Three questions RevOps leaders should ask about their company brain

The short answer. A company brain starts as a folder: the strategy, sales process and customer context a team works from, opened in Claude Code. Christopher O'Donnell gives the RevOps leader who owns it three questions for the team. Which number is the AI strategy meant to move? How do you get from baseline to 2x on it? And who controls, and who can see, each agent, skill and data source? He treats every part of the brain, from meeting transcripts to the agent control plane, as a separate build-or-buy decision.

What the video covers

• 5:53 Hard problems now start with a folder of everything you know, opened in Claude Code or Codex. A harness is that folder, and the company brain is the one about how the company makes money.
• 8:32 Sharing a folder through GitHub gives teammates access to it, but no say over who sees what inside it.
• 11:42 Choose the number the AI strategy should move before building anything, for example cutting rep ramp time from seven months to four.
• 15:31 The rep stays human, as the person who represents the company and makes promises. Research, recall and record keeping go to agents.
• 21:18 The alternative to bot-sitting is agents that act before anyone asks, each with a name and a job, reporting to a person who gives them feedback and gets the credit for their work.
• 25:31 The parts of the brain taken one at a time: meeting transcripts, data sources, the agent control plane, harnesses, work surfaces and knowledge management.

Excerpts

[0:00] Listen, the 10 years spent at HubSpot building the thing, that's been talked about a bunch, and also the whole idea of taking the product usage and rotating that into sales, which kind of occurred to us at the same time as a number of companies. Like, whoa, here's an idea. We could have product actually get involved. And that led to us owning a number. […]

[0:34] But there's another story. That's not the whole thing. There's a story that we've never talked about, the origin story of my real relationship with RevOps. So I had been part of the ELT, we called it Helm, for several years, and I kind of took one for the team and was the unified senator representing engineering, product and design for years.

[1:05] And I kind of had my fill of that, and I said, I need to bring a buddy. Somebody else has to kind of do this work with me. So I convinced one of my closest friends, who was head of engineering at the time, now he's CTO, still over there, Whitney, to join ELT. And it was his first or second of these meetings, and Ops was presenting. And in the time-honored tradition, while these nine-hour QBR-type meetings are happening, we spend the whole time Slacking each other, trying to figure out what the hell is going on.

[1:37] He texted me: "Wait a second. This doesn't make any sense. How does it take 90 minutes for a web lead, a form fill, to actually get a rep owner and get rotated to them and be notified?" And I said, "Be very careful what you ask. Be very careful. I will tell you the answer, but then you cannot unknow it." Because of course, the Ops team had a lot of kind of machinery, and they didn't have necessarily an engineering team doing this stuff.

[2:09] And so I told Whitney the answer: this is the deal, and we got to do this enrichment lookup, and then we got to go check for channel conflicts, and we got to do this, and we got to do that. And he's asking these engineering questions. Are they doing FIFO queues? Are they doing this? Are they using Kafka and all the rest of it? And I'm like, see what I said? Be very careful. So once he knew it, though, that was it, right? A year later, we were the proud owners of all of corporate IT, including RevOps, which led to me having an incredible experience managing a guy who's still there, EVP of Ops at HubSpot, Chris Hogan, who is a legend and a gentleman and a scholar.

[2:41] And I had the honor of managing him for two or three years, which, talk about imposter syndrome, but I learned a ton from him. The punchline here is, and this came up in the breakout session, a lot of what's in here came up in the breakout session, the punchline here is you are no longer just RevOps or StratOps or whatever.

[3:15] You are product engineering teams. You take a new role, you find out about 13 engineers that you're responsible for, and what on earth are they doing, and who has decided what's a vendor and what's OpEx and CapEx and all the rest of it? And then you throw AI into this whole situation. So that's kind of where we're going with this today. Good news and bad news. The good news is I intend to illustrate to you why literally your seat of every single person at this event today, genuinely, I believe that your AI strategy represents the number one opportunity to build a multiplier effect, bend the curve, at your company, more than any other department, even more than engineering.

[3:45] The bad news is you're gonna have to take my traditional role of being a product leader, go out, buy the black T-shirt and the New Balance and the whole thing. Robert Barakett has great black T-shirts. They're very durable. You should get them dry-cleaned. Anyway, I love this job, and a lot of the product leader job, especially as you scale and the stakes get high, spoiler, is knowing what questions to ask and then just kind of reading the response, right?

[4:16] So that's gonna be the takeaway today. I wanna give all of you where to look, who to grab, and what to ask them to start a conversation on the whiteboard with. I'll give you some of the answers and some of what to look for and how to think about it, but a lot of it I'm gonna be leaving to you to kind of figure out in that room when you get back to work.

[4:49] So let's start with this idea of a company brain, which is being talked about to death on LinkedIn, on Twitter, Hacker News, wherever you look. What actually is it? What's a harness? What is an agent? Different vendors have different concepts of what agents are, and skills. How does automation fit in? I wanna demystify some of that and kind of zoom out and say, a lot has changed, but the new world is very approachable. Just demystify it entirely for you.

[5:21] Here's the entry point. The way that we solve hard problems in 2026 is very different from how we solved these problems before now. That's the hard part. The easier part is once you learn this pattern, you can use it for any hard problem that you have. I think our most recent marketing campaign that we did was how to solve CRM by solving everything, because this is pattern-based thinking, and what your starting point is, and then you'll kind of find your way like water. You'll find a way to the solution, especially the kind of folks in this room, with your level of expertise and resourcefulness.

[5:53] Anything you're doing, you're planning a wedding, you're building a house, or trying to get a kid into private school, any hard interdisciplinary problem, you start today by making a folder. You create a new place on your laptop, and you go to Codex or probably Claude Code, and you navigate to that folder, and you begin.

[6:27] If you wanna get a kid into a private school, what do you do? You grab all of the target schools' websites. You convert them into markdown. You grab all the report cards, everything. You do a Wispr Flow dump of what you think about your kid, and you put that all in the folder, and then you start to go, okay, let's make sense of this. And you can actually approach that problem with: where should my kid go to school? If you're building a house, you can put in all of the town ordinances, all of the information about how to apply for easements and all this kind of stuff, lead certification, contractor information, reviews from Yelp. Dump it into a folder, and then you begin. And we all know if you do that, you are going to get a spectacular result.

[6:58] So where is the folder at your company that really matters the most? Spoiler, it exists today, I guarantee you, and you own it. So let's go find it. Then how do we approach it like a product leader and own a roadmap for it? All of the little decisions that are being made. Who's gonna maintain this thing? Is this OpEx or CapEx? What do we want it to be? Why did we build this thing? I had one particularly wild thing where I found out that the team at HubSpot that was working on the mobile app was off building a database technology themselves from scratch. It's like, why on earth are you doing this? This is not what we are paying you for. This is not where the dollars are supposed to go into R&D. This is a solved problem. What are you doing?

[8:02] So start here. There are brains scattered everywhere. That sounds a little wrong when I say it that way. Okay, everyone has a brain. That's a little bit more palatable. I still talk to people that I managed years ago, and I'm like, what's your rig? What's going on? Well, my team has this repo, this harness, whatever, and then the VP has one, and then they're trying to do this thing at the company level, blah. How do you fit those all together? I'm spending a lot of time in my personal life just helping people with their AI rigs.

[8:32] By the way, a harness is just a folder. It's a repo. Once you start putting it in GitHub, which I'm sure all of you and your teams have figured out: oh, wait a second, I could take a folder, I can use this GitHub thing to share it with my buddies. Yes. Okay, great. Does that give you permissioning on what's inside it? No. Okay, so there's basically this maturity curve happening of taking ideas from core engineering and applying them, which you've always done in RevOps, to be sure. But now you're gonna have to go get the rest of these concepts and bring them in, and these best practices.

[9:03] So there are all of these harnesses, all of these folders. Many of them are not gonna be in version control. No one's gonna know about them except the owner. I have one just for myself on my laptop. Laura in the back has one just for herself on her laptop. We vibe code whatever we want, right? But everything is connected into the right company strategy on the back end. There is one folder that sort of supersedes all of them. That is the question of how we make money. This is what gives you and your AI strategy an overarching, superseding precedence over everybody else's harnesses and rigs. So let's talk about how we kind of get them to all fit together.

[10:07] I am accustomed mostly to giving presentations to engineering audiences, and so I have a Lord of the Rings reference. We'll see how well it plays in this room. What do these things look like? What's even in one of these harnesses, in these folders? A company brain is, again, just a folder. It's a directory on your hard drive. It is going to have things in there that never existed in a CRM, that may have existed in a company wiki, that may have existed on a laminated, printed-out handout for sales kickoff of "Can everybody please follow the sales process?" And then most of those, like, right by the bar on the way out, in the trash can. You have all the sales kickoff sales process cards.

[10:37] But really nicely detailed examples of the strategy, when you are having your internal ops meetings, and when the company meeting, the CEO is talking about "Here are the big hills we have to climb as a company." The bear case from Morningstar on your company, if you're a public company. Documenting all of this, the same way you would document all of the building restrictions if you're building a house, or the pedagogy of a school you're trying to get your kid into. Basically, just putting all of that in one place, you now have a company brain. Why? Because you can open up Claude Code in that folder and start to have a real conversation about what's going on.

[11:12] So how do we go from good to great? I'm here to tell you how to build the right company brain, what questions to ask your team when you get back. Here's the entry point that I would challenge you with. Think about the number that you have given your team that represents success for your AI strategy in RevOps. What changes in your FP&A, in your financial plan, in your P&L? What changes in quota roll-up, in hiring, headcount plan? I have never asked this question to somebody in RevOps and had them give me an answer.

[12:15] I'm sure there is somebody who says, "I am going to take ramp time from seven months to four months by Q2 of next year." I have not yet met them, and I talk to RevOps folks all day, every day. You cannot improve what you can't measure. This is like classic Peter Drucker management advice. If you look at engineering, what's different in 2026 from 2023? '23, '24, the beginning of '25, you all saw the news stories. Engineering teams are seeing 15% productivity improvements, and then a couple of weeks later, engineering teams are 15% less productive because of AI. And it was all these little puts and takes, right? You remember all those articles? You don't see those articles anymore. Those articles would be ridiculous in 2026. […]

[13:23] Two and a half pull requests a week used to get you promoted. You'd be able to afford the fancy fixed-speed bicycle, so you could come into the office at 11 a.m. and complain about the cold brew. Today, two and a half pull requests a week is ridiculous. Whatever chart is on the wall is going to go up. We're gonna start probably with just signs of life. How many people are doing AI in any regard in go-to-market? That's not gonna be good enough for the people in this room.

[13:55] You get to a point, if you do this right, where the improvements start compounding. Our software engineering rig improves itself today. We pay for tokens and people to solve a problem one time. That problem is solved anytime we see it, forever moving forward. What is the go-to-market equivalent of that? The opportunity you have is that your fellows across go-to-market operations, go-to-market engineering, RevOps, Sales Ops, strategic ops, nobody is doing anything.

[14:25] The chart on the left is from Andreessen Horowitz. Coding, okay, Cursor, there are a million of these, Claude Code. Legal, sure, Harvey, Legora, et cetera. We're starting to see finance come in. We're starting to see IT come in. Sales and marketing even, but sales and RevOps are not yet expecting the kinds of things that you absolutely can get right now for not a lot of money. Anthropic on the right: I disagree with this one a tiny bit, but it's close enough, I think. It's a little hard to see. The red is by department, how much value they've seen realized by each of these departments with AI, and the blue is the potential. I would argue that sales is probably another click out, but their research suggests there's a lot that can be done there and almost no usage. So let's think about this. This also came up in the breakout session.

[15:31] What stays with the human, and what is delegated and given to the AI in the right company brain, as we think of ourselves as product leaders in a revenue operations role? I try to give nice little alliterations on both sides of these to make them memorable. The things that stay, the human things: I believe firmly that the rep stays. I think we're gonna have four archetypes in the future companies. One of them is the representative, the human UI that is going to talk about the company, be credible, make the promises. A lot of people are saying, "We don't wanna buy from AI." I think that's absolutely right. But let's let that person do more of that. Let's let them be more persuasive. Let's let them be better listeners. And again, that credibility: having a person that can actually make a good promise and stand behind it.

[16:03] What goes? Research, recall, record keeping. Trying to find a lead list seems like a solved problem. We have datasets. We have web agents. There are great companies doing these kinds of things, some of whom are in the crowd. Parallel AI is a great company. […] It is a solved problem because you can put agents on it.

[16:35] Recall. With these meeting recorders, one of the first things you see is that you're just able to pay more attention in the meeting, because you're not responsible for remembering all of the details. This is the new expectation. Everything is going to be remembered. Every shred of information and sentiment that matters is going to be captured. And then the record keeping, and kind of don't get me started on this […] That is a solved problem. You should have absolutely perfectly clean data, and nobody should ever have to spend any time on it, and your team should have full governance over all of that.

[17:38] Okay, so this is what the human does. This is what the AI is gonna do. What is the relationship between those things? What is the ergonomic form factor, the day-to-day? What does it look like? I've never seen a team come to me for the first time having gotten this right. There are a million of these kind of mono-agent Slack bots, or a Vercel app that, "Hey, everybody, go ask your questions to Jarvis," or whatever. By the way, they're all named Jarvis. Anybody who has a mono bot named Jarvis, I hate to break it to you, but you're not the first person to come up with that.

[18:41] What are the ergonomics of that kind of handshake? The easiest way for me to illustrate this for you is just a before and after. The before is probably where most of your teams are. This is where engineering was in 2023, for most teams in 2024, for some teams in 2025, and is now a ridiculous world to live in. Humans are bot-sitting. Oh, we got Claude, and then we got the connector into Salesforce. So, hey, everybody, what you're gonna do, this is gonna be amazing, you're gonna love this: instead of entering your stuff into Salesforce, enter it into Claude, and Claude will make sure it goes into Salesforce. Okay? […] Sitting there, and then you end up with like 20 of these things running, and that whole kind of fatigue, that AI burnout that happens. This is not how you wanna live.

[19:46] The second point here is that these things are ad hoc and ephemeral. They go away. They are happening because it occurs to somebody in the moment: oh geez, I should write an email. Oh geez, I should pull a report. The work is done in the thread, again, with the human driving all of that, and then it is gone. If somebody sits there wrestling to pull an accurate, verified Salesforce report for their role, for their team, and they have to try it a bunch of ways and figure out what the fields mean and all the rest of it, the next person who comes along and wants to do that same thing is going to have that same problem.

[20:17] This does not happen in software engineering. If somebody is debugging an issue or optimizing UI rendering performance, and they spend an afternoon on it, everybody gets that for free the next time any task like that comes up, and it goes like this. You are spending human hours only on the things that actually matter, and the knowledge of your overall harness and system compounds over time in wild ways. And the third I talked about a little bit: a faceless void. No names, no personality, no way to respond to something and have it validate you and say, "Okay, you want this in this format every Tuesday, and then you want me to do this and that? Sure, count on it." In engineering, faceless void, totally fine. We're used to it. In go-to-market, it's not good enough.

[21:18] The after here is that the intelligence is absolutely proactive. I have an incredibly cool thing I get every day. I don't think I've shown you this, Mike. I gotta show you this. And it came up on a call. I was talking to a CRO, and he said, "You know, I've always wanted as a CRO to just know the number one thing that would surprise me about my business that I don't know about." […] And I tell my agent off to the side, I'm like, "Do this for me." And so I get these every morning, and they are just wildly interesting and entertaining.

[21:49] And it puts me in a situation where, around the water cooler at work, and my friends who are here can confirm, I'll overhear conversations about deals. And my little party trick, which is getting annoying now, and people are like, "Dude, we get it. Stop." But I'll just start chiming in about the deal […] First couple of times, it played really well. Now it's just really annoying.

[22:19] The self-improving loops. Are you documenting somewhere what's working in a particular vertical, a particular kind of deal or situation? Do you know what is helping reps who ramp faster than they should? What's different about their experience than the reps that never get there and have to get PIPed nine months in? Do you know? This is a solved problem. You should just be able to ask, and then document that in one place, and have everybody's agents pull off of that to coach people and make sure they're on the happy path, and update it as more information comes in.

[22:51] Again, in engineering, we have this. We have self-improving loops. We can auto-resolve a bug when we find it. We can also run a QA bot that goes and gets us the bugs. So we can have a web app or a SaaS product that is just actually getting better 24 hours a day, and we're sitting there, like, putting the credit card in and deciding how hot to run the system. This is absolutely possible in go-to-market.

[23:23] And then again, this idea of agents that people actually form a relationship with, and have more than one. Grok Bot is the closest I've seen to this, where, if you go on Twitter, you'll see examples of this. The Grok Bot sales team, the way they use their own product is extremely cool. They have a coach agent. They have a research agent that's going out and finding leads and coming up with email drafts. They have a data entry agent or a reporting agent: prep me for my forecast meeting, give me a CSV, whatever. And it's a little bit spit and tissue paper and duct tape, but extremely right. Why? Because each of these agents is a long-lived entity that the human can give feedback to, and they have a defined job role that the human who manages that agent can take credit for. And this is a really important concept.

[24:26] If we're not deploying the AI in a way where productivity and credit can accrue through that human org chart, it's gonna be really, really awkward. Because at the end of the day, we are keeping score. In this side of the house, we are keeping score. Who is productive? Who is closing deals? Who has more capacity, right, in the financial plan? Every one of your teams, I imagine, has some form of an engineering team. One particularly wonderful example came up in the breakout session. But even if it's just you using Claude Code, that counts.

[…]

[25:31] Build or buy? Understanding the components within this company brain will help you figure out what you should build and what you should buy. The good news and bad news on this that I observe is that all of you, I think, are on a similar set of thinking. That's the good news. The answers are going to be obvious and will align. I don't think it's a mystery to anybody in this room that meeting recording transcripts are absolutely gold, because they can really tell you what's happening. Whether you're forecasting, coaching, trying to drive rep productivity, whatever it is, they are absolutely gold.

[26:06] Granola, internal meetings: where are you capturing the decisions? You might have workflows where you capture an internal meeting and you have some nice machinery that spits out some action items and all the rest of it. Great. But where is the company strategy? Here are the things we're trying to do as a company. Can we push those out and have everybody's behavior ladder up to that somehow? The data sources. How do you take all of this information and put it into one place? I whiteboard with RevOps practitioners, the folks that many of you are managing, and I whiteboard out the company brain stuff that they have.

[27:10] Very interestingly, traditional CRM is never in the top five most useful, valuable data sources. It is just low resolution. It's not the number one priority. Meeting recordings, Granola, customer Slack, down the list, email. Let's get the Gmail in there. They don't know how to permission that. Fine. All of that stuff in one place, so that you can interact with it in a way, if anybody's used Perplexity, like Perplexity puts the entire web into context. […] So that's the difference in those harnesses and in those data sources. Have a data source that gives you that instant recall and the data privacy, which I'm sure came up in everybody's breakout session as well.

[27:42] The agent control plane. You could build this. You could go get Vercel eve. You could have guardrails and tell people that they can use […] or whatever they want, but here's the MCP, and here is a set of instructions you need to include, or whatever. You have a bunch of options here, but you're going to want to be able to drive behavior change, because so much of our work is gonna be done by agents. How do you decide, here's a detail about, we have a new product, everybody should be looking for attach opportunities in these circumstances? How do you drive that across everybody's agentic rig, while letting them continue to personalize their agent experiences?

[28:48] Harnesses. When do you use a first-party harness? When do you build your own? Who should be using what? Work surface. There was a great conversation in our breakout about this, and different perspectives. There's a perspective that, you know what, I can vibe code whatever UI I want. Which, I mean, listen, I feed my family by making these UIs, or at least I have for the better part of 20 years. I don't know, man. I really don't know. I mean, it's humbling for us. We've spent a lot of time on UI stuff and trying to get everything perfect. And if somebody shows up and they go, "Look what I built," which happens with us all the time, they're like, "Look at this cool thing I built." It's kind of hard to say, "Don't do that." It's pretty cool.

[29:20] Knowledge management. You could do this in Markdown files, or you could buy Glean, or you could do Notion, or you could do this, or you could do that. It's a very interesting open conversation. I think all of these genuinely are open conversations. I have my own feelings on each of them, but look, each one is a separate decision, in a real research project, to think about what you want to do.

[29:52] So here's, as promised, what to walk away with. The first thing is to ask the team, or maybe you all are senior enough, you could actually probably just go tell your teams: what are we measuring? What is success? Whoever on your team is the Claude Code sicko that you're really hoping will get those wins for you, if you promote them in six months, what is the slide they showed you that you then showed to the CEO and said, "Look at how amazing we are with this AI stuff. We're really moving the needle"?

[30:25] Pick a metric. There are metrics in the financial plan that you have never been able to move, that finance has never been able to move, that are now very movable. I'm not in a position to tell you all you should double quota, right? And there's a little bit of a standoff right now where nobody wants to say something that sounds ridiculous. But whatever number is possible for you to move, if somebody told you that number and how much you could move it in reality today, it would seem ridiculous.

[30:57] Second thing: once you have that number, how do we get from baseline to 2x? That's gonna be the biggest lift. […] And then whiteboard for me: who has control of what? Who can see what? These agents, these skills that are running in the background, who owns that?

[31:31] "Oh, well, we have a thing where we push these skills out to people, and we tell them to put it into Claude," and all the rest of it. It's like, get a really good answer for that. And then, what I'm saying to you is, I think your judgment will lead you into the right sub-conversations for whatever your business priorities are, whatever your stack is at the moment, and what you want it to be. If anybody gives you an incredibly good answer for this third one, I would just say, I'm a little out of line saying it, but lock them up and bubble wrap them with a huge equity grant, because that person is an adult and is really thinking about your company deeply. I would promote them and keep them around forever.

[32:03] So that's it. You are no longer just RevOps. You are product engineers. Hopefully, the idea of a company brain makes sense. Hopefully, you have conversations to have, lovingly, with your beloved teams when you get back to the office. And yeah, that's life after CRM.