PostHog: Pivots Were The Real Lesson In Building A Startup

1. 0:00 – 0:46

   Intro

   1. SPSpeaker0:00

      [upbeat music] Today I'm joined by James Hawkins, the co-founder of PostHog from Winter '20. Um, PostHog started as product, open source product analytics, but over time it becomes so much bigger, to the point that now it can even build a product, your product for you while you sleep. Isn't that right?

   2. JHJames Hawkins0:23

      Uh, yep. I would say.

   3. SPSpeaker0:25

      [laughs] Welcome, James.

   4. JHJames Hawkins0:26

      Uh, thanks for having me.

   5. SPSpeaker0:26

      Great to talk with you today. That's awesome. Uh, you did YC in 2020, but you didn't start with PostHog as it is today, right? I heard about a story about, like, five pivots?

   6. JHJames Hawkins0:36

      Yeah, we had-

   7. SPSpeaker0:37

      How did you-

   8. JHJames Hawkins0:37

      PostHog was the sixth, I think, that we worked on. So-

   9. SPSpeaker0:40

      The sixth company in a row

   10. JHJames Hawkins0:40

       ... five pivots. Yeah. Um-

   11. SPSpeaker0:42

       And, and how long did you spend, like, in that pivot hell before?

2. 0:46 – 2:13

   The Rule: Sleep on It Once

   1. JHJames Hawkins0:46

      Not as long as it sounds like it could have been. Uh, I think we, uh, probably spent in total maybe seven or eight months.

   2. SPSpeaker0:52

      Okay.

   3. JHJames Hawkins0:52

      So we're kind of averaging maybe, like, a month and a bit per idea. We didn't give each idea very long before we moved on.

   4. SPSpeaker0:58

      And, like, and you didn't give up. Like, it's kind of, like, impressive in a, in a way.

   5. JHJames Hawkins1:02

      I, I kind of f- feel like at the, in the early days we weren't thinking about scaling or getting close to revenue. We were just like, "Well, we need to build a machine that can kind of learn." [laughs]

   6. SPSpeaker1:09

      Mm-hmm.

   7. JHJames Hawkins1:10

      Um, like, we just need, like, a single customer and so on. So we'd, we'd fail at various stages, like no one wants to talk to us about this idea, no one wants to use the software that we're building, no one wants to pay for it. There were different hurdles that we kind of stumbled through, but we-

   8. SPSpeaker1:22

      Each time there was a reason to give up and move on.

   9. JHJames Hawkins1:24

      Yeah, basically. Or, like, we would kind of get stuck somewhere or, like, some ideas we didn't like that we worked on. Um, others just felt unnecessarily hard. But we did, I would say we were very, we would really commit to each... We wouldn't spend kind of a couple of weeks existentially debating what to work on.

   10. SPSpeaker1:38

       So you would, you would pick an idea, go full on that idea.

   11. JHJames Hawkins1:42

       Yeah.

   12. SPSpeaker1:42

       Like burn the ships, do only that.

   13. JHJames Hawkins1:45

       Yeah.

   14. SPSpeaker1:45

       And then after some time you realize it's not going to work.

   15. JHJames Hawkins1:47

       Yeah, our only rule was if we felt it was something wasn't, wasn't working, we would make ourselves sleep on it for one night, and then if we still felt like that the next day, we would just start afresh.

   16. SPSpeaker1:55

       Okay.

   17. JHJames Hawkins1:55

       Um, I think it was because we were quite, Tim and I, Tim, my co-founder, and I were quite bad at product, and so we hadn't read The Mom Test. We didn't really know what we were doing very well.

   18. SPSpeaker2:02

       [laughs]

   19. JHJames Hawkins2:03

       And so we kind of learnt by... We, I think we, we, in practice for us, we got much more information by just getting stuff into people's hands quickly versus, uh, intellectualizing it a little bit too much.

   20. SPSpeaker2:12

       Yeah.

   21. JHJames Hawkins2:12

       Um-

3. 2:13 – 2:31

   Why Pivots Were the Real Education

   1. SPSpeaker2:13

      And in a way it feels like, uh, you're going to tell us, but that you probably would never have found the last idea, the one that worked without going through that journey.

   2. JHJames Hawkins2:20

      Yeah, and there are definitely other ones we worked on before that, like, maybe would've worked if we'd persisted, but I just don't think would have grown to the, anywhere near the same extent-

   3. SPSpeaker2:27

      Mm-hmm

   4. JHJames Hawkins2:27

      ... or has been as fun to work on, or we would have got bored and sold or something. So,

4. 2:31 – 5:14

   Birth of Open Source Analytics

   1. JHJames Hawkins2:31

      um-

   2. SPSpeaker2:31

      How'd you end up on a open source product analytics?

   3. JHJames Hawkins2:33

      So we, [laughs] in the process of trying to do all these ideas, um, every time we'd create a new product we'd have, like, a new website, new-

   4. SPSpeaker2:39

      Mm-hmm

   5. JHJames Hawkins2:40

      ... like, login page, all this stuff. Um, and we just felt, we got frustrated at using the existing tools. There are tons of tools already that would help you understand user behavior, but we were getting annoy... We were just getting frustrated at the process of buying and implementing them. Um, my co-founder is very technical. I'm a bit technical.

   6. SPSpeaker2:58

      Mm-hmm.

   7. JHJames Hawkins2:58

      Um, and we're both united in the front that we just wanted to have, like, pricing on the pricing page. We just want to put a credit card in, start using it. We don't want to talk to sales team. So you can kind of consider these sort of brand things really, but, um-

   8. SPSpeaker3:09

      So-

   9. JHJames Hawkins3:09

      We just want something more developer-y, basically

   10. SPSpeaker3:11

       ... your background is more, like, uh, is, like, more sales.

   11. JHJames Hawkins3:14

       Yeah.

   12. SPSpeaker3:14

       And despite that you didn't want to have any-

   13. JHJames Hawkins3:16

       So-

   14. SPSpeaker3:17

       ... thing to, to do with sales?

   15. JHJames Hawkins3:18

       Uh, so we, I, we, some of the early ideas we worked on were for sales. So I'd been a VP of sales before.

   16. SPSpeaker3:23

       Okay.

   17. JHJames Hawkins3:23

       I used to sell compliance software to large banks for, uh, millions of dollars a year. Um-

   18. SPSpeaker3:28

       That was kind of like you got vaccinated against [laughs]

   19. JHJames Hawkins3:30

       Yeah, and it kind of grass is always greener [laughs] on the other side. So, uh, yeah, no, honestly, it was a choice. Like, we, when we st- That past has haunted me multiple times. So we, um... I really learnt by doing that how hard it is running a pure enterprise-

   20. SPSpeaker3:45

       Mm-hmm

   21. JHJames Hawkins3:46

       ... um, company where no deals would happen for a couple of months, then you would double your revenue in one go. Um, just inherently made it super hard to scale because, you know, every investor meeting will be about, "Well, what about this deal, this deal, this deal?" Rather than, "How do you go build the future," basically.

   22. SPSpeaker4:02

       Mm-hmm.

   23. JHJames Hawkins4:02

       So we made loads and loads of mistakes. Um, and the thing I found kind of from working in a company that had a really, like, had a succe- like, a good exit in the end, but not kind of like a true venture scale, big, big, big outcome. Um, but kind of the combination for me that was very useful was going and working somewhere that struggled and we really had to try, we just had to try and stupidly hard to get anything sold there.

   24. SPSpeaker4:22

       Yeah.

   25. JHJames Hawkins4:23

       Um, like, we would do maybe, like, 50 meetings plus to get a deal over the line. Some of our deals took four years. [laughs] Some of them-

   26. SPSpeaker4:30

       Wow. But, but-

   27. JHJames Hawkins4:30

       The minimum was probably about two years

   28. SPSpeaker4:32

       ... like, big, big deal size or whatever?

   29. JHJames Hawkins4:33

       Uh, yeah, up to, like, 5 million euros a year kind of size. Um, so it's, like, the biggest contract-

   30. SPSpeaker4:37

       Okay
5. 5:14 – 5:49

   How They Knew This Idea Was Different

   1. JHJames Hawkins5:14

      stuff.

   2. SPSpeaker5:14

      And so how did you know, uh, this idea was right, and how soon after pivoting?

   3. JHJames Hawkins5:20

      Um, I think there were signs. So we, um... Of all the ideas we worked on, it was the one that we had the most fun building the early version on-

   4. SPSpeaker5:28

      Okay

   5. JHJames Hawkins5:28

      ... I would say. Um, uh, so I think that was kind of one, and it was very clear.

   6. SPSpeaker5:32

      There were already, there were already product analytic software out there.

   7. JHJames Hawkins5:35

      Yeah, there was already... Kind of like Amplitude was super successful. I don't know what a- amount of revenue they had, but I'm pretty sure they'd been over 100 million when we started, I would guess.

   8. SPSpeaker5:44

      Okay.

   9. JHJames Hawkins5:44

      Um, there was Heap, uh, Mixpanel, a bunch of others. [laughs]

6. 5:49 – 7:17

   Competing Against Amplitude & Heap

   1. SPSpeaker5:49

      And so your differentiation was what? Like open source was the big deal?

   2. JHJames Hawkins5:52

      Yeah. So, um, open source, and we believed that, so you could run the whole thing, in the early days you could run the whole thing in your infrastructure.

   3. SPSpeaker5:59

      Mm-hmm.

   4. JHJames Hawkins6:00

      Um, so all your data would stay on your side. Um, so we kind of believed originally that we're gonna build something a bit like GitLab- ... compared to GitHub where-

   5. SPSpeaker6:06

      Oh, and didn't, uh, play out that way [laughs]

   6. JHJames Hawkins6:08

      No, and then we started doing it and I was like, "Oh, m- man, like, I'm gonna end up doing enterprise sales again." [laughs]

   7. SPSpeaker6:13

      Yes. [laughs]

   8. JHJames Hawkins6:14

      So, um, so Tim and I were honestly, we were like, "We're gonna end up having to do really top-end deals to huge banks. They're gonna buy our-

   9. SPSpeaker6:19

      It's like the open core play where you have the enterprise version of the open source software.

   10. JHJames Hawkins6:22

       Yeah, and it worked. Like, we got demand for this, we got... Like, we started making a bunch of revenue. Um, but we kind of realized we're selling to the, the most conservative companies that are slow-moving, that haven't moved, don't really want to move to cloud. Everyone was asking for this stuff. But then we also realized that it was really hard to run our software on other people's infrastructure. Um, we've often got an order of magnitude more data than their production database would have because we're-

   11. SPSpeaker6:48

       Mm

   12. JHJames Hawkins6:48

       ... collecting kind of every click, every page-

   13. SPSpeaker6:49

       Yes

   14. JHJames Hawkins6:49

       ... view, whatever. Um, and we built some features that would do that automatically, but it kind of meant that we're always an order of magnitude outside of the competence of the engineering or the DevOps team that our customers would have. And so the reality of running this business was, like it's growing, growing pretty well, and we're basically debugging other people's Kubernetes over Slack. [laughs]

   15. SPSpeaker7:07

       Oh.

   16. JHJames Hawkins7:07

       Um, or from like literally, like we had one customer who, like, took a screenshot of their AWS account on their phone and then sent it to us over Slack through a Gmail account-

   17. SPSpeaker7:15

       Sure

   18. JHJames Hawkins7:15

       ... because they can't, they haven't got permission to let us

7. 7:17 – 8:57

   The On-Prem Nightmare

   1. JHJames Hawkins7:17

      do anything.

   2. SPSpeaker7:17

      So, so most people are like, like you don't do, uh, on-prem anymore, right?

   3. JHJames Hawkins7:21

      No.

   4. SPSpeaker7:21

      And so h- how did that transition happen? Like-

   5. JHJames Hawkins7:23

      Far too slowly [laughs] is the reality. Uh-

   6. SPSpeaker7:25

      How do you, uh, convince these big enterprise to use the cloud version?

   7. JHJames Hawkins7:28

      Uh, what we kind of noticed w- was we had a whole bunch of customers who were running us on their infrastructure for no reason. So we had a bunch that were using us for good reason, and a bunch who were just, were just like, "What are you doing?" [laughs] Like-

   8. SPSpeaker7:37

      Yeah

   9. JHJames Hawkins7:37

      ... and then we were like, "Oh, we should have a cloud product to, so we can run it for them because they're having a rough time hosting PostHog." And so that was actually the problem we ended up solving was, oh, we've shipped this open source thing, bunch of people hate running it. Let's build a kind of nice, invent a cloud product for them. And so people ended up not really comparing Post because at this point, PostHog's c- cloud product was just objectively worse compared to the competition. But people, our customers were kind of just asking themselves like, "I want to use PostHog because I'm a developer. Do I pick open source or do I pick cloud?"

   10. SPSpeaker8:04

       Mm.

   11. JHJames Hawkins8:04

       And we realized, we're like, "Okay, let's just make the cloud product" And I, we were also getting frustrated seeing people saying like, "PostHog's buggy, it doesn't work very well-"

   12. SPSpeaker8:12

       Yeah

   13. JHJames Hawkins8:12

       ... "blah, blah," because they're self-hosting, and so they have issues with their infrastructure mostly. Um-

   14. SPSpeaker8:15

       At least, uh, it's in your control to make it work well and you're-

   15. JHJames Hawkins8:18

       Yeah, we couldn't do anything often

   16. SPSpeaker8:19

       ... only one in five to monitor [laughs] not every time.

   17. JHJames Hawkins8:21

       Yeah. We're like, "Man, we know it's free software and people are complaining comparing it to cloud products."

   18. SPSpeaker8:25

       Yeah.

   19. JHJames Hawkins8:25

       And so we kind of thought, "Oh, it's better for everyone, for us, for, and for our customers and for our marketing if, um, we control the experience." So we said, "Okay, we'll make the cloud product, um, we're competing with our open source product really, and so we'll make the cloud product cheaper." And the way to do that is like, well, they have to run their own infrastructure, and we'll put a big free tier on the product. Um, because we're like, "Well, if you have no money, there's no point us trying to get it from you." [laughs] Um, so we should just make it free for people who don't have product market fit.

   20. SPSpeaker8:51

       Okay.

   21. JHJames Hawkins8:51

       And so we kind of set it all on... We looked at, like, usage-based pricing. We tried to set everything so that, um, there's no kind of

8. 8:57 – 10:33

   Going Cloud & Killing the Friction

   1. JHJames Hawkins8:57

      friction.

   2. SPSpeaker8:57

      So it's very easy for any new project, new company to, uh-

   3. JHJames Hawkins9:01

      Yeah

   4. SPSpeaker9:01

      ... just add up PostHog.

   5. JHJames Hawkins9:02

      And we made it all self-serve as well, which are, at the time none of our competitors were doing very well. And so, um, and that just led to absolutely not capturing all the value we're providing, but, um, we're kind of making it up in terms of numbers basically.

   6. SPSpeaker9:13

      And the fact that it was open source was not as important anymore in a way.

   7. JHJames Hawkins9:16

      No. And so, yeah, open source-

   8. SPSpeaker9:17

      It became more like a communication community.

   9. JHJames Hawkins9:21

      Yeah, like it's kind of something that we'd never reverse, um, because it just feels like the wrong thing to do.

   10. SPSpeaker9:27

       Yeah, for sure.

   11. JHJames Hawkins9:27

       Um, so we kind of maintain the open source, but I would probably say I- we're pretty confident it's under 5% of people self-host our products. There are some really large companies that do. Um-

   12. SPSpeaker9:36

       Still do? Okay.

   13. JHJames Hawkins9:37

       But-

   14. SPSpeaker9:37

       But you don't, but you don't sell them any like-

   15. JHJames Hawkins9:38

       We just, we literally put, we sell literally nothing to them. People who paid for the self-hosted product in the end, um, I think the vast, vast majority of them moved to cloud. We got like HIPAA compliance-

   16. SPSpeaker9:47

       Mm-hmm

   17. JHJames Hawkins9:47

       ... SOC 2 Type 1 and 2. Virtually everyone moved over. I think 95% of people end up moving over despite saying they could not do that. Um-

   18. SPSpeaker9:54

       And so those 95% of people were the ones who I think never hold the cloud before.

   19. JHJames Hawkins9:58

       Yeah.

   20. SPSpeaker9:58

       They end up, like, moving to cloud.

   21. JHJames Hawkins9:59

       Yeah, they found it, and it's kind of because they could justify it because they're like, "Well, we like the product. It really solves the use case, so we're happy going, working with you to get this across." And then the other 5% I think we ended was giving it to them for free and just saying, "We can't support this, but you can keep, you can just have it."

   22. SPSpeaker10:10

       That's brilliant.

   23. JHJames Hawkins10:11

       "Keep using it."

   24. SPSpeaker10:11

       Yeah.

   25. JHJames Hawkins10:12

       Um-

   26. SPSpeaker10:12

       Then, then you did some big moves after because, like, you started with that product analytics-

   27. JHJames Hawkins10:16

       Yeah

   28. SPSpeaker10:16

       ... uh, software.

   29. JHJames Hawkins10:18

       Yeah.

   30. SPSpeaker10:18

       But then you expanded to so many other things like, uh, like most companies, most startups that just do one thing well-
9. 10:33 – 13:06

   The Engineer Who Forced Multi-Product

   1. SPSpeaker10:33

      so many different verticals.

   2. JHJames Hawkins10:34

      Um, it didn't really go in my mind. It went in the mind of one of our engineers.

   3. SPSpeaker10:36

      Okay.

   4. JHJames Hawkins10:36

      So one of our engineers just literally built Session Replay-

   5. SPSpeaker10:39

      Okay

   6. JHJames Hawkins10:39

      ... as a second product, um, because he could see it was in a hackathon, and he could see a bunch of users on the open source project had requested it.

   7. SPSpeaker10:46

      Yeah.

   8. JHJames Hawkins10:46

      Um, and we, uh, Tim and I both thought it was a bad idea, um, but he just shipped it, and it was popular straight away. Um, and then we started realizing, "Oh, this actually gives us a way to compete. This strengthens our position in cloud-"

   9. SPSpeaker10:58

      Yes

   10. JHJames Hawkins10:58

       ... "because we have competition in cloud. Everyone else is a point solution. Um, we'll just build all their products."

   11. SPSpeaker11:03

       Yes.

   12. JHJames Hawkins11:03

       "And we can already sort of feel that we're quicker at shipping." Um, we think software development in general is speeding up, so this is kind of before AI. We were looking at open source going, well, for example, we were able to use ClickHouse as an open source database, um, rather where our competition have all got like-

   13. SPSpeaker11:19

       Yes

   14. JHJames Hawkins11:19

       ... I, I think have all got kind of custom databases and stuff to cope with analytics, um, which means that we don't have to have many people. We do have to have people looking after it, but nowhere near as many as had we-

   15. SPSpeaker11:28

       Yeah

   16. JHJames Hawkins11:28

       ... self-built everything. Um, so we're more able to focus on kind of feature development. I was like, "Well, if open source just continues shipping more free stuff into the world, we can go, software will get commoditized, so we better build all of it [laughs] because that's the only way to compete, I think." Um-

   17. SPSpeaker11:41

       But you went pretty far, like you even did a, a data warehouse, like, which is kind of like-

   18. JHJames Hawkins11:45

       Yeah

   19. SPSpeaker11:45

       ... very different. I don't know.

   20. JHJames Hawkins11:47

       Yeah, the data warehouse is a super full-on product to build. Um, like I-

   21. SPSpeaker11:49

       Yeah

   22. JHJames Hawkins11:49

       ... again, we've got this engineer called James Greenhill, who has been with us for ages. Um, [laughs] and he was like, "There's no way we should build this. This is a stupid idea. It's really complicated." Uh, and then I think, I can't re- I think Tim again said to him, "Imagine I held a gun to your head, and I compelled you to build a data warehouse. Walk me through how you would build it." And then within about 10 minutes, James was like, "We should totally build this thing. I can see how to do it now." Um-

   23. SPSpeaker12:11

       [laughs]

   24. JHJames Hawkins12:11

       ... and then he just built it. Um, so, uh, yeah, we kind of found that at scale Uh, as our customers get, our customers who are-

   25. SPSpeaker12:18

       Mm-hmm

   26. JHJames Hawkins12:18

       ... kind of beyond about 100,000 a year in spend or so, a lot of them start hiring data teams who, um, want to pull in, want to combine, like, the data that we capture through SDKs-

   27. SPSpeaker12:27

       Mm-hmm

   28. JHJames Hawkins12:27

       ... with other data from, like, Salesforce, for example. Um-

   29. SPSpeaker12:29

       So it, it seems like it was, like, a, a lot of things happened. Like, you didn't have a, a plan of creating, like, a huge compound startup from the start.

   30. JHJames Hawkins12:36

       No, it just happened.
10. 13:06 – 15:45

    When Ambition Exploded

    1. JHJames Hawkins13:06

       tools.

    2. SPSpeaker13:06

       It, it feels like, um, uh, as you were going, your ambition kind of exploded in some way.

    3. JHJames Hawkins13:13

       Yeah. We kind of realized that, um, f- m- to- totally, basically

    4. SPSpeaker13:16

       Like why stop with, why, why stop with a single product that, that's, uh, kind of like a normal market-

    5. JHJames Hawkins13:22

       Yeah

    6. SPSpeaker13:22

       ... and then go after everything?

    7. JHJames Hawkins13:23

       Yeah, 'cause we're, um, 100%. Like, we're just following, like, the generic startup advice of, like, pick one thing, do it really well, um, and, like, find a gap in the market, whatever. Whereas actually what we found worked for us was, um, we do, we did much better when we shipped products that had big competitors than when we built products pre-pivot that had no competition 'cause no one cared. [laughs] It kind of meant that, oh, this is, like, shortcut for us, product market fit. It's like other people have got product market fit for this, so we need to. The, the challenge is standing out. Like, maybe we're bringing it to a different audience. It's a branding thing. It's a tweak on something that's there.

    8. SPSpeaker13:55

       Mm-hmm.

    9. JHJames Hawkins13:55

       Um, and I think because we weren't kind of... Especially at that stage, we were very poor from a, like, product vision perspective-

    10. SPSpeaker14:01

        Mm-hmm

    11. JHJames Hawkins14:01

        ... um, it was a bit like, you know, going, like, 10-pin bowling with the lanes up or something is how it felt on-- It got so much easier 'cause we're like, "Well, what are we good at?" And it's like, well, okay, we are willing to just grind our way through this and to do tons of work. So, um, we don't care if we have to work until 3:00 a.m. every day through the YC batch, whatever, to get this out the door, so we'll just focus more on outworking everyone-

    12. SPSpeaker14:19

        Okay

    13. JHJames Hawkins14:19

        ... rather than, um... Uh, so we kind of-- And s- uh, yeah, so it was k- like when we decided, okay, let's build open source product analytics, it was super clear to us what we needed to ship. Um, which just meant that we could just ship every-- like we just shipped basically really, really hard for four weeks to build, like, the website, bas- like, the product, like the SDKs, documentation, um, and so it was kind of usable. Um, but it was a totally... Uh, considering it was all pre-AI as well-

    14. SPSpeaker14:46

        Mm-hmm

    15. JHJames Hawkins14:46

        ... it was totally wild how much work we did.

    16. SPSpeaker14:47

        Yeah, that's what I was going to say. Actually, it felt like today it doesn't seem as crazy as it did when you did it. [laughs]

    17. JHJames Hawkins14:52

        Yeah. [laughs] Same. Yeah, there's more than prompting. Tim was, like... Yeah, basically the method we had was, um, Tim would, um, sit in the house and code all day, and I tried to make sure he stayed there. [laughs] So my job as the less technical one would be, like, go get the groceries, um-

    18. SPSpeaker15:07

        [laughs]

    19. JHJames Hawkins15:07

        ... make him breakfast, and do the washing up. Um, and just try and, like, not get in his w- 'cause I found that if I was coding as well, I'm so much slower that by the time I've, like, done creating a mess in the code base, I'm actually slowing him down. So it's better just to go, "Hey, I'll get-- recruit users, do all the marketing, um, and make sure we're talking to people"-

    20. SPSpeaker15:22

        Yeah. That's, that's for real

    21. JHJames Hawkins15:23

        ... um, basically.

    22. SPSpeaker15:24

        Okay. So, um, ambition increased. You started to build a crazy big surface area product.

    23. JHJames Hawkins15:30

        Yeah.

    24. SPSpeaker15:31

        Um, huge success. And again, you're inventing yourself. [laughs] It kind of feels like, uh, like, uh, that, that crazy ambition was not enough, and now you are like, you want to go to a full self-driving software, the, the new way you look at the world.

11. 15:45 – 17:18

    The Shift to Self-Driving Software

    1. JHJames Hawkins15:45

       Yeah. We, um... So yeah, now our focus-

    2. SPSpeaker15:46

       How did that happen? Like, how did you come here?

    3. JHJames Hawkins15:48

       Um, [laughs] it w- not all necessarily choice, honestly. Like, I think we, um... Kind of the arc we went on was, like, product an- open source product analytics, then doing, um, all the products in cloud. Um, then we started, we started actually getting complaints from customers. Like, I think AI may have been a saving grace for us-

    4. SPSpeaker16:05

       Okay

    5. JHJames Hawkins16:05

       ... um, where customers were kind of saying, "Oh, this is getting a bit overwhelming." Um, so some users adored it. They'll go, "Hey, everything is in this magic box."

    6. SPSpeaker16:12

       Ah.

    7. JHJames Hawkins16:12

       And I was kind of telling myself, well, AWS is doing, like, 100 billion in revenue, and they have a lot of buttons.

    8. SPSpeaker16:18

       Yes.

    9. JHJames Hawkins16:18

       Like, that's fine. Seems to be fine for them. So we'll be fine. But it did, it was not, um-

    10. SPSpeaker16:22

        I'm sure, I'm sure everyone loved [laughs] interacting with AWS-

    11. JHJames Hawkins16:25

        Yeah, yeah

    12. SPSpeaker16:25

        ... products.

    13. JHJames Hawkins16:26

        Uh, I, I think it's kind of, um... I don't know. But it, it, we kind of-- I felt like, okay, it, it's reasonable to ignore this feedback, that it's complicated. Um, but then when AI came, we were like, oh, actually this, we feel like there's an opportunity here to use all these products together because they're all working off the same data. Um, so there are some really cool use cases that started appearing, like people debugging, where you can look at, like, error data, log data, replay data, um, flags, settings.

    14. SPSpeaker16:50

        When, when was that?

    15. JHJames Hawkins16:51

        Um, so this is very-

    16. SPSpeaker16:52

        Back last year or?

    17. JHJames Hawkins16:52

        ... 2025.

    18. SPSpeaker16:53

        Yeah.

    19. JHJames Hawkins16:53

        Um, uh, so we kind of built an agent. We had to work a lot on the architecture. We wound up realizing we needed a, an single query loop, so kind of an iterative approach-

    20. SPSpeaker17:00

        Yeah

    21. JHJames Hawkins17:00

        ... where the agent could, like, do something, assess if it had worked or not, try something else. Like, "Oh, here's a graph. Oh, the graph's empty. Let's try something else." And once we got the agent iterating like that-

    22. SPSpeaker17:09

        Yeah

    23. JHJames Hawkins17:09

        ... the results went from being kind of mediocre-

    24. SPSpeaker17:12

        Yeah

    25. JHJames Hawkins17:12

        ... like kind of slow, slow and in-inaccurate to, um, slow but much better than a human could do, frankly.

12. 17:18 – 19:29

    The Agent That Beat All Other UX

    1. SPSpeaker17:18

       Yeah.

    2. JHJames Hawkins17:18

       And then suddenly we just saw of, of the feedback on, like, we monitor our brand mentions. We have this, uh, Slack channel-

    3. SPSpeaker17:23

       Yes

    4. JHJames Hawkins17:23

       ... um, that monitors what we, how we're talked about online across, like, Reddit-

    5. SPSpeaker17:26

       Yes

    6. JHJames Hawkins17:26

       ... and Twitter and stuff. Um, and we just realized, like, almost, like, 95% of the positive feedback is coming from this agent that we've built that's now sort of became the main UX in the platform. The fourth step was then, okay, people seem to really like this agent. We think it's great internally. Like, this is how I like using our products now.

    7. SPSpeaker17:42

       So yourself was, like, yourself when you look at PostHog as a, as a-

    8. JHJames Hawkins17:45

       Yeah, it was like very quickly

    9. SPSpeaker17:47

       ... owner of this case.

    10. JHJames Hawkins17:47

        Yeah, I was like, "This is the only way I want to use the product now."

    11. SPSpeaker17:49

        Like, like-

    12. JHJames Hawkins17:49

        It's so much better

    13. SPSpeaker17:50

        ... like, early on, like, as soon as it was good enough, like, there was no reason to use the old UI.

    14. JHJames Hawkins17:54

        Yeah, exactly. And then we're like, we've built this UI for no reason, which is fine. [laughs] Um, and then, uh, but the step we're now at is, like, well, why am I... I feel like this agent knows more than I do. It's a better product manager. It has... You can sort of squint and see how this is a better product manager, a better growth engineer than I could ever be, um, if you can pull this off properly. So we've been working on, uh, we've shipped something recently called self-driving mode-

    15. SPSpeaker18:13

        Mm-hmm

    16. JHJames Hawkins18:13

        ... where, um, we've built all these different pipelines for different data, all these different data types we collect. Um, we've built pipelines for them. Um, and then we have these scouts that then read across them. So if you basically just surface, like each new error to a user, you can say like, "Okay, this e- does this error seem important?" In isolation, each error seems kind of important.

    17. SPSpeaker18:33

        Yes.

    18. JHJames Hawkins18:33

        But if you start looking across all of them and you start building kind of company-

    19. SPSpeaker18:36

        Mm-hmm

    20. JHJames Hawkins18:37

        ... context and having kind of a memory system, scientific context on like what the customer and what the individual user and the customer cares about-

    21. SPSpeaker18:44

        Yes

    22. JHJames Hawkins18:44

        ... you can really start prioritizing the same way a human would, 'cause a human you would hold, like, "Oh, I know there's loads of problems, but these two are the ones I actually need to focus on." And instead of it being the case that we just tell you academically, here's something that seems broken, we've gone through-

    23. SPSpeaker18:56

        With the chair and not, why not-

    24. JHJames Hawkins18:58

        Yeah, why don't you solve the actual problem?

    25. SPSpeaker18:59

        ... fix the issue.

    26. JHJames Hawkins18:59

        Yeah. This is kinda my grand theory [laughs] is like, well, it's better to solve the problem than to surface the problem. [laughs] Um, like that's the same if you're dealing with an employee or a company where people who just like solve the problems are more valuable than ones that just complain. Um, and so we don't wanna be a data company, we want to be a doing company, and we should try and fully automate stuff.

    27. SPSpeaker19:15

        But the data is what enables you to, to the doing.

    28. JHJames Hawkins19:18

        Yeah. And get there first as well. Um-

    29. SPSpeaker19:19

        Like, yeah, you get the, you get the... You sit on a kind of like a treasure in a way of data about the usage of the product.

    30. JHJames Hawkins19:25

        Yeah.
13. 19:29 – 21:15

    Fixing Bugs While You Sleep

    1. SPSpeaker19:29

       And so-

    2. JHJames Hawkins19:29

       And then kind of, I would say competitively as well, like I think, um, we're like, "Oh yeah, why not do the same Claude or something?" We're like, well, if we or we know before a human does there's a problem, so we can ship the fix-

    3. SPSpeaker19:41

       Yeah

    4. JHJames Hawkins19:41

       ... um, before Claude can get there. Um, you can also use it through MCP or whatever, which is also very popular [laughs] . But, um, if you want it to be proactive, it's like, well, we're just constant monitoring what's happening so that whilst you're in bed, we're fixing-

    5. SPSpeaker19:51

       That's awesome

    6. JHJames Hawkins19:52

       ... P0 issues for you, for example.

    7. SPSpeaker19:53

       Which is huge. And so that's a recent, uh, evolution of the product-

    8. JHJames Hawkins19:56

       Yeah

    9. SPSpeaker19:56

       ... that like people have started using already or like is it-

    10. JHJames Hawkins19:59

        Yeah

    11. SPSpeaker19:59

        ... available to everyone?

    12. JHJames Hawkins20:00

        Uh, so it is, uh, you can go sign up for this. We haven't, uh, marked it super hard, but-

    13. SPSpeaker20:05

        Well, by the time the video will be released, maybe it will be available. [laughs]

    14. JHJames Hawkins20:06

        Yeah. It should, it better be. [laughs] Um, but yeah, it is. Like we have like lots of users using it already. Um, it's started-

    15. SPSpeaker20:12

        And what kind of a, what kind of adoption have you seen?

    16. JHJames Hawkins20:13

        Uh, so, uh, it's looking very, very healthy. Like it's, um, like we, within two days we sent it to, we emailed to 7,000 of our users roughly. Um, we got past a couple of million run rate, um, within about 24 hours.

    17. SPSpeaker20:26

        Wow.

    18. JHJames Hawkins20:26

        Um-

    19. SPSpeaker20:27

        7,000, is that the majority of users? Like is that a small portion? Like I don't know how many users you have.

    20. JHJames Hawkins20:31

        Um, a tiny portion.

    21. SPSpeaker20:32

        Okay.

    22. JHJames Hawkins20:32

        I think in total we probably have about a million-

    23. SPSpeaker20:33

        Okay

    24. JHJames Hawkins20:33

        ... uh, users or something, uh, like email addresses or whatever. Um, but we're kind of, um, and we picked good ones. Like we picked ones who were like using the right kinds of data that we thought-

    25. SPSpeaker20:42

        Yeah

    26. JHJames Hawkins20:42

        ... worked particularly well and stuff. So there's like skew in this, but in short, um, we have, we've got a ton of usage from... We, we started using it internally first.

    27. SPSpeaker20:49

        Okay.

    28. JHJames Hawkins20:49

        Started solving a decent chunk of problems for ourselves. Um, then-

    29. SPSpeaker20:55

        So-

    30. JHJames Hawkins20:55

        Did private beta, then did this. Um-
14. 21:15 – 23:02

    Half of PRs Are Getting Merged

    1. SPSpeaker21:15

       for your, um, customers. Uh, how much for you?

    2. JHJames Hawkins21:18

       Uh, it's, it's similar. It's similar across both. Um-

    3. SPSpeaker21:21

       Okay

    4. JHJames Hawkins21:21

       ... uh, because the product's the same [laughs] basically. Um, but yeah, we're kind of, uh ... It might be higher for some of our customers, I think because our cust- a lot of our customers just in terms of numbers are gonna be smaller than us because we have in terms of ... Like, they definitely are getting ... Like I can see a bunch of them who are basically just like merge, merge, merge [laughs] when they wake up, and they kind of assume. That behavior I think is healthy, um, for a small company. Um, but we're trying to build, we're not trying to, we're starting to build now more tooling around kind of, um, the really obvious things like code review or-

    5. SPSpeaker21:47

       Yeah

    6. JHJames Hawkins21:47

       ... um, like agentic testing.

    7. SPSpeaker21:48

       Mm-hmm.

    8. JHJames Hawkins21:49

       And we want to get to the point where we're shipping things proactively behind flags-

    9. SPSpeaker21:51

       Yeah

    10. JHJames Hawkins21:51

        ... and so on, using the other tools that we've already built, um, or shipping stuff as an experiment, for example. Um, so we're investing more heavily to try and get the, like, merge rate higher. So if you feel like we've given you a pull request, you can trust it, um, which isn't solved. Yeah, like you definitely should look at it probably quite a bit.

    11. SPSpeaker22:05

        I don't think what, I don't think what you have been, you've done so far would, I would even have imagined possible, that possible last year. Uh-

    12. JHJames Hawkins22:13

        Yeah. It wasn't. Um-

    13. SPSpeaker22:14

        And so like just, uh, thinking about six months from now-

    14. JHJames Hawkins22:16

        Yeah

    15. SPSpeaker22:16

        ... um, like I'm not worried about that being solved, so it's more like, okay, what's next? [laughs] Going to be crazy.

    16. JHJames Hawkins22:21

        Yeah, I think, yeah, basically if you can go from, um, typing in English to like a 20,000 line of code application-

    17. SPSpeaker22:26

        Mm-hmm

    18. JHJames Hawkins22:26

        ... I'm pretty sure once you've got your code, you should be able to double check it carefully.

    19. SPSpeaker22:30

        Yes.

    20. JHJames Hawkins22:30

        Um, it's just, I think it's just a matter of time. We actually-

    21. SPSpeaker22:32

        Yeah

    22. JHJames Hawkins22:32

        ... we're starting to also think about a bit beyond this, like we're... So I think we've learned as we're in the AI, which again, I would give as a general advice, if you're doing AI stuff now [laughs] or if you're building software-

    23. SPSpeaker22:40

        Yes

    24. JHJames Hawkins22:40

        ... um, the pace is so much higher that we're having to go like some other standard startup advice, right, is go where the puck is going. Um, we didn't used to do that. We used to just go where the puck already is really, really fast. And we've kind of realized actually, no, we do now when we're competing against comp- people who are building AI natively-

    25. SPSpeaker22:56

        Yeah

    26. JHJames Hawkins22:56

        ... they move so much quicker that you actually do need to place a bet on the direction of travel a bit more.

15. 23:02 – 24:27

    Betting on Where the Puck Is Going

    1. JHJames Hawkins23:02

       So we're working on things like, so we've gone, okay, let's, we have a team we're working on, we have people working on like code review, getting stuff merged.

    2. SPSpeaker23:07

       Yeah.

    3. JHJames Hawkins23:07

       There's a lot to build there. But in parallel we also have people now working on like, okay, let's assume that stuff works.

    4. SPSpeaker23:12

       What's next?

    5. JHJames Hawkins23:13

       What's next? And then we're like, okay, we think it's about kind of, um, a different harness that captures kind of goals and intent from a product. It's like, well, if you go from like one pull request a day in the olden days to five to 100. At 100 it's like, okay, now you're limited by ability to like describe something in English. Um, so how can we build a harness that will work with a human to kind of pull out of them [laughs] what the actual requirements are?

    6. SPSpeaker23:32

       Mm-hmm.

    7. JHJames Hawkins23:32

       So you can let the agent run for longer. We have a better idea of what the product's supposed to do, which makes it easier to know what the goals are of the product. So we do a better job of like shipping fixes or taking them in that direction or shipping like experiments for example. So, um, we are starting to get quite a lot more aggressive with how far forward we're willing to go, which is new for us. Like we haven't, we weren't very innovative at all for, by design for like the first few years. Um, we're starting to build more confidence. Again, it's a confidence thing I think where we're like, actually, like we built some stuff that we consider ahead of everyone else in our industry. Um, but then we think the next step is-

    8. SPSpeaker24:01

       Mm

    9. JHJames Hawkins24:01

       ... okay, let's go like further, further, further in.

    10. SPSpeaker24:03

        Yeah.

    11. JHJames Hawkins24:03

        Obviously, if this goes horribly wrong, we like waste loads of time. [laughs] Um-

    12. SPSpeaker24:05

        What does that mean for the... So it's super interesting what you're doing, everything you explained about the product. What does it mean for the company, PostHog the company?

    13. JHJames Hawkins24:11

        Yeah.

    14. SPSpeaker24:12

        Because there is like kind of, I guess everyone at PostHog is using agentic coding, something, something.

    15. JHJames Hawkins24:16

        Yeah.

    16. SPSpeaker24:16

        But, uh, the way you described your harness for the next iteration of the product-

    17. JHJames Hawkins24:21

        Yeah

    18. SPSpeaker24:22

        ... uh, does it apply to the company itself, how you run the company, how you become yourself an AI native company?

16. 24:27 – 29:06

    Elephant Carpaccio

    1. JHJames Hawkins24:27

       Yeah. The way I would kind of say it is-

    2. SPSpeaker24:28

       Because you are like more a legacy company at this point, but-

    3. JHJames Hawkins24:30

       Yeah, I think it's kind of, um- Well, I'm trying to con- concept- conceptually, like often thing referred to in a meeting, like I, I spend all my t- time hanging out with engineers building AI stuff and, um... Or at least like 95% of it, I would say. I'm trying to get the engineers in a position where their job feels like my job today, where my job today is, like, have a bit of a discussion with an engineer usually-

    4. SPSpeaker24:52

       Yeah

    5. JHJames Hawkins24:52

       ... going like, "What are we thinking about building here?" And, um, uh, like, that, the harness we're building is kind of supposed to reflect what it feels like to talk to a product engineer. So someone who is... 'Cause, uh, with an engineer, if I said something like, "Hey, I think this error could be really important," they won't come back. The, the question they will ask me isn't gonna be like, "Should I mock up the data in the back end? Which database should I use?" Which is currently what the harnesses would do. Um, but instead they'll be like, "Well, why do you think that?" Um, or, "Really?" Like, "What about this alternative?" Or, um, w- "Wait, are you trying to make this simpler than everything else? Are you trying to make it more complicated or more powerful, or what's the deal?" Like, and I, I think that interaction I have with an engineer, like I want the engineer to be having with the agent-

    6. SPSpeaker25:28

       Agent

    7. JHJames Hawkins25:28

       ... instead. So it's trying to like abstract everyone upper [laughs] level, is the way I would think about it. Um-

    8. SPSpeaker25:34

       Mm-hmm

    9. JHJames Hawkins25:34

       ... it's similar to, um... We also think that the same applies to go-to-markets. There's a whole load of interesting use cases there-

    10. SPSpeaker25:40

        Yeah

    11. JHJames Hawkins25:40

        ... that are very similar. Um, it's the same style of loop. Um, but yeah, it's more like the way I would-

    12. SPSpeaker25:45

        Do you have any example of-

    13. JHJames Hawkins25:45

        ... kind of see AI building as abstracting people from their jobs

    14. SPSpeaker25:47

        ... do you have any example of what you're doing on the go-to-market side?

    15. JHJames Hawkins25:48

        Um, so we've basically found that by using one agent that uses a lot of tools together, um, it's made everything much more accessible. Like, I think there's a thing where historically you're supposed to have kind of a persona that you target inside of a customer. Um, and we're like, well, you know, it's either like an engineer or it's a salesperson or a CS person or someone else. Um, these roles are still different to each other, um, but I do think they've stepped, everyone's stepped towards everyone else, where, okay, like for example, our support team will fix bugs in the product if they're reasonably straightforward. They would never have done that pre-AI. Um, or our product managers will change something small in the product themselves because the fact that you can program a lot of things in English now means that, um, everything's got more accessible, which means that like slightly sort of software adjacent people-

    16. SPSpeaker26:35

        Mm

    17. JHJames Hawkins26:35

        ... will ship bits of software. Like, um, we have a Slack bot for PostHog, um, that lets you code or ask questions or whatever, or use our tools. Um, 62%, I think, of pull requests that come through Slack are not from software engineers at PostHog anymore. It's from, like, the marketing team, for example-

    18. SPSpeaker26:48

        Okay

    19. JHJames Hawkins26:48

        ... going like, "Okay, we've created an experiment," or whatever it might be. W- we've seen a pattern, that pattern happening in our customer base, so we've realized that, like, back in the olden days, PostHog was virtually all engineers using it and quite technical founders. For our AI tooling, it's 25% software engineers, 25% founders who are mostly technical. The other half is these, like, software adjacent kind of roles, so it's like PMs, technical marketing people, um, growth engineers. Um, we think there's a very similar loop concept for them, where in engineering, your self-driving product means-

    20. SPSpeaker27:18

        Yes

    21. JHJames Hawkins27:18

        ... um, like find prob- find UX issues, find bugs, ship experiments proactively, this is the direction of travel, and then, like, set higher level goals or work on like bigger things. Um, I think similarly, like in, um, customer success, uh, I think there's a world where, uh, like the stuff we're building is like, okay, well, we shouldn't have outputs that are just pull requests. We need to give the same thing, the ability to, like, book a meeting with a customer, for example, um, so that a CS, like a customer success person's job can be like, "I'm literally on meetings all the time," or, "I'm configuring how customer success should work." Um, there's nothing else that is required, kind of thing.

    22. SPSpeaker27:51

        Okay.

    23. JHJames Hawkins27:51

        Um, because we've seen so much usage of, like, our agent already, that's a subset of this use case. It's like, is this customer gonna churn? How's this customer using us? Is this customer using this feature or this feature? I want to try and cross-sell them, basically. We can go, "Hey," instead of telling that small slice, like giving them the information, we want to go, "Well, what's the job you're actually trying to achieve?" It's probably to like cross-sell a customer or-

    24. SPSpeaker28:10

        Sure

    25. JHJames Hawkins28:10

        ... to make sure a renewal happens or to realize there's a problem and help them debug it or something. And so we're like, okay, let's just take the whole sli- like the whole job kind of thing. Um, uh, this is like, uh, engineer... This, this is an engineering consultant in London who's kind of like a underground celebrity in the London tech scene.

    26. SPSpeaker28:28

        Okay. [laughs]

    27. JHJames Hawkins28:28

        Um, and he used to use this phrase, elephant carpaccio. So if you, um, take an elephant and you slice the elephant up, you can slice it horizontally-

    28. SPSpeaker28:36

        Yes

    29. JHJames Hawkins28:37

        ... or you can slice it vertically.

    30. SPSpeaker28:38

        Okay. [laughs]
17. 29:06 – 31:49

    The Founder Who Became a PM

    1. SPSpeaker29:06

       I wonder how you g- like you got there pretty fast. It feels like you actually took on, uh, that PostHog AI team yourself, right?

    2. JHJames Hawkins29:14

       Yep.

    3. SPSpeaker29:14

       Kind of like, um, in a way I was surprised because, uh, I would have expected that to come from more the tech, uh, leadership-

    4. JHJames Hawkins29:20

       Yep

    5. SPSpeaker29:20

       ... uh, not someone who has a sales background.

    6. JHJames Hawkins29:23

       Yeah.

    7. SPSpeaker29:24

       Like, how did you... how did that happen? Uh-

    8. JHJames Hawkins29:26

       Yeah. [laughs]

    9. SPSpeaker29:26

       Like, PostHog has always impressed me in terms of like, uh, like division of responsibility between, uh-

    10. JHJames Hawkins29:31

        Yep

    11. SPSpeaker29:31

        ... between you and team and, and how you work there.

    12. JHJames Hawkins29:33

        Yeah, you would have, uh, you would have thought you'd want someone competent doing this. Um, but yeah. [laughs]

    13. SPSpeaker29:37

        No, no, that's not what I was saying at all.

    14. JHJames Hawkins29:38

        [laughs] No, it was true. Uh, I think the... It's just one of the things that I think Tim and I learned that, um, again, I think there's a very strong human behavior-

    15. SPSpeaker29:44

        Yes

    16. JHJames Hawkins29:45

        ... to do what you've done before, and the reason you're doing that is perhaps because you just slightly randomly made a decision when you were 22 to go be a product manager or-

    17. SPSpeaker29:53

        Yes

    18. JHJames Hawkins29:53

        ... a sales leader or something like that. And so I think Tim and I have sort of wound up in slightly the wrong career path, and we've kind of realized this at PostHog where-

    19. SPSpeaker29:59

        You mean you should have been tech, you should have been the sales?

    20. JHJames Hawkins30:01

        Probably, yeah. I'm not a very g- I don't think I'm actually a good developer. I, I naturally feel like I'm quite a-

    21. SPSpeaker30:05

        Yeah.

    22. JHJames Hawkins30:05

        I feel naturally like a very bad developer, honestly. I have done a bunch of software, and I'm just bad. Um-

    23. SPSpeaker30:09

        But you're good at, uh, guiding AI.

    24. JHJames Hawkins30:11

        Uh, yeah, like the product side, I think I'm just, like, I... The product side I enjoy an awful lot-

    25. SPSpeaker30:15

        Okay

    26. JHJames Hawkins30:15

        ... um, and from that have got to the point where I feel more competent. And like a lot of, like Tim and I do, um, like we do these like 360 feedback dinners-

    27. SPSpeaker30:22

        Mm

    28. JHJames Hawkins30:22

        ... at PostHog. So, um, probably every six months or so, um, every team will have dinner where one-- you sit there. So like the, like exact team we have will sit down, but this applies to every single team at PostHog, and I sit there, and then everyone else gives me mostly constructive feedback. [laughs] Um, and then it moves on to the next person, and you just listen. You don't-

    29. SPSpeaker30:39

        Yes

    30. JHJames Hawkins30:39

        ... challenge it, you just ig- You can ignore it, but you just listen to your team. Uh, and this exercise has been super helpful for us. Um, and the feedback I was getting from Tim was like, "You should..." He's like, "I really think you should just back yourself more on these bigger product ideas that we're having, and just go make them happen, basically. Don't, because you have not done this before, don't half-ass it"-
18. 31:49 – 34:41

    Why Going Wild Gets Easier

    1. SPSpeaker31:49

       It feels like it's even existential. If you don't do it-

    2. JHJames Hawkins31:51

       Yeah

    3. SPSpeaker31:52

       ... uh, your company may risk becoming irrelevant in time.

    4. JHJames Hawkins31:55

       Yeah, exactly, and you're kind of framing it a little bit as like-

    5. SPSpeaker31:56

       Yes

    6. JHJames Hawkins31:56

       ... yeah, this feels like a big bet, but it's like, oh, no, actually, like, we're forced into this. [laughs] Like, we have to, um-

    7. SPSpeaker32:00

       Well, you're forced, you're forced, but, uh, but you embraced it, so it's not like, uh-

    8. JHJames Hawkins32:03

       Yeah

    9. SPSpeaker32:03

       ... it's not like you were forced to do something against your will.

    10. JHJames Hawkins32:06

        [laughs] Yeah.

    11. SPSpeaker32:07

        Uh, but, but for, for everyone out there-

    12. JHJames Hawkins32:09

        Yeah

    13. SPSpeaker32:10

        ... um, I guess, uh, you need a founder-

    14. JHJames Hawkins32:12

        Yeah

    15. SPSpeaker32:13

        ... to, to actually lead the charge here.

    16. JHJames Hawkins32:16

        Yeah, I think it's really-

    17. SPSpeaker32:16

        Is that fair?

    18. JHJames Hawkins32:16

        Yeah, there's no way this would have, uh, for me at least-

    19. SPSpeaker32:19

        Yeah

    20. JHJames Hawkins32:19

        ... I would struggle to see us managing to, to, like, outsource a shift like this where, um-

    21. SPSpeaker32:22

        Like, if the, if the, if the founders are not, like, uh, fully bought in-

    22. JHJames Hawkins32:25

        Yeah

    23. SPSpeaker32:25

        ... uh, kind of that kind of huge seismic change-

    24. JHJames Hawkins32:29

        Yeah

    25. SPSpeaker32:29

        ... I guess it's not going to happen.

    26. JHJames Hawkins32:31

        No, and the, and the way I think about it as well is just like, uh, you know, if you look at Anthropic who are, like, the fastest growth company in the history of mankind or whatever, I'm pretty sure it's not because the sales team are really good. Like, it's an exceptional product.

    27. SPSpeaker32:41

        [laughs]

    28. JHJames Hawkins32:42

        Um, thus they are... And it's the most leveraged place, I think, to be really strong, so I'm just a very large believer in... I, I just, I struggle to make myself care about stuff that isn't product, so increasingly as we're growing, because when I look at the, like, very, very top-end competition, the, like, product work is outstanding. Um, and so I think that's where, like, I really want a big chunk of, like, founder focus to... But I think we're kind of losing the battle the moment that we, um, get focused on... Like, we have a sales team. Sales matters to us.

    29. SPSpeaker33:09

        Yes, of course.

    30. JHJames Hawkins33:10

        We're, like, foc- we're, like, increasingly working on, like, enterprise and stuff, but it's not, that's not my focus. Like, that is something that consciously we have other people that we've brought in to do this. Like, we're not trying to... We do sales somewhat uniquely-
19. 34:41 – 38:51

    Free Up Everything That Isn't This

    1. SPSpeaker34:41

       should they start?

    2. JHJames Hawkins34:41

       Yeah, it's, it's like you have to be ruthless. Like, we, um, like I literally have stopped doing... So it got to the point where I think, like, start of 2025-ish or, like, early summer or something.

    3. SPSpeaker34:50

       Yeah.

    4. JHJames Hawkins34:50

       I think, like, I was... Basically, Tim and I got to the point where we each had roughly half the team somehow or another reported to one of us.

    5. SPSpeaker34:56

       Yeah.

    6. JHJames Hawkins34:56

       Like, I had, like, half the exact team. He had the other half, and then that would, like, then you follow the lines down to, like, everyone in the company. So it's like, ultimate, I'm responsible for, like, half the business. He's responsible for kind of the other half. And so we went, "Hey, this is just not gonna lead to, um... There's too much noise, too much random stuff going on."

    7. SPSpeaker35:11

       Yes.

    8. JHJames Hawkins35:11

       Like, literally, this product needs to exist, just, like, go make the sales IP.

    9. SPSpeaker35:14

       So you need to make the time for it.

    10. JHJames Hawkins35:15

        Yeah. So Tim went, like, "Okay, I will..." And then we're like, "Well, do we both just do AI stuff?" And our conclusion was, if we both try and build the product at the same time as each other, we're just gonna get in each other's way. Like, we're gonna both pull the com- the teams in slightly the wrong directions and stuff. So, um, Tim basically had my back and went, "I'm gonna look after everything, um, so that you're, I'm, so I become completely redundant."

    11. SPSpeaker35:35

        So you've got the time.

    12. JHJames Hawkins35:35

        And I can-

    13. SPSpeaker35:35

        So you can, like, spend all your time on that initiative.

    14. JHJames Hawkins35:37

        Yeah, so now I'm just like, "Okay, I need to, um, I need to get us to self-driving." Um, and in my head I'm like, "What is the current blocker to self-driving happening?" And that is where I'm gonna spend my time today. Um, and I've just been doing that for maybe a year and a half or something like that. And then it's just meant that, like, it's lots and lots of time with engineers and, like, pulling people internally from one team into another and, like, rejigging stuff. Because again, we have lots of engineers. Like, we have, like, I'd say we probably have, like, 125 engineers or something. And, um, again, it's like, well, I don't think this is gonna go quicker if I put 125 people in at the same time. Like, we've had to, like, ramp lots of people on this project. Um, but it started with maybe, like, two people or three people-

    15. SPSpeaker36:11

        Yeah

    16. JHJames Hawkins36:11

        ... put into post-AI, and I've been trying to paralyze as much of the work as possible. Um-

    17. SPSpeaker36:15

        So maybe that the extreme version where you freed up, like, pretty much all your time. [laughs]

    18. JHJames Hawkins36:18

        Yeah, but it's like as much as humanly possible.

    19. SPSpeaker36:20

        Yeah.

    20. JHJames Hawkins36:20

        Like, I'm not doing, like, even stuff like interviewing, Tim does all of. Like, he's taken... We were doing half the interviews each.

    21. SPSpeaker36:25

        Yeah.

    22. JHJames Hawkins36:25

        Now he's doing, like, the vast majority of it. I'm doing, like, very rarely interview people anymore. Um, I've, like, not done any investor stuff. Um, it's, like, really nothing. I don't do one-on-ones with anyone. Um, so I've just ditched everything that I felt wasn't contributing to this. Um, so it's really quite-

    23. SPSpeaker36:40

        Okay

    24. JHJames Hawkins36:40

        ... far that you have to go, I think, to get to change.

    25. SPSpeaker36:41

        So that's, that kind of like the extreme version, but I think that's good advice here, like, uh- In a way because you need to free up that time.

    26. JHJames Hawkins36:47

        Yeah. And it's also-

    27. SPSpeaker36:48

        Otherwise you cannot really-

    28. JHJames Hawkins36:49

        And frankly, it's very fun [laughs] to-

    29. SPSpeaker36:51

        Well, exciting to work on, right?

    30. JHJames Hawkins36:52

        Um, but yeah, I kind of hope we keep this, um, going. And you can see proof points to this. Like Elon Musk does this, I think. Well, that's the impression I get anyway. Um, so like, well, it works for him. He's like, kind of successful [laughs] . So, um, we ought to be able to pull this off ourselves. Um, but it has been much more work than I would have expected to change 'cause like we've got products that are mature doing tens of millions or like multiple tens of millions of run rate each. Um, and we've hired people that will look after these, and so it can be very unnatural for people to like totally-

Episode duration: 38:53