224: Keith Jones: How OpenAI’s GTM leader structures teams and spots standout candidates

[00:00:00] Keith: since I first started hiring at OpenAI, I open every hiring manager screen this way: "I'm gonna provide you some context on the role, and then I want you to ask me questions." And it is the questions [00:00:11] they ask that ultimately steer the rest of the conversation. [00:00:15] showing your work here's what I did, here's what the model did, that's the differentiation that candidates need to [00:00:21] call out, and say: ChatGPT did not write this. what you are reading right now, are my real human thoughts about who I am as someone who should work at your company." [00:00:30] go-to-market ops is really the heartbeat of understanding the needs of our stakeholders. We're doing, ads now, as everyone knows. enveloped B2B marketing into this org. my mental model around org structure had always been: [00:00:46] be very close to the people who control our budget. put me in finance, put me in IT, like it makes business sense, but my head and heart's always gonna be with go-to-market, [00:00:55] ​ [00:01:22] In This Episode --- [00:01:22] Phil: What's up, everyone? Today, we have the pleasure of sitting down with Keith Jones, head of GTM Systems at OpenAI. In this conversation, we explore how OpenAI hires and ships at such crazy velocity, how to stand out as a candidate when everyone is using AI, how Keith's team is orchestrating agentic code deployment with Symfony, and we'll also talk about what separates GTM ops from GTM systems, how the GTM org moved from decentralized to centralized model, and why the best home for GTM systems at OpenAI is actually under finance. [00:01:53] All that and a bunch more stuff after a quick word from two of our awesome partners, and just a quick disclaimer that Keith is joining the podcast as [00:02:00] Keith the technologist and human, not the employee at OpenAI. The views and opinions he expresses in this episode are his own and do not represent OpenAI. [00:02:07] Sponsor: GrowthBench --- [00:02:07] ​ [00:03:08] Sponsor: GrowthLoop --- [00:03:08] ​ [00:04:11] Phil: Keith, thank you so much for your time again, sir. Appreciate you coming back on. Uh, really excited to chat. [00:04:18] Keith: Excited to be here, Phil. Thanks for having me back. [00:04:20] Phil: So lots of stuff has changed, obviously, in our world, uh, since the last [00:04:26] time we spoke. Like, feels like just, I think, eight or nine months that, that we had you on. But in tech and AI, that probably feels like a couple different decades. Um, you know, there, there's a lot of stuff that we could chat about here. [00:04:38] One of the reasons that you kinda reached out, big thing that was kinda changing in your life is the GTM systems landscape at OpenAI was kinda split up and decentralized, but then you're kinda bringing the team back together. Uh, before we get into that, I wanted to, like, help orient folks back in. I chat with a lot of listeners that are still split between, like, what do we call marketing ops versus rev [00:05:00] ops versus, like, GTM ops? [00:05:02] What are the differences there? So last year, [00:05:04] 1 — What Separates GTM Ops from GTM Systems --- [00:05:04] Phil: you kinda broke it down as, like, GTM ops and GTM systems. Your team owns the GTM systems at OpenAI, and you kinda said GTM ops supports the people directly behind the process design training, um, everything that's like frontline support for the humans in the GTM org. And then the GTM systems side of it obviously working hand-in-hand, but you own the tools and the tech and the back-end technical work. You own salesforce.com, all scaling, all that tool. Is that still, like, a fair definition? Um, w- walk us through that. [00:05:36] Keith: Yeah, I think it's, it's still very fair. and what I would add to sort of, um, kind of just provide a little bit more, like, color and sort of layered context to it, right? Is that, um, you know, our, our friends in, in go-to-market ops are sort of our first line in terms of understanding the needs of our stakeholders, which we share with them, right? [00:05:59] Because [00:06:00] they're truly those business partners to the business and to the leaders of the different sales segments, the different parts of the go-to-market organization, which is morphed and evolved quite a bit in the time since I was last on your show, right? We've since welcomed Denise Dresser as CRO. [00:06:16] We've, um, sort of, um, involved, or I would say enveloped B2B marketing into this org. We're doing, uh, ads now, as everyone knows. So, You know, a lot of things have changed, but the point here being is that because go-to-market ops is really the heartbeat of the go-to-market organization, they're the ones that provide to us really clean requirements what, what we need to do. [00:06:41] And so the line will get a little blurry at times, and probably the best way where even I have a few folks from the go-to-market ops team who will interact and do certain things with the data, uh, even in some of the tooling. And I'm sure we'll get into a little bit more in how we're doing some cool things with Codex and everything, and really starting to blur the lines on [00:07:00] where, you know, ops can play a more direct role in systems, um, while systems still owns the infrastructure, um, if you will. [00:07:08] So, um, we'll get a lot more into it, but, um, yeah, it's, it's that business partnership and, like, lens of here's what the business needs systems, um, that's key to our success and ultimately the success of a good tech stack. [00:07:21] Phil: Gotcha. So marketing ops, rev ops, like are, are those even labels for teams at OpenAI? Like, do they sit under the GTM org? Like are, [00:07:30] Keith: Yeah. Yeah. They-- I mean, yes and no. I mean, it's, uh, there's, there's marketing ops, there's rev ops, go-to-market ops is kind of within that. The, the labels are themselves are, are less important for us today. It's more about the outcomes and the impact, right? Um, but there are several kind of different groups that, you know, we work with to understand the, the needs of the business and then realize that as best we can, um, in the technology that, uh, that we own and deploy. [00:07:57] Um, but, uh, at the end of the day, [00:08:00] it's, you know, put whatever you want on-- whatever label you want on it. There's, there's folks that are doing the job that most of us and most of your audience would expect when you hear the terms marketing ops or rev ops. [00:08:11] Phil: Gotcha. Makes sense. All right. So you-- we kinda teased this out at the start here, but [00:08:16] 2 — The Cost Center Problem That Reunited OpenAI's GTM Systems Team --- [00:08:16] Phil: part of the reason you reached out to, to come back on is there was a big change in the GTM systems org specifically. Uh, the change was from, uh, or you're actually in the process of bringing the, the team back together. [00:08:28] So the GTM org, half the team was split up between-- Or GTM systems, sorry. Half the team was split up between some folks who were on the GTM org still, and then some [00:08:38] folks moved over into finance/IT. You're bringing folks back together under finance/IT, not the GTM org. Walk us through that. Like, uh, it sounds crazy. [00:08:50] Must have been stressful, maybe a bit painful. [00:08:53] Keith: Yeah, it was. And the funny part that we'll get to in a moment is that the, the real reason for why we [00:09:00] are, you know... And, and actually, uh, to clarify, like, we have brought the band back together. Like, that has been completed. So I'll walk through the, the whole journey here, and feel free to, to, you know, poke along and ask me questions to, to clarify. [00:09:12] But, um, the reasons why we ultimately ended up having to do it were, um, just so wildly practical. Um, I'll, I'll get to that in a moment, but to, to, you know, to remind your listeners, right? So- Yeah, we had, um, we, we had a need to start accelerating a number of, uh, really deep integrations, uh, specifically related to downstream financial data coming out of s-Salesforce, right? [00:09:40] And that was what was initially, um, truly kind of predicating, uh, what was originally intended to be a wholesale pull over. But there was a lot of discussion and, uh, somewhat, you know, um, concern around like, uh, you know, are we still gonna be able to honor all of go-to-market's needs if we create distance [00:10:00] away from the go-to-market organization, right? [00:10:03] And so we ultimately settled on a compromise, which was to, you know, distribute the team and peel off the folks who were closer to the things that finance cared the most about then. Um, and they were able to successfully help us do a number of transformational, uh, efforts in regarding, you know, ERP systems integration, billing data, all of that. [00:10:25] Um, but as we started to ex-- uh, learn how to navigate that, right, with the teams in two separate orgs sharing the same surface, a different problem that again is, I'll just, uh, kind of preview, is so wildly practical in nature emerged, which was that we had now two different cost centers in the business that we were reporting off of and drawing resource from and having to, like, negotiate and share, [00:11:00] you know. [00:11:00] Oh, well, you know, okay, this, this, uh, this, this resourcing comes from the go-to-market cost center. This one comes from the... It's actually called Enterprise Platform Technology is the name of the org that we roll into. Um, uh, and that, and that rolls up to our CFO ultimately. Um, but, uh, you know, it was like, okay, how do we, we make sense of this? [00:11:20] And the fact that they're sharing resources, but some of the folks from, like, this consulting firm go to this side, some folks go over here, tooling budgets left, right. Headcount. Headcount was where it really got, you know, 'cause it was like, okay, we have a new chief revenue officer. She's got the incredibly hard but ambitious and wonderful task of sort of reimagining the go-to-market org in her vision that she knows how to, uh, having come from such a, a, an amazing career at Salesforce then Slack, right? [00:11:51] Um, and she's gotta justify a lot of new things to realize that vision and take go-to-market org into the future she's [00:12:00] casted, and I was competing against her needs for more systems resources. And so again, it was just so wildly practical, and it was like, why are we competing against this? And in fact, it was our CRO who didn't offer even a shred of, uh, of detraction from the proposal that if we move over. [00:12:20] It wasn't even a conversation. She was like, "Yeah, go, go." Like I know you'll continue to support us. I know I can hold you accountable, [00:12:27] um, but, like, I can't sit here and try to decide between revenue-generating roles and your roles, which obviously have an impact on revenue, but not as directly as me hiring, uh, more account directors or adding different functional roles that we still need despite the growing agentification of go-to-market, right? [00:12:47] Um, and so it just kinda came down to that practicality of, look, this is just gonna be a lot simpler and cleaner for the business if the group is back together, right? Um, but was-- what was really [00:13:00] wonderful about it was as we brought the band back together, and I'll pause here in case you have any follow-up questions for me, was that we didn't choose to, like, completely, like, rebuild it the way that it was before. [00:13:11] Um, so I actually now have a peer in that group who still leads that, like, quote to cash and, like, um, more of the stuff that's directly tied to our revenue. While I am more focused now on, like, top-of-funnel data enrichment, pre- and post-sale workflow, as well as, uh, having the support systems org still roll up to me. [00:13:30] And so we have a little bit more division of responsibility, which is helping sort of, you know, share the load a little bit, and also creating some, you know, interesting challenges and opportunities on how we sort of make decisions together, right? Um, instead of it being, you know, a particularly kind of narrow pyramid, we've-- we're a little bit flatter, uh, now as an org. [00:13:50] Um, and that's been really interesting. [00:13:52] Phil: Super cool. Yeah, I'm sure a lot of side stories and side quests that we could go down on, uh, during the transition there. But [00:14:00] initially, like my first question is, what happened to GTM ops during the transition? Like, are they still under the GTM org? [00:14:07] Did some of those folks move over so you're s- fully [00:14:10] Keith: No, they, they [00:14:11] all stayed They all... So we, we moved over, [00:14:13] um, and then we still maintained a really tight relationship, and in fact, probably the single most important and productive meeting that I have is this biweekly session, and it is me, it is my fellow lead in go-to-market systems. [00:14:31] It's our manager who runs all of enterprise platform tech, which includes the people systems, supply chain, revenue systems, uh, and a few other, uh, uh, working groups. And then we have on the other side of the table, proverbially speaking, we have the most senior leadership of our stakeholder groups. So we have our most senior leadership from growth, uh, from, uh, go-to-market ops, rev ops, uh, whatever we wanna call it, right? [00:14:53] It's, it's, you know, six to seven of us in a room, right? And we are just talking about priorities. [00:15:00] What, what are the priorities that we need to get done? Uh, has anything shifted? Here's the latest and greatest, right? And we really just like-- we hash through it in a very Organic way. Like, uh, the way I run that meeting is, um, we show up and I say, "Okay, hey, let's spend five to ten minutes jotting down our top of minds, like what is like-- what's really keeping each of us up at night right now in terms of the, the priorities, and then let's figure out where the overlap is, where, where we're a little bit apart, right? [00:15:28] And let's make sure that our teams are working on the right things cross-functionally together and ensuring that we're not actually sort of spidering out, so to speak, if you will." [00:15:38] Phil: Gotcha. Super cool. Yeah, I'm sure there's, there's benefits to having... So like I, I feel like part of your story is we centralized GTM systems back together after having been dist- decentralized, [00:15:51] distributed, but the GTM systems plus ops is still decentralized. So you're still getting like some of the benefits of rolling up [00:16:00] into two different departments, seeing different sides of the org, which you kinda had before at the transition [00:16:06] period, 'cause like that was gonna be one of my questions. [00:16:09] Like you fought to get the team back together in one central GTM systems org, but you had a bit of time to see the decentralized model. Like obviously there were some pains, you were advocating for it, but were there some positives that you were, that you got to kind of experience from having folks under the GTM org and then finance plus IT? [00:16:28] But I feel like the new model is still benefiting from those potential benefits 'cause you still have the GTM ops folks under the GTM org. [00:16:37] Keith: think, you know, I would, I would hinge this on like, you know, this idea of like, you know, challenge and failure can be the best teachers that you can ask for, right? And the reality is like beyond just some of the less sexy practical problems like the funding and resourcing and, and headcount matters, we just-- we ran into some challenges, right? [00:16:56] Like, we had two different orgs with different sets of priorities and, [00:17:00] uh, slightly different ways of working, right, because of that distance. And it really just like made real the challenge that was doing this in a, in a highly distributed fashion, and especially at the velocity that OpenAI moves, right? [00:17:15] I'm still of the, of the opinion that there's a, there's a world where distributed systems can work, but it can only-- it, probably only works really well in businesses where there's a lot more structure. They're, you know, having to move as insanely fast as we have to, given the, uh, constant evolving nature of the technology we bring to, to market. [00:17:37] Um, and so I would say it's not ideal for us. It may be ideal for others. But really the-- what was productive about the time that we spent as a distributed org was that it taught us why that wasn't going to be the most optimal future for our now combined organization. [00:17:58] 3 — Being Close to the Money Means Being Close to the Budget --- [00:17:58] Phil: So folks listening [00:18:00] that are hearing you say GTM Systems has a better home under finance/IT, I feel like part of that is gonna be controversial a little bit. Like, you hear those battles all the time, like where should GTM ops, GTM Systems sit in, in that org? Um, I'm gonna, like, try to tease out what your thesis is here, and I'd love for you to, like, confirm whether that's the case or not. [00:18:23] But is it fair to say that you've picked finance and IT as the home for GTM Systems because you see GTM Systems as, like, the shared revenue infrastructure that needs, like, strong data governance, single source of truth, and GTM is, like, acting as the primary customer rather than the owner of the systems work that you're doing? [00:18:44] Is that part of the thesis? W-what are your thoughts there? [00:18:46] Keith: It's absolutely part of the thesis, but I would expand on it and I would say, uh, and I'll, I'll phrase this somewhat, uh, colloquially, if you will. Um, because I was chatting with a fellow operator shortly after, like, you know, [00:19:00] it was official, like, "Hey, we're actually doing it." Um, and I had said to them that, you know, my baseline mental model around org structure and systems and how best it fits into the org had always been be very close to the money, [00:19:17] right? What was interesting, though, is that I had always, um, codified that in my mental model as be close to generating revenue, [00:19:28] be really close to those stakeholders, be really close to them. And, and we still are, to be clear. But where my model has now been updated is that I'm being close to the dollars, to the bank account, to the people who control our budget, [00:19:44] who see the wider lens of the company and where our expenditures are going, right? [00:19:51] And so it gives me an opportunity to advocate directly to finance on behalf of my cust- [00:20:00] my internal customer, right? You can ima-- I'm sure your listeners can imagine given, you know, just the rapid evolving state of our technology and then our organization continuing to grow, uh, with no, no slowdown in sight, is that, you know, we, yeah, we, we do all the things that any corporate environment does in setting budgets and, uh, forecast and spend and all that. [00:20:22] But when things change so much so quickly, those budgets have to remain in either somewhat of a fluid state or constantly sort of, you know, shifting things around, right? And we had expenditures this year that were certainly not planned when we did budgeting last year. Now, that particularly, you know, um, sort of X factor was Dinesh joined after budgets were set too as well. [00:20:46] So, you know, keep that in mind, right? New, new, new CH- uh, new CRO means new everything, right? But at the same time, I'm able to- To advocate when I get asked questions about, you know, the, the resourcing and the funding, and [00:21:00] even though it still ultimately comes from go-to-market's budget, right? I am their advocate closer to finance and speaking to the strategic finance partners that are then embedded in that organization and with other people so that, you know, I'm sort of providing a little bit of connective tissue, um, and, um, helping kind of facilitate those conversations around new unplanned expenditures that candidly we never could have planned for to begin with, that sucks. [00:21:28] Phil: Gotcha. Super cool. I love it. Like splitting up how you're thinking about being close to the money as opposed to driving revenue and seeing the budget side of things and, [00:21:38] like, financing and, like, two sides of, of the revenue engine, I guess, [00:21:43] if you will. So you said, like, it's, it's finance, but it's also IT. [00:21:47] Finance/IT is kind of the home, right? [00:21:50] Keith: Well, so it's, it's actually a little interesting. Um, so we have two major groups. We have IT still, and that is a separate group that has more of a [00:22:00] security lean to it, right? And is run by our, our, our, our CISO. Um, and they're responsible for global IT infrastructure, right? So things like our access management is owned by them, and obviously there's some democratization where we can control certain things around who gets access to stuff, but the, the infra is built by them, right? [00:22:21] Um, they're also responsible for all of the things that frankly, all of us probably take for granted, right? Networking, hardware, uh, meeting room, you know, things like that, that, that IT, you know, the, the unsung, uh, uh, the unsung her- heroic work, uh, that organizations like that do. And so that's, that's still a separate group, and we stay very, very close to them. [00:22:44] Our group, I would think of it more as like almost a internal like forward deploy business systems org, if you will. Um, so separate from the core IT infra and like access management that has global application across all, you know, tooling, [00:23:00] data, et cetera, et cetera, et cetera. And then we are owning very specific core business applications and third and first-party workflow, uh, that is specific to certain pockets that we own, all with the sort of general bent of, you know, continuing to evolve OpenAI into a, um, a more mature organization as it relates to the hardening of these systems and these data, uh, so that we can also operate in a, in a fluid sort of agentic state too, where, you know, a seller can go to Codex and ask it to do something to its pipeline in Salesforce, right? [00:23:40] Or, or build themselves a little dashboard for managing their pipeline. But we are heavily focused on ensuring that even if they're not using this-- the UI of Salesforce to do something That their permissions are in line, that they can only do certain things. Uh, making sure that we have, you know, a, a complete, uh, [00:24:00] continuously developing list of like public- published actions and, and write permissions through the NCP of Salesforce and other tools as well. [00:24:08] Um, so that's like kind of like where the, the focus sort of shifts, right? IT, infra, uh, b- enterprise platform technology. How do we, how do we mature these business applications so that then go to market and other stakeholders can just do their thing, and we don't have to worry about, you know, them doing something that they shouldn't do, if that makes sense. [00:24:29] Phil: Yeah. Makes, makes a ton of sense why they're kinda like housed, uh, together. So Keith, I, I wanted to like also spend a bit of time asking you about some of the open roles you have, uh, at OpenAI. When I see you posting on LinkedIn, uh, half the time it's like new open roles, exciting roles that are, that are coming up. [00:24:46] Some of them is like contract-based stuff for Salesforce, but you have like new roles opening up. Um, by the time this comes out, like maybe you'll have like a bunch of other ones and, and these ones will be filled. But I wanted to ask you about some of the requirements for the roles [00:24:59] [00:25:00] themselves. Like [00:25:00] 4 — The 2 Profiles Every GTM Systems Team Needs in the Agentic Era --- [00:25:00] Phil: I get why one of the rules for even being considered for some of the roles on the GTM Systems team is that candidates need to have a developer/software engineer background [00:25:11] A lot of the work that you do end up doing is deep developers, Salesforce architects work, right? [00:25:17] Like in the big platform side of things. And if you don't have that background, it might not be as easy for folks. But because of how much nuance and context there is to some of these roles, um, like you have one that is dedicated to just data enrichment and another one that's like lead to opportunity stages. [00:25:34] Why wouldn't you also potentially open these roles up to broader non-technical developer folks that maybe have spent like a decade plus in the Salesforce world as a user, as an admin? Maybe not the developer writing code, but th- they've been integrated to stuff. They've like lived in data enrichment and lead [00:25:56] to, uh, open opportunities. So like, would love to get [00:26:00] your thoughts there. Like why, why wouldn't you o- open that up? [00:26:03] Keith: Yeah. So, well, I mean, the, the reality is I have those roles. [00:26:07] I just don't need them right now. [00:26:09] I hired those people, and they're amazing. They're, they're, um... It's a wonderful privilege to have them in the org and, and, and to manage them today. So my, my, my org today is comprised of two profiles. The-- there's the, what we call the enterprise systems manager role, which is sort of a technical product manager, if you will, and they do ship quite a few things. [00:26:32] Um, and then the other role is that software engineer, systems engineer, developer-type persona. The reality is that I need both, 'cause there are very few people who overlap significantly enough or sufficiently enough To be able to do both sides of that job, right? And so in a rather sort of ideal work stream, right, you have the TPM who, uh, is to your exact, your, your exact [00:27:00] description is ideally someone who, um, has spent years in systems architecture and, and, and might call and refer to themselves as a platform architect or something to that degree with a particular focus area perhaps. [00:27:14] Um, and they are the ones that are meeting with go-to-market operations, and they are the ones who are thinking about the, um, the, the actual like requirements and, and the milestones that we need to hit and planning that out, right? Then they're working in tandem with the more technical staff to inform how we're actually going to do that and hit those milestones. [00:27:36] And so it's not too different from, you know, a lot of PDE orgs where you have product and developer working together, right? Where it gets really blurry in the best way possible is when you start using things like Codex, and I'm sure we'll get more into this, talking about like how we're allowing not, you know, less developer-type personas to ship developer-type [00:28:00] work. [00:28:01] Um, but we need... You know, at the end of the day, this is, you know, kind of a, a simple analogy that I like to use, like when thinking about how we're implementing agents and agentic, uh, orchestration and things like that into our day-to-day execution and technical delivery, is that, uh, entity, if you will, is a intern, [00:28:23] a junior developer, [00:28:25] Phil: Yeah. [00:28:26] Keith: What's the best way to make a junior developer successful? Give them a senior developer to teach them, right? And so that's what my, that's what my, my developer types that I hire are doing. Um, and they are the ones who are looking at the PRs as they're being drafted and, and, and some of them are coming agentically, some of them are coming from the TPM types, right? [00:28:48] But they're the ones who are ensuring the integrity of our systems and ensuring the th- and that things are being held to a certain standard in terms of code quality and things of that nature. Um, and the [00:29:00] reality is that unless you have spent time in a pre-Codex world writing software, writing code in systems like Salesforce, understand some of the implications of that, you're not going to be effective on reviewing something that an agent has written. [00:29:19] But you can, if you haven't had that experience, inform the agent in what it should do, but then you need that second piece of the puzzle, uh, on the backside, if that makes sense. [00:29:30] Phil: Yeah, very cool to hear your, your perspective there on, like, the two roles and you know, like, I, I'm, I'm not looking for those people, Phil, because they, uh, already part of the team and, and, and I have them, [00:29:41] and that, that makes a ton of sense. [00:29:43] Sponsor: Knak --- [00:29:43] ​ [00:30:51] Sponsor: MoEngage --- [00:30:51] ​ [00:31:47] Phil: I feel like you, you kinda teased this out, and I'm, I'm guessing your answer is yes, but, like, those two roles that you kinda talked about, the engineer, the one writing the code, and then the technical product manager that works [00:32:00] on GTM systems, I feel like those two separate roles for the last decade have been very distinct entities, and nowadays they're very much... [00:32:09] The lines between them are, are blurring more and more [00:32:13] with things like [00:32:14] 5 — How OpenAI Orchestrates Agentic Code Deployment with Symphony --- [00:32:14] Phil: you guys just released, like, Symphonic Orchestration. I still butchered the pronunciation of that despite practicing it before recording with you. Uh, generative code orchestration, right? Like, [00:32:25] talk to me about that. Like, what's the translation element to that there? [00:32:28] Like, this is all, like, uh, because we're open sourcing this stuff, right? [00:32:32] Keith: Yeah, 100%. So let's, let's paint this some sort of a practical example of how we would employ it, right? Um, so, you know, there's a precursor to this idea of Symphony that we've, we've open sourced for orchestrating, AI generative code deployment, um, which is harness engineering. Um, I'm not going to be able to speak to it to the nth degree to the technical level, as I'm sure as your audience knows, I'm not the developer type, right? [00:32:56] I fit more into that technical product manager profile. That's who [00:33:00] I am at my core. Um, but, um, to that point, like, you know, harness engineering at its simplest, right, is building the infra that a coding agent needs in order to know where to slot things in and how to do certain things, right? And so we have a harness specific to Salesforce. [00:33:19] But now with Symphony, we're able to set it up so that if we have something on a ticket that is scoped well enough and has enough specificity and enough very tactical requirements, and, you know, we need to do this here and that there, right? Just like you would deliver something to a junior developer where you know they can do the legwork if they have enough guidance and enough prescription. [00:33:44] And we're not needing to go to the nth degree of like, write the code like this, although there are some elements to it that I'll talk about in a moment. Um, but we've got the ticket, we've got the clarity. Awesome. We put a label on it, which sort of kicks off the process so that it can [00:34:00] use the harness and then orchestrate the code and, and start writing it and start drafting a PR that then an engineer reviews and looks at and sees, "Okay, great. [00:34:12] I see what it did here. I see what it did there. Oh, it made a mistake." Common mistake. A junior developer would make that mistake. Again, that's what the agent is, right? And so how do we stop that from happening in the future? The engineer goes back in, writes a new skill, says, "Hey, when you're seeing something like this, don't do this. [00:34:30] Do this instead," right? Roll back the change, run it again. Guess what? It doesn't make that mistake again. [00:34:38] And that means the next time we have a requirement like that that requires the same specificity of skill in order to maintain our standards- We don't have to go back and write the skill again because it's already there, right? [00:34:50] And so the model that we essentially have running at that point is essentially we have our two profiles, right? We have our TPM and we have our engineer, and in [00:35:00] between both of them, we have essentially an army of junior developers that are those agents writing that code that's being informed by the crisp requirements that the TPM has provided, and then constantly being reinforced and upgraded by the engineering staff who understands how to make these agents better over time by continuing to reinforce their practices with those different skills, as well as continue to make, you know, changes to the actual harness itself, as well as Symphony to help just inform that infrastructure, if you will. [00:35:32] Um, and that's, that's the model. Like, that, that's how it works. And it's, um, it's going in the right direction. [00:35:39] Phil: Yeah, it's, uh, it's crazy. So on top of the crazy, like restructuring that your, your team was going through, there was also a technical restructuring from trying to figure out how do we make use of generative code orchestration within the way that we work and between our stakeholders and, and those ticket systems. So like, can you [00:36:00] share, I, I don't know how much you can share of like real use cases, maybe like a hypothetical use, use case. What you just walked us through like 10 years ago, five years ago when you were at, um, some of the other companies, you were at like Mural. Um, like walk us through the difference between you had a stakeholder that wanted to make a change in lead enrichment in Salesforce, they created a ticket, versus what that looks like today. [00:36:26] Like paint us how different that picture is. [00:36:30] Keith: Yeah. So, um, I would say that, you know, in the, you know, world that we existed in, you know, you know, five, 10, 15 years ago, you know, a lot of it kind of came down to really strict domain expertise that was particularly, um, you know, departmentalized within different types of profiles and roles that people would have, right? [00:36:51] Yeah, I would... You know, in, in prior lives, I would get requirements from the business like, "Hey, we wanna, you know, incorporate this new data. We want to, you know, change [00:37:00] the way that this automation runs when we move from stage A to stage B," right? And that's all they would do, is just provide it in somewhat of a, I don't wanna say casual, but in a business context and not a technical [00:37:15] context, right? [00:37:16] And that was my role. My role was to translate the business context into technical context, and then ultimately, because I also managed the teams and prior to, you know, even being a leader, did it myself, right? Although, you know, again, very declaratively because I'm not a developer myself. Um, but then we'd go and, and make those changes, right? [00:37:35] Whereas now today, the real difference is that, you know, there is still that need to translate from business context to technical context But I don't need to go to the nth degree of this is exactly how we're going to do it. We're going to, uh, write this Apex class that does these things and build this flow that does these things because of the harness, because of Symphony. [00:37:59] We can [00:38:00] just say, "Hey, we need to do these things," and it has all the... already has all of the baseline technical context of our standards, how we do things, and what we have today, right? That's a key element here. Like, none of this comes easily. I need to be really clear about that. There is very firm prerequisites in the form of disciplined, uh, operations, especially Dev Ops in this case, right? [00:38:28] Like, you don't get away with this building in production. I'm sorry for any of your listeners who like to do that. Those days are over. Um, at least they should be. That I will refuse to soften my stance on that. Um, but that said, if you have that discipline and have that structure, have a repo, have everything, um, set up that way, then you are on the precipice of being able to implement harness engineering and something like Symfony that we've open sourced so that you can sort of leapfrog [00:39:00] into this more agentic future of executing changes so quickly. [00:39:04] Like I've, I've got contractors who within less than 30 days of being with the business are shipping cha-- the same changes as someone who's been here for two years [00:39:15] Phil: Hmm. [00:39:17] Keith: Think about that. [00:39:18] That is... That's an acceleration in productivity that has not been heard of in a technical landscape before. [00:39:26] Phil: Super cool, Keith. I really appreciate your, your thoughts there. We'll, we'll link out to some of the stuff that, that you shared with me on, on Symfony and, um, yeah, like, uh, the, the pace of change here is, uh, is breathtaking for a lot of [00:39:39] folks. But I did wanna also let you chat about some of the team principles that you have on the job postings. Um, you just talked a lot about the tech and the changes and, and you [00:39:51] spent some time talking about the restructure, but there's, like, [00:39:55] 6 — How to Hire and Ship in a High-Velocity GTM Systems Team --- [00:39:55] Phil: two things that you kinda called out on LinkedIn when you were sharing some of these posts, and the two principles that you shared were fast where it's safe, and that you don't trade chemistry for hiring brilliant jerks. So I wanna spend a bit of time on the first one. So, like, fast where it's safe, slower where it counts. Reversible decisions move fast, irreversible ones take the time they deserve. Walk us through why that is, like, one of the core team principles for you. [00:40:23] Keith: Yeah. And, and to be clear, I didn't invent these. I, I co-op-- I, uh, um, uh, appropriated them from some other leaders in our space who I was particularly inspired by. But the, the principle at, at, at the base of, of them is, is pretty simple Um, if it's a really big decision, if it's something that is going to be difficult to unwind, then you do need to take your time on it because the reality is that if you made the wrong decision or any series of micro decisions within the bigger framework, then you're going to have to do something differently. [00:40:58] But you can't [00:41:00] approach it at the same velocity you would something that you can easily roll back, right? If I wanna ship a, a new field, I'm not going to contribute any ounce of brainpower to that, right? Um, of course, we're gonna do a little bit of, uh, due diligence on, you know, is there any overlap with this? [00:41:19] Do we really need this separate data point? Is this something that even needs to be persisted in our system, right? Maybe it should just be virtualized data from somewhere else, something like that. Um, but if we're talking about like, for example, um, we need to introduce whole new swaths to our data schema to represent new and emerging parts of our business, no, we're not gonna do that quickly because the sheer labor that's going to be required, even with agentic assistance and agentic code development, is going to be significant. [00:41:51] So it's kind of at the end of the day, like, you know, you know that the juice is gonna be worth the squeeze, but you also know that you're gonna have to put some serious [00:42:00] grip strength into that squeeze. So, like, why bother putting that much effort into it, um, if you're going to make such a fast, rapid decision only to find out that you missed something and that you need to clean up the mess that you ultimately created, right? [00:42:15] So it's just a matter of being, uh, really mindful and thinking about what happens if this goes wrong and how quickly can I fix it. That has to inform how much time you spend on the actual decision itself. [00:42:29] Phil: I love it. And the other one was, "We don't trade chemistry or harmony for talent or complexity." Essentially, we don't hire brilliant jerks, and I love that. Uh, y- you also said, like, "We don't chase shiny solutions that aren't durable." Walk us through that one. [00:42:45] Keith: Yeah, for sure. So, you know, on the hiring element, right, we just-- we try to make sure that we're gonna bring people in that we feel confident, you know, whether or not it's with, you know, some particular guidance or coaching, um, that they're gonna work well together as humans, [00:43:00] right? And that, I have to say, I think is increasingly important as a lens to have, right? [00:43:06] Especially as we start to, you know, create more and more pro-productivity through the labor that these agents are doing, right? A lot more work than has ever been done before by the same size of human workforce is able to be done on the back of agents. And so that means now that the people who aren't doing the work that the agents are doing need to be really collaborative, and they need to be able to work well together on that human level, right? [00:43:34] So if you're brilliant at what you do, but you're going to, you know, be a jerk to everybody you work with, then I don't have a place for you in my org. Now, it's okay if, like, working styles, there's a little bit of, you know, friction here and there, and that-- that's, that's humanity, right? Um, my wife and I have, you know, areas where we, where we, we, we, we have friction, right? [00:43:56] But that, that's any pairing of any two or more humans in [00:44:00] any context. Uh, and you have to navigate that, right? So there, there are always gonna be little bumps and things like that. But at the end of the day, I'm not gonna hire someone simply because they're really good at what they do if I don't also think I have a way to integrate them as a human into our human organization, right? [00:44:19] Um, so I think that's, that's a key element. Um, and then I would say kind of on the other side of, like, not chasing shiny solutions, right? It's that, like... And, and this is-- I, I think this is actually a funny, um, uh, a little bit of recency bias on something that we're working through right now, right? But, um, be skeptical of the solution [00:44:38] in a healthy way, right? [00:44:40] Um, we, uh, I won't be able to go into a hyper-specific context here, but suffice it to say that we had an ask from a stakeholder, very reasonable ask, very interesting ask, right? And on paper, in the way that they described it, in the way that they had been informed by a third-party vendor, it seemed [00:45:00] really, really simple, right? [00:45:02] Uh, and even the vendor had gone as far as to say, like, "This should only take about 30 minutes [00:45:06] to [00:45:06] Phil: Oh, yeah. Good, good sales demo. [00:45:09] Keith: Yeah. Wasn't the case. Um, we're like four weeks in. We're not done yet. Um, and, and that's because it turns out it was way more nuanced than that, right? So, uh, I would say it's not just not chasing the shiny, uh, solution, but not getting oversold [00:45:27] by its shininess, not, uh, you know, being healthily skeptical of its shininess, right? [00:45:32] And being like, "Is it really shiny, or am I gonna [00:45:35] rub that off and have nothing but a dull finish underneath of it," right? [00:45:39] Mm-hmm. [00:45:40] Phil: Yeah, healthy skepticism w- with everything in life, I think is, uh, very good advice. Something else that you do in your, your hiring posts, uh, that I love, and I was also a big fan of doing it back in my in-house days when I was hiring, I would love to sneak in a question in the job posting, and it's kind of like a [00:46:00] prompt, and it really gets you a taste of, like, who that person is, how they think about a certain something. But it's also used as a really nice gatekeeping mechanism. Like, i- if I'm only hiring for someone that is good at attention to detail, and if they didn't bother reading through the full job description and they missed the part where I'm highlighting, "You need to answer this question to be considered for this role," and they didn't [00:46:24] bother doing that, like, for me, that's, like, at least a yellow flag. [00:46:28] Maybe they're, [00:46:29] like, super deep in the hiring process and, you know, you're just one of, like, 30 applications that day You know, everyone's got different stuff going on in, in their lives, and I've chatted with plenty of [00:46:39] folks that have been on the job hunt forever. But you're-- Like, [00:46:43] 7 — How to Stand Out in Job Applications When Everyone Is Using the Same AI Tools --- [00:46:43] Phil: one of the hiring prompts that you're asking folks is, um, like, one of the most recent one was, like, your most pointed, controversial, forward-looking AI native opinion on GTM tech stacks. How many applications do you look through that don't even bother answering the hiring prompts that you put in there? Is there a lot? Is-- 'Cause it's open AI, is, uh, does everyone do it? What, what, what's the reality? [00:47:07] Keith: I would say most of it. I, I, I respect that there is a certain amount of gravitational pull with the brand equity that sort of commands that, right? Um, but it's a really interesting differentiator between, uh, you know, the candidacy of the different people who are reaching out, right? Um, and, and the best part is, like, it's sort of a semantic filter as I'm like, just sort of, you know, chewing through my inbox, right? [00:47:34] If someone just says to me like, "Hey Keith, I'm really interested in the role. I do X, I do Y, I do Z." Like, okay, great. I really appreciate that you're interested in the role, all of that. Like I do with the bottom of my heart, right? I'm a, I'm a human first leader, so I really, really do appreciate that. But if your message starts with hot take, hyphen, and then your hot take, I'm reading that message [00:47:55] Phil: Hmm. [00:47:56] Keith: l- more, way more likely than I am reading the, [00:48:00] "Hi, my name is so and so. [00:48:01] I do X, Y, and Z," right? And that's because again, to your point though, like I did call that out specifically in the post and it, it is a litmus test to a degree, right? Um, but I'll also share something that I do additionally, kind of just a couple steps further downstream of that, right? That is also a particularly helpful thing for me in my hiring. [00:48:21] I've been doing this, um, almost since I first hired, um, first started hiring at OpenAI, um, is I open every hiring manager screen this way. And I say, "All right. Hey, here's how I'm going to structure this conversation. I'm gonna provide you some context on the role, why we're hiring it, why now, a little bit about the team, uh, that sort of thing. [00:48:43] Um, and then I want you to ask me questions based on that context," right? Um, and it is the questions [00:48:50] they ask that ultimately steer the rest of the conversation, right? Now I always have a, a series of, you know, uh, things that I'm screening for through more conventional interview [00:49:00] questions and things like that, but I will adapt those questions and more specifically or more often I should say, the way I phrase them [00:49:08] based on the questions they ask me in light of the context I've just provided. [00:49:14] Phil: Very cool. I love that. Letting people kinda navigate or steer, like, where the interview is gonna go based on their curiosity, where their brain goes when you're giving [00:49:23] out that context. Yeah. [00:49:25] That's really cool. You're, you're probably getting a lot of insight just based on the questions themselves, and then you can kinda skip to the, the meat of the interview. [00:49:34] Like, sometimes that's what I've found, like, [00:49:36] one hour hiring screen, and then, like, by the time you're finishing giving them context and asking them, like, the, the classic first couple of questions, you're, like, 40 minutes, and you're just like, "I don't-- I didn't get anything from this person yet." Um, but I think what I love the most about your hiring prompt is that it, it's helping... [00:49:56] It's forcing people to, like, differentiate themselves from, [00:50:00] like, I don't wanna call it AI slop, but, like, one thing that a lot of people are struggling with right now, especially for remote roles, is like, "Dude, there's, like, 936 people that applied for this opening, and it's been up for 24 hours. And what I'm doing to apply to it, I took the job posting and my resume, and I asked ChatGPT to, like, personalize it and tweak [00:50:21] stuff, and so did 900 other people that applied for that role. [00:50:26] And so ev- everyone's cover letter looks the same. Everyone's job titles and description looks [00:50:31] the same. How can I stand out from that crowd of folks?" I think you're, like, forcing people to do that. Do you think even when you're, like, hiring for roles that don't ask you for, like, that hashtag hot take or hashtag that was fun, like, how do people stand out from, like, the sea of applications right now and everyone using similar AI tools and sounding the same? What do you... What advice do you have for candidates on, like, standing out from the crowd? [00:50:56] Keith: Well, I think, I think first and foremost, um, [00:51:00] we, we have to remind ourselves that we're representing ourselves and not the agentic version of ourselves, right? Um, you know, I'm particularly optimistic about whatever the evolving world of work looks like with, you know, agentic productivity at its heartbeat, that there's an increasing need to differentiate ourselves as humans, as you know, where our capabilities as humans with real reasoning, um, you know, um, experience, uh, set us apart from what agents can do, and then what sets us apart from our fellow humans, right? [00:51:34] And so, um, you know, I kind of-- I, I really like, for example, in fact, this is sort of a thought that's occurring to me in real time, so apologies in advance to your audience if it's not perfectly polished. But I recall way back in the day, pre-ChatGPT, pre-LLM, you know, fully deterministic, if this, then that, uh, logic, right, in our business systems. [00:51:55] But we, [00:51:56] um, we had this little, uh, sort of fork in [00:52:00] our inbound automation at a company I worked at where we made a point to say, uh, when someone signed up for like a free trial after hours or on the weekend when our staff typically wasn't gonna be working, we'd still send them an email in relative, you know, near real time, and it would come like, you know, say from Phil, right, uh, at the company. [00:52:21] But we would put in the message Um, like, "Hey, Keith, this is Bill from company XYZ. Here's the thing, I didn't write this email. This is fully automated. It's being sent on my behalf because you signed on, uh, when I'm typically not online. Um, I can definitely accommodate your schedule in the future if this is the best time when, when you need to, uh, to work with us. [00:52:43] But, uh, right now, this is a fully automated message." Right? I'm raising this because I think there's an opportunity for clients to differentiate themselves and to explain what they did and what they did with the technology. Right? And I think showing [00:53:00] that, that, that-- showing your work in that sense of here's what I did, here's what the model did, here's what I had the model do, here's what I did based on the model's output, I think that's the differentiation that candidates need to, to capture and, and be explicit about it, right? [00:53:18] Call out and say, maybe in your cover letter, "ChatGPT did not write this. Thought did not write this," right? Like, "I wrote this. There are parts of my resume that have been streamlined by the models, great, awesome. But what you are reading right now, if you are reading this, are my real human thoughts on paper about who I am as someone who should work at your company." [00:53:39] Right? I think that sort of explicit call-out, um, and then being able to show your work and how you augment your output with the assistance of these models is the way it's different James, though. [00:53:51] Phil: Hashtag hot take. I wrote this cover letter and this DM, Keith. That, that is my hot take. But yeah, just for [00:54:00] fun, like, uh, what are the most memorable hot takes of, like, pointed, controversial, forward-looking AI native opinions on GTM tech stacks that you've received in your inbox, uh, like DMs from, from folks, uh, looking at some of those roles? Are there any that, like, jump out that you remember being like, "Oh, damn, this is, this is spicy"? [00:54:21] Keith: Um, nothing, uh, like super spicy comes to mind, but there were a slew of ones and people who definitely earned, you know, time, you know, in the interview process as a direct result of it, who I'd like to think were, a a little bit ahead of some of the other machinations that we're seeing, uh, in the industry. [00:54:41] For example, you know, there's this, you know, new emerging marketing from our, our friends at Salesforce about Headless three sixty or, or Headless CRM, right? We were already moving in that direction, right? Where, uh, to some of my earlier comments about how in enterprise platform technology, we are really [00:55:00] focused on hardening the systems and making them, you know, in, in lack of full technical jargon, more API ready, right? [00:55:08] Um, so that people can interact with them conversationally with the assistance of the models, right? There were a lot, uh, not-- sorry, not a lot, but there were, there were a few people who called that, like, hot take UX doesn't matter. Data matters. Hardening matters. Something to that degree, right? There were a few folks in my inbox who called that out and earned time in the interview process because of it, because they were not looking at the problem as it exists today. [00:55:36] They were looking at the solution that needs to exist tomorrow, even if we're not fully there just yet, if you will. [00:55:44] Phil: Super cool, Keith. I got a couple last questions for you. Uh, really appreciate your time. Um, y- y- like, obviously, [00:55:51] 8 — What Building at OpenAI's Velocity Teaches You About Decision-Making --- [00:55:51] Phil: you're at one of the coolest companies on this planet, and, like, you've built a career kind of at the intersection of revenue and systems and MarTech and RevTech. That vantage point is really rare for a lot of folks. [00:56:05] Like, people look at your career, and they're just like, "Man," like, "that's, that's crazy." Um, throwing flowers at you here, Keith, but well-deserved. W- what does, like, being specifically at OpenAI teach you about that intersection that you don't think you could have really learned anywhere else right now? [00:56:23] Keith: You know, I think the answer to your question, Phil, is the same thing that I found myself saying to my newer colleagues as they were joining the organization back in uh, late '23, early 2024, after I'd kinda gotten, you know, my, my feet wet, so to speak. Um, actually that's not true. I dove right in, so I was soaked head to toe, right? [00:56:42] Um, but, um, I, I remember saying to several folks, uh, it is the thing that you will find that you have to figure out how to do for yourself as fast as possible, and to the earlier kind of ideas of team principles and decision-making, was figure out [00:57:00] how to index really quickly on what is the thing that you know you can do right away or the thing that you should say yes to right away versus the thing that you have to pump the brakes on and work your way through more diligently or more, uh, um, decisively, if you will. [00:57:15] Um, I don't think I would have the same sort of improved baseline operating model for doing that if it weren't for this very unique environment that is OpenAI that has accelerated so quickly and grown like bang busters, just left and right, up and down, right? Like when I came to OpenAI, it was less than 100 people in go-to-market, and now there's close to 1,000 of us, right, in that org. [00:57:42] Um, I say honestly, You can tell I'm still not like fully mentally separated from go-to-market. I never will be, right? Like put me in finance, put me in IT, like it makes business sense, but my head's always-- my head and heart's always gonna be with go-to-market, [00:57:54] Phil: best friends for life. [00:57:55] Keith: especially as a former salesperson myself. [00:57:57] Like I'll never not be that. Um, it's [00:58:00] integral to who I am. But again, at the end of the day, like I would not have been able to gain that, um, that experience of having to make so many decisions So quickly while also figuring out what decisions I can't make as quickly. Um, and frankly, I-- uh, it's been improved because I made some mistakes, right? [00:58:19] There are decisions I made so quickly that were not the right ones. Turns out should have slowed down and thought about that a little bit more before just yeeting that Slack message or making that decision. Um, and so I think that's the thing that I'm taking away from my time at OpenAI so far that I don't know that I could have gotten anywhere else if, if not for anything other than the sheer velocity and volume that I've experienced in the [00:58:42] time here. [00:58:43] Phil: I love it. Love the self-reflection there, Keith. [00:58:46] 9 — How an Endurance Athlete Mindset Applies to Cognitive Work --- [00:58:46] Phil: You're obviously a GTM systems leader and advisor, well-traveled speaker. Uh, last year when you came on the show, you also shared that you're a Lego fanatic, a cyclist, and a TV show buff, as well as a new runner. Maybe not so new anymore. Maybe you've, you've kept that up in a year. But as you know, we ask the same question about happiness and balance to everyone that comes on the show. Like what's your personal system for staying aligned with what actually makes you happy? Last year, your answer was that fulfillment requires problems that challenge your brain and time that protects your energy. [00:59:16] You kinda need both of those. Is there anything you would update in that answer? What are your thoughts there? [00:59:21] Keith: Yeah. So I mean, all of that is still true for me today as it was a year ago. I would say where I put a little bit more practicality into it is that, you know, um, the cognitive load that especially my role at OpenAI requires is not one that can be sustained for endless hours on end. [00:59:39] Um, and so I have to be really decisive about pulling away from work at a reasonable hour. [00:59:47] Uh, and I know this may not be as popular as some folks who maybe operate really well in that endless hustle, working twelve plus hours a day. That's not me. I'm an endurance athlete, which means that I know when I need [01:00:00] to recover. I know that, for example, on my bike when I'm in zone five, zone six, in regards to the amount of power that I'm generating on that bike, those are things that can only be done for a short period of time before physically my body will shut down, [01:00:15] right? [01:00:16] And while the mind is definitely more stronger than the physicality of the body, the body is still an important factor, right? I get tired, I get hungry, I get sleepy, and so I have to pull away from work so that I can do things outside of work, um, that still offer me, you know, good, good things for my brain, good problems to solve, whether I'm trying to bike code some new fitness app or, or do something else. [01:00:41] But I will physically disconnect myself from work. That's key for me. Again, not trying to be prescriptive for, for you, your audience or anyone else. But for me, what that means is when I decide I'm done working, the monitor is turned off, the keyboard is unplugged, the keyboard is put away, [01:01:00] the laptop is de-- disconnected from the monitor. [01:01:03] I have to create some friction there And, and the reality is I'm better for it because what's gonna happen when I come back a, a few hours later, the next morning, uh, you know, maybe, uh, when Monday rolls around, if it's the weekend, right? I'm gonna be fresher for it and have whatever maybe I did outside of work now inform what I'm doing there. [01:01:22] Um, but all the same with a little bit of extra practicality in terms of how I draw that separation. [01:01:29] Phil: I love it. Put the keyboard away, lock it away in a drawer. The combination doesn't receive to your inbox until Monday or the next day. Whatever you can to force your brain to, to shut off there, I think is, uh, super cool. Are you actually vibe coding a, a finance, uh, or f- fitness app you said? [01:01:46] Keith: I'm trying, yeah. I haven't, I haven't, I haven't found something that I like just yet. I, I'm kind of going in a bunch of different directions while still experimenting with different apps that people have across the board. I've got my Garmin, I've got my Apple Watch. I'm, I'm, [01:02:00] I'm experimenting, put it. that way. [01:02:02] Phil: Awesome. Keith, really appreciate your time, [01:02:04] man. This is super fun. Thank you so much for coming back on. Really appreciate it. [01:02:09] Keith: Happy to be here, Bill. Thank you so much for having me. [01:02:10]