about. Oh, man. We're using cursor and cloud. And all this stuff.
Speaker 2And they're still at two and a half. Yeah. And you see these articles in the news about, oh, the AI is supposed to be making everybody 15% more productive. Yeah. And actually, we think it's 15% less productive. My take on all that noise is, it's going to be pretty bimodal already. It's pretty bimodal where you have people who are known as 15% more productive, right? Like that would be people using auto-complete or something, but still coding manually. That's not what anybody is really doing.
Speaker 1Yeah. That's happening all the time right now. We're in the boardroom and they're like, our sales team is so AI native. Our engineering team is so AI native. How do you know? What are you basing that on? Because you're using open AI to write emails? That's the key we have to get to now is quantify it. How will you know in 2026 whether you're truly an AI native sales team? I'll throw out some metrics. Number one, selling time goes from 25% to 75%. Selling time is the percentage of a week that a sales person is in front of a buyer or a customer. It goes to 75%. Get through to all the admin work, the CRM prep, a lot of the meetings with pipeline reviews with your manager, et cetera. Number two, the rep to manager ratio doubles. It goes from say 7 to 1 to 14 to 1. Why? Because one of the most time consuming aspects is coaching. But today's AI is a better coach than humans. They can look across 70 calls, not just two. And they can look across a number of coaching calls to know the type of coaching that the human is most receptive to. And ultimately, you double productivity per rep. So Bob and Susan who for years and years and years have averaged 250k a quarter will start averaging 500k a quarter. That's how I'm though, if you're in it. AI enabled sales team in 2026. -
Speaker 2I think you have people who are two to five, in some cases more than that, more productive. And you have a lot of people who are still at the baseline. -
Speaker 1So that's what I would say at the board meeting is like, okay, fine, you're bragging about how like AI native you are, what was your poll request per week in the last quarter? And I'm looking for five, which would be a two X, up to 20, which would be like, you know, a nine X or something like that. -
Speaker 2Yeah, exactly, and has it changed? I mean, whatever, whatever that particular company's baseline is, you know, it's just work product. -
Speaker 1Can that be cheated? Like, can you be like, oh yeah, we've crushed poll requests, but it's not actually driving so and so productivity on the-- -
Speaker 2Oh, sure, sure, sure, sure. But it's pretty obvious. I mean, if you start working in this new agent a kind of way, you're not going to have to really cheat anything. And it's kind of an enormous amount of work to try to show that you're using the AI. It's a lot more work than actually using it. So it's dramatic in the amount, that's kind of the lens for all of this, sales, engineering, anything, is how much work product are you, you know, the colleagues sitting next to me in charge of and getting credit for versus the baseline of 2023 or before that. And I think across all these departments, it really is, you know, radically more work product. -
Speaker 1It's crazy that like, we under appreciate both in the on-prem and the cloud era, there was a decision around a database structure, tables, rows, columns, that the entire industry had to conform to. And that's essentially what you're saying is, we don't have to have that, that, you know, abstraction there. -
Speaker 2It's not that, you know, you couldn't build something like Palantir, using Oracle relational database. - Exactly. - There are things you can't build on top of even Salesforce. - Salesforce, you know, ultimately is extremely good at the things that it does. HubSpot is extremely good at the things that it does. And now people are turning not necessarily away from CRM. They're saying, yeah, yeah, CRM, like it has my opportunities and people get paid because they keep it up to date and there's some value in all of this. But okay, if I'm gonna run agents, I need all the gong data. I need all the Slack data. I need all the email data. And then you start to get into this world where it's like, okay, we're gonna ask the CEO to share their email inbox. We're gonna ask the header product and this support wrap and everybody at the company to just pool their data together. And you realize pretty quickly, okay, well, we're gonna need to give people some controls, you know, so that they understand what is being shared, what is visible to everybody else at the company. In some cases, if you have a customer facing agent, boy, you better make sure that they can't, you know, delete all the meetings from Mark's calendar just by asking it to. So that's gonna be a really interesting story to watch on the phone.
Speaker 1That is such a challenging thought exercise. It's how the level of abstraction changes with time. You know, we could think about like something very minorly abstracted from today's point of work, like write this email and that's where we are today. Will we ever get to a point where we just say, build a billion dollar company and it just builds a billion dollar company? Who knows how far abstract we will get on which time frame, but I will say this, those level of abstractions will likely define the phases of evolution of our post AI world. For example, in sales, we already talked about how the first phase may just increase selling time, but the actual process is humans selling to human buyers. And perhaps the next level of abstraction will be the functional boundaries that we know today, start to blur. In a way, we have finance, sales, marketing, engineering, product because of limitations of humans. There are very few people who study finance and write code and vice versa. And yet those functional divides create suboptimal inefficiencies within organization. We would benefit from finance being closer to sales, product being closer to support, HR being closer to product. Perhaps over time, those functions will blur in the optimal organizational design. -
Speaker 2This is the new thing that people are trying to roll around. There was a brief period of people talking about vibe coding a CRM. And look, you can vibe code a CRM, single player. Could you really do that in a weekend? It's like if you're just trying to build yourself a CRM, you don't need a weekend. You need 45 minutes. The minute you go multiplayer, and now you can see your leads, I can see my leads, but I can't see your leads, then you go, I think maybe I'm gonna go to a vendor. I think I'm gonna go buy this one. I'm gonna go buy sales. Now in this customer memory, which is sort of a superset of what CRM tries to solve, it gets really intense really quickly. You need to be able to populate that customer memory almost instantly to get going, and you need it to be built originally for the AI. We no longer judge whether the interface or the feature set or whatever is to our liking, because we are working with the AI. The AI is working with those tools. See what I'm saying? - Totally. - And the AI will tell you what it can and can't do based on those tools. You can connect a bunch of different tools to something like Cloud and say, what do you think of all of these? What could I use these MCP servers for? Try some of it, tell me how well it goes, and it will give you lots of ideas. And some of those products are incredible products, and they won't really give Cloud what Cloud really wants. So yeah, I mean, there's absolutely room for a new data structure. When people are building these customer memory layers themselves, they are not using relational databases. It's not their first instinct. They go to vector databases. They try property graphs. They try all of these different interesting things. I do think ultimately they'll realize that this is the kind of thing they should use a vendor for, and that's where we come in. -
Speaker 1Let me ask, this is a cool call and apologize, but more like where the level of abstraction ends, and this is something I'm struggling with, I want your take on. I'll make a simple analogy on a toy manufacturer. So there's all these people assembling toys. They come up with the ideas, et cetera. And the first level AI just automates the assembly part of a bar, and then the next level AI might be like, create an idea for a toy. And so we're abstracted even further. And then you go to first principle, what does a toy do? A toy creates entertainment for children. So this is abstracted. What's the best way to create entertainer for children? And then you make a billion dollar company. So you apply that to sales, and it's like, okay, we already see the first level of abstraction, and we talked about selling time, right? So it's like, okay, we're getting rid of all the admin stuff. And they're like, how far does this get abstracted to like, okay, closing customers, okay, inventing product, okay, creating a billion dollar company? I'm struggling with that level of abstract, is that where we get to? Or where does it end? -
Speaker 2I don't know where it ends. I started making agents, and I made a user researcher agent and a data analyst agent, and a marketing copywriter agent. And then I made an agent to manage them. - Right. - And I was really blown away by that VP, who immediately started asking for all of the things the real VP would, like, hey, wait a second. I need more data. I need more of this. Like, you gotta give me this. And I'm like, this really is like humans. And then I realized, okay, this is awkward. I could make a CEO agent. - Right. - And I did, you know? There are still decisions that need to be made that have human consequences. You know, we have investors. I can't just give away the strategy of the company. We have employees. I can't count on the AI to inspire the employees and to answer their questions and to meet them in their train of thought and try to explain how we see the world as a company so that they have the piece of the puzzle they need to continue. Now, as far as sellers go, I think it's a good example. I think we sort of agree maybe that we want a seller to be making the human connection, providing that trust and that credibility, building rapport, being the face that is answering these questions. It's hard for us to develop trust with the agents. And we can talk about that, particularly with somebody else's agent. But with sellers, there are some pretty obvious opportunities. If you imagined you had infinite time to review your deals and to look at the internet and to just imagine all the possibilities into pattern match, well, the AI should be able to say, you seem to like generating pipeline. Why don't we go develop a plan to generate more pipeline? Starting with, here are the customers that close the fastest. Here are the patterns and the similarities between them. Here is what they said in their own words. They came to you, looking for, and. And here is the moment where your solution made sense to them. Here were the words that were used by the seller. Here was how the prospect reacted to that in their own words. So imagine taking the entire history of everything and just extracting from that history, just those bits. Who are these people? What are the patterns? And then being able to say, get me 100 more of these. And imagine that when you reach out to them, you are using all of that customer success and all of the previous prospects, previous customers actual words. Their actual feelings, not a bunch of us sitting in a room coming up with marketing slogans, but actually the source, the record of the reality of what happened. That turns into your outreach to these people. I get these cool emails from AI, SDR apps all day every day, and none of them are any good. I mean, they seem to kind of pull my LinkedIn, and that's pretty much it.
Speaker 1In your role at AI and found, it's like the same generic concept.
Speaker 2My favorite one though, my favorite one was somebody talking about the cathedral, because I live in Winchester Mass.
Speaker 1Okay. And it's like, I think you're
Speaker 2talking about, like, Westminster maybe, and that's definitely a different country. Right? We have a very modest, like, Anglican church in downtown. So that's where I think this kind of goes, is hyper accelerating the part of our job that we really like. In that case, being on the phone with somebody who really needs the thing you have, I mean, it's a great feeling. Right? When you just have those, like, Bloomberg, just, oh, my God, they're going to love this thing. And you go home to your family, and you're like, I crushed it today. That's what we want to really speed up.
Speaker 1The future of sales is an approaching. It's here. Yet, many teams are still running a tired playbook. This guide breaks down the full cycle seller model, where AI fits into modern workflows, and how to give sellers more end-to-end ownership without stretching them thin. You'll get a hiring profile you can use immediately, a framework for measuring whether your team has truly restructured around AI and a transition plan for making the shift. Grab your free copy of the full cycle seller model at the link in the description. All right. Let's get back to Christopher. Is that, I think, in the last year we've kind of regressed, we've gone back to the human? Like, I think in a lot of situations, like, the last year, it's like, oh, yeah. Like, human SDRs are still outperforming the AI, SDRs, you know, like, if you, the human that writes this no, or like, the human demo, et cetera, is that just because of limitations of the AI tech today that goes away, or is that instructive to the future of human's existence in, like, their role in this new architecture?
Speaker 2I got a call last night from somebody using one of these SDR systems, and said, hey, is this Chris? I'm calling from such and such, real well-known company, you wouldn't know. And I said, are you using this app, and he's like, how did you know that? I'm like, dude, you have so much to work with, like, we have shared investors. I have done demos of your products. We decided to use this service provider instead of your app, and it didn't go well, like, you have all of this history. And so humans and AI are trying to do this thing, in this case, to establish an initial connection from cold, from zero. The AI is there, the humans are there. All of the technology is there, and yet, this outreach is just really kind of vapid. What's missing is some instant access into every data point, and every clue, so that these hyper-intelligent AI models can come up with something to say. If you give them the right starting information, they will come up with something to say. And very good at writing, very good at helping people communicate with each other. This is one of the biggest things people use ChatGPT for, is, like, marriages and dating, there are personal relationships. These models are very good at personal relationships. They don't have the starting material. They don't have how do we think about this person we're writing to in the context of everything we've ever said to anybody. There's no ability to find patterns and start from those patterns when you decide to reach out to people. So in many ways, I think it's a problem that should be solved by now. But these emails are terrible. These emails, these calls are terrible, but I can give you, I can give you an AI BDR that is not terrible. I mean, give you anything. I love that. A lot of the vendors that I have, and I love them dearly, love them. I don't mind buying things. I like buying things. It's super fun. But these folks, as it goes from the BDR to the AE2 all the way through the renewal manager and everything, they have, it's like the goldfish with the five-second memory. They just have absolutely no idea who the hell I am. That is inexcusable at this point. Everybody should know every detail that matters.
Speaker 1Okay, so like before we get off it, picture 322 year old MIT and Caltech engineers building something right now, you walk in, you know there's going to be massive holes in something. They're just like probably building too much in a vacuum. You need to convince them with a very tactical tip on like how do you stay closer to the customer? What's that one thing you tell them to do?
Speaker 2I don't want to be too self-promotionally. Do it, you can. Yeah, no, I mean if you just set up day AI, recorder meetings, connect it to Slack, whatever, everything comes into what we call the customer memory layer. And then from there, it's pick your poison. If you're more comfortable with a web app environment and seeing tickets fine, if you want to automate the whole thing so that it just flows through to GitHub, the personas that you're talking about, you could just wire it up all automatically, or you could do something in the middle of Cloud Code, you know, whatever that's. But whatever you do, you know, you could connect GONG or you could connect Fathom, you could roll your own Slack thing, email gets a little bit tricky. But yeah, I think we're definitely past the kind of ticket concept, you know, level. And you should be able to have some sort of automated action off of that. And then start to roll that up into higher level, more sort of synthetic strategic insights. And be able to ask questions about, you know, who would this impact if we did this feature? You know, what are the biggest gaps for this new persona that we're going after? I think that's the expectation, regardless of what you use, I think that's the expectation that people should have, is that a lot of that prep work is done for you and the raw material is there for you.
Speaker 1So like, all that said, let's talk about the future CRM for a minute. And what I've extracted from your conversation so far is like, obviously relational database goes away. We're talking about this unstructured data that's really close to the front line, personal of the customer. We're talking about removing a lot of like the tasks that are, you don't want to be spending human intellect on, it can be done by the AI. Can you just, am I right, directionally, and like, can you solidify that? I'm like, what is the future of the tech stack that supports all this?
Speaker 2There needs to be some sort of instant memory layer for whatever it is that you're doing. You know, if you are doing legal work, you need to be able to grab all of the relevant case law instantly. If you are working in medicine, you know, or pharmaceuticals, biotech, you need to be able to grab all of the most relevant data points to the thing you're trying to do, apps like Open Evidence, unbelievable product. People love this thing, helps doctors and researchers because it just instantly grabs all of the exact pieces from canon, whether that's, you know, this insurance company's policy about such and such, or whether it's the best treatment or diagnostic method for this other set of clues, and it's instant and it's boom, it's right there. You know, this is information that exists somewhere, maybe, and can be assembled and recalled perfectly in time for a person to get the answer that they need. So the memory layer, no matter what you're doing, becomes the single most important thing. You can get agents, you can get skills, you can get eventually automation from any vendor. You can get it from clot or Chachy PT, you know, you can get it from cursor, you can get it from perplexity. The part that people will need to get right that is extremely hard is this memory, this access layer, CRM, I think, could play a role in that. But if you look at Jack Dorsey's piece on the world model of a business, it's only one of the inputs. Does it help? Does it hurt? I think it depends on how clean your CRM data is, how diligent you are. If you have a team that has kept everything up to date in your pipeline is in perfect order, that's really valuable information for the LLM, but it's not enough. This superset of information, of every fact ever, being able to ask a question and have all of the most valuable pieces of data presented to the AI when it tries to answer your question, when it tries to draft an email or give you a way to read an x-ray or whatever it is, that retrieval is really the big layer that's going to matter in the future.
Speaker 1Remember the zoom out, it's like the local maximum issue, that's pretty much the same. So I do want to zoom out, and before we get back to the HubSpot archives, one more point on this team design. And where I'm going to go. I'm going to zoom out to here for our product founders is you got to be very attentive to innovators, dilemmas, that you can exploit right now. That is so key. And if you don't know what I'm talking about as a product founder, go do a quick reading on innovators, dilemmas. Essentially what this is, as you know, Clay Christiansons, the Lake Clay Christians work. When we have these massive technology shifts, it gives a huge advantage of the startup relative to the incumbent. The startup usually wins. It's very different for the incumbent to defend themselves. As a founder, I have to be aware of what those opportunities are and exploit them. Not enough founders are aware of this stuff. And that's the big question these days is like, we know looking back on cloud what they were, what are they going to be in AI? And I have a hypothesis that one of the strongest ones is not necessarily how you AI your product but how you AI your team in organization, okay? And we're talking about this potential move from mass specialization in R&D and GTM to a single person that does everything. And what I want to, that's a massive opportunity for us to exploit. I want to hear what it is on the R&D side, but I think on the sales side, if I'm hiring someone directly into this AI-enabled team, my hiring spec is so different than what it used to be as hiring an A and SDR and CSM, that's a massive opportunity, because if you're a huge incumbent with a thousand people on your go-to-market team, and suddenly the right answer is like, oh, this is how it works now, you don't have many people that fit that spec. So you have to have to retrain people or get rid of people and replace people. As a startup, I can hire into that profile right from the get go. That is a massive innovator's dilemma that product founders can exploit. And I will tell you, I want to hear what your spec is on the R&D side. I will say, on the go-to-market side, if I had to promote like an SDR or an A or a CSM or whatever, it's probably an A-E, that's the best. But it's like, you're very tech-enabled as well. I could see myself promoting SDR. I mean, that's something I got to do some work on, is like, what is the optimal go-to-market hire in this AI-enabled world? I think it largely grows out of A-E.
Speaker 2Christian Sin says the startup founders have it kind of easy. The challenge is, the startup founders are going to innovate but don't have distribution. And the incumbents need to innovate faster than the startups can get distribution. And it's kind of a fair fight, right? It's a game we can get out on the field. I have tons of friends that incumbents, you know, and we can all go out and have a lovely dinner and, you know, everything else, like NBA players who hug after a game because they were college roommates, like it's all good. So that's the fair fight. And if you want to make it a little less fair, I think the org design of a brand new company is the most obvious place to do this. Why not have people in whatever department able to write code, you know? Well, we're nervous about, we're nervous about what? Deal with your nervousness. Put the right process in place. Have clogged, triple-check this and put it in front of a human. Have a new engineering role that is in charge of, you know, release engineering for ideas that come from other departments, like experiment, try stuff. But that's principally, I think, the innovator's dilemma, you know, evolution in advantage that we have today. Okay. I wish Christiansen could see.
Speaker 1Amazing. I'll go to one more thread here that I was actually inspired by watching some of the changes R&D is the concept of specialization and role. And let's look at that. I'd like to hear your take on R&D and I'll tell you what I'm seeing on go to market. As you are very well aware, because we live through it, in the 80s, in sales, there was one role. It was salesperson. And they set the meeting, they worked the customer, they renewed the customer, they upsold the customer, and probably triggered by the work of Aaron Ross in predictable revenue coming out of sales force. That went to like a highly specialized environment. We should have an SDR that coal calls and sets the meeting, an account executive that sells the meeting, a CSM that onboards the customer, an AM that renews and expands. And there's always like, we got into too much specialization last decade, like there were a lot of contacts where people did it just because it was invoked, but it wasn't optimal for their contacts, I predict we're going back. Because like, there's always a trade-off like yes with specialization, you can hire the perfect person. But now you have these handoffs that are very inefficient for the customer and the organization. With AI, AI can empower a single individual to handle all those roles. So I've been talking and inspiring a lot of companies to get rid of the concept of an SDR and AM and a CS, it doesn't mean you're getting rid of those people. It just means we're going to a single role. And I was kind of inspired from some of the reason I've been doing an R&D where I'm hearing that they're having PMs do more of the stuff that engineers used to do. Can you speak to whether you think that's going to happen on the R&D side too?
Speaker 2Yeah, the same kind of diamond-shaped thing, and the back half of that diamond, I think already. When I was working with you back in the day in, let's say, 2011-2012, I was the product manager, the support expert, the user researcher, the CSM liaison. I was the go-to-market lead on those products. I was the front-end developer on those products. I was the designer on those products, and all of those, even by, geez, 2016, were individual
Speaker 1teams. Yes, exactly.
Speaker 2And then those things started splitting further, and you had UX copy content writers. And it got kind of, probably, obviously past the point of recent ability, and these small autonomous teams that we wanted to have, all of a sudden, were 7, 8, 9 people. And this happened to everybody, and it turned into I heard it described once as like a quaker meeting for what it's supposed to be three people planning what they're going to do, and the moves they're going to make on this part of the product, again, very customer driven. Now, all of a sudden, you have this huge conference room full of people, and you lose that effectiveness. Now going back to generalist, you absolutely have PMs, not just coding, but doing some pretty sophisticated stuff in their coding. And that, I think, the last six months. I think six months ago, we started to realize, okay, especially early stage. But even later on, the PMs are going to need to have much more detail up front, which is interesting. I didn't kind of see this coming. We were always a little bit kind of run and gone. Be very customer driven, make a decision. It's small. You get it out. You get it live. Back, you're A/B testing, you're gating, you're kind of working that way. And now, there's really a reward for having all of the details of a particular product change that you want to do every error case, like all of the polish around this. If it occurs to you, and you can even say it out loud, that can be in a spec. The cost of writing the code from the right spec is asymptotically approaching zero. Like, you know, the right spec for a thing you want to do with the planning phase and the research phase and everything that we're doing in agent to coding, that spec becomes almost like an iron on decal that you just iron on to the code base. You know what I mean? All the details are there. I think, you know, everybody's awareness of design and taste and design's kind of an interesting topic because everything's sort of converging back to a set of clean based defaults, which is interesting. It's not about doing something new. It's about doing the thing that people really expect the most. And across all of that, you basically have just product generalists in a lot of cases who can do a lot of coding, a lot of design really can get involved and go to market, right? And so that kind of interaction is much more tightly bound and much more organic and much more detailed and ambitious in what it can do. Again, because you can both look at the problem you're trying to solve in the customer base, you don't need an analyst to quantify it. You don't need a designer to mock up a bunch of options. You don't need any of that. You need somebody on the phone and somebody in product to sit together over a coffee and get a lead on something you could do, verify that it's going to have this impact and ship it 45 minutes later. So the way of working is really changing. In our office, we mostly sit around a set of couches looking at the results of the data and looking at the customer feedback and discussing high level what our options are down to the very specific details of how we want to do something. And we may talk four or five of us for three hours. And at that point, the software is written, dude.
Speaker 1But what's your answer on the R&D side? You know, you get what I'm saying?
Speaker 2R&D is not that complicated, I think, because you have people who are very curious and people who are not. Okay. Interestingly, I mean, this is magic wand level stuff and not everybody cares. There are a lot of skeptics and you have to find the people who are authentically really, really curious about this. And across the whole company, I mean, there's nobody at our company that is not using quad code to do most of what they're doing. We had a marketing join recently and the first day he's there at the office Monday morning and I have him install CMux, which is like a total quad code sicko way of using Mac terminal, right? This is like even engineers are like, blah, blah. That's a little much. And I'm like, first thing, go to cmux.dev. Okay, second thing, Cloud Code, you know, auth, boom, boom, boom, third thing, do you understand agents and skills and MCP and you know, all this kind of stuff. And we just start building this rig, you know, by Thursday afternoon, he was deploying to production. - Crazy. - Meaning on our commercial web app, he was deploying to production, had never written a line of code before, had never opened a code editor, actually still hasn't opened a code editor. So that, you know, on the sales side, we were having this conversation literally on the way over here because we are really interested in the people who think the AI stuff is incredibly cool. Like that's the number one thing. I mean, yeah, we sell an AI product, but also just operating. I mean, it doesn't really matter what your business is. The people that you want to hire as a startup are the people who just think all of this clawed code, you know, whatever else, AI, you know, agentech harnesses. The people who think that stuff is unbelievably cool and are already playing with it nights and weekends and then figured out from there. We created a new role that is agent engineer and it's kind of a post sales CSM sales engineering, I don't even know what you would call it. And as these folks come in, they have a clawed code rig where they can do extremely deep analysis using our product as the customer memory layer of the entire install base. We're at about 500 customers right now. And one of the really interesting things here is that the go-to-market roles and the R&G roles are starting to kind of merge. Exactly. And so we have this agent engineer role that is coming from a sales background, super comfortable being on the phone, super personable, polished, business acumen, you know, quota, et cetera. But 80% of their time is in clawed code or in day AI, whatever, and using that customer memory to basically find insights, find pockets, find, you know, oh, we could get the marketing departments into our product. How would I go about that? Let's create a new notification that they can get, you know, wild stuff and they're able to pursue this stuff autonomously and they're coming back and slack with results of AB tests. And it's just like this wild hybrid. But the number one thing for us is just, do you think this is cool? Do you want to be using AI at work and becoming more creative? Are you willing to spend more of your time directly with customers, with your friends at work, having extremely detailed debates that cause the room to sort of stop and stare at the wall for five minutes because you have reached that area of the very hard problems you break through that, then you ship the code, you improve this for customers, like you're unblocked. The all that's left is taste, details, thinking the problem all the way through, being extremely honest about impact and data, you know, understanding your customer-based incredibly well, and the sort of trade craft of putting it live into the world every day gets closer to free. -
Speaker 1Well, man, what a blast. It's so good to see you. I think when people think about, like, I grew up with my siblings, I went to college with my college buddies. I, like, what we went through together is just stories and memories I will bring to my grave. So I'm blessed to have had that opportunity and to have this moment to kind of, you know, remember those times together and think about the future as well. -
Speaker 2It was a lot to go through, and I'm really glad I went through it with you. -
Speaker 1All right, bud. - Mm-hmm, all right. - Awesome. - What is in this coffee? - This coffee is disgusting.
Speaker 2I love you. I love you, Matthew. -
Speaker 1It's coffee. - Yeah. - What kind? -
Speaker 2From when? - It's been coffee. I just made it like. -
Speaker 1I go into 7-Eleven and drink coffee, and I have no problem. -
Speaker 2Yeah. - You're like. - It's not good. (laughs)
PodcastTranscript AI