Des Traynor: How to Survive and Thrive in a World of OpenAI | E1082

1. 0:00 – 0:54

   Intro

   1. DTDes Traynor0:00

      (instrumental music plays) ... OpenAI will hit some sort of minimum viable kind of what's good enough for everyone. You're never gonna make money (cash register rings) filling and in the gaps in a platform. I liken what we're going through right now to like the (cash register rings) beginning of the internet. It's very big what's happening to both our world of customer service, technology in general. We don't have the luxury of sitting back and watching. There's gonna be a lot of winners and losers and there's gonna be a lot of market share up for grabs, so people are gonna be looking for who's got the good shit.

   2. HSHarry Stebbings0:22

      If you are CEO of Google, what would you change?

   3. DTDes Traynor0:25

      I kinda hate what I'm about to say. I'd be-

   4. HSHarry Stebbings0:29

      Des, I am so excited for this. Dude, when, when we did our last one, I remember, it was the first one to break an hour and 15. And I was like, "Fuck, the reasons like this show is why I love doing what I do, and I'd love to do it in person." So I'm thrilled that you're here in person.

   5. DTDes Traynor0:42

      I'm happy to be here. I love London and it's awesome to actually do this in person and see the three-dimensional Harry.

   6. HSHarry Stebbings0:46

      (laughs) Most people are like, "Wow, you're actually quite old."

   7. DTDes Traynor0:49

      (laughs)

   8. HSHarry Stebbings0:49

      I'm like, "Thank you, yeah."

   9. DTDes Traynor0:50

      Yeah.

   10. HSHarry Stebbings0:50

       Age is this kind of strange thing that happens.

   11. DTDes Traynor0:52

       Yeah.

   12. HSHarry Stebbings0:52

       Uh, before we dive in, Des,

2. 0:54 – 4:16

   Founding Stories and Personal Dreams

   1. HSHarry Stebbings0:54

      what did you want to be when you were younger? When you were growing up in Ireland as a kid, what did you want to be?

   2. DTDes Traynor0:58

      I think it was either a, a professional video games tester, but I kinda knew that wasn't a career, or at least definitely wasn't in the '80s. A sports journalist. I loved soccer and I loved writing, so I really wanted to like write about soccer. I just thought like if I could do that, that would be like the, genuinely the easiest job. I used to actually do it. I used to like, I'd watch a game and I'd write down my match report and I'd get my mom to read it and she'd be like, "Oh, it's very good." I was just passionate about like, I, I think, I love soccer, watching it, and then I literally love writing as well. And I love to, like, I'd always read all of the cover to cover, every sports newspaper, everything, just trying to get a sense of what's the language, how you're supposed to talk, all that sort of stuff.

   3. HSHarry Stebbings1:30

      (laughs) When your mother was like, "You know, that's amazing, darling." And more or less like, "Ah, thanks, Mom." (laughs)

   4. DTDes Traynor1:33

      Yeah, yeah, well totally. I mean, thankfully I didn't end up doing that 'cause I have friends who are sports journalists and it doesn't pay as well as they thought it might.

   5. HSHarry Stebbings1:39

      You know what? It's better to do software. (laughs)

   6. DTDes Traynor1:41

      Yeah. (laughs)

   7. HSHarry Stebbings1:41

      But Intercom, just for those that don't know, how did it come to be?

   8. DTDes Traynor1:45

      It actually started life inside a different product, uh, believe it or not. We, uh, had a different product called Exceptional. W- we wanted to talk to our customers, so, uh, the logo for Exceptional sat in the bottom right hand corner of the product. And one day, ptt, it had a little speech bubble that came out and it was like, "Hey." Probably something like, "Hey, we're sorry that we fell over, uh, or the server wasn't working last weekend." Or something like that. And that was kind of like, uh, the beginning of like, that was a much better workflow to have that bubble pop than the old world. The old world if y- if you, some of your viewers might be old enough to remember, would have been, you get a database export, you import that into Mailchimp or Campaign Monitor or something like that. You sync it like with S- PayPal 'cause there was no Stripe, to work out who were on a paid subscription. Just bear in mind, PayPal didn't know anything about SaaS. It just thought these were subscriptions. So you, you'd have to figure out who was paying. You'd import all that information. You'd send out the mail to say like, "Hey, our, you know, here's some preannounced downtime," or more likely some apologizing for some downtime we didn't know was coming. And then you get a shit load of replies in your personal email. It was crap. And then with Intercom, it was like one day it was like, "Why don't we just talk inside the product?" And like the... Another sort of inspiration for us was we were working out of a coffee shop called 3Fe in Dublin at the time. And the dude who was building 3Fe, this new recurring software busin- or recurring subscription business if you like, like where he'd be selling coffee every day to the same people, he was talking to people. He was, you know, "Hey, I've got a new blend in. Do you wanna try it out?" Or like, "Hey, is your latte nice?" Or whatever. He was getting like real time in-context feedback, although h- in his head he was just running a coffee shop. But for us we were felt really like divorced from our users, uh, there was no obvious way, like in our heads we're like, "We're sitting here in Dublin, Ireland. They're all in San Francisco." Like, there was a single street in San Francisco where we had more customers than we did in all of Ireland, to give you an example of-

   9. HSHarry Stebbings3:29

      (laughs)

   10. DTDes Traynor3:30

       ... of the gap. But, um, what we realized pretty quickly was it wasn't a unique problem to us 'cause we were in Dublin. Everyone who runs software businesses wanted to be closer to their, they wanted to have that personal connection. So we started to like iterate on this idea of like, maybe, maybe users should be able to talk back. Maybe you should choose who you send a message to. Maybe you should see who's s- who saw the message and who hasn't seen it yet. Uh, and maybe the user should be able to start conversations. And that ultimately, uh, became what we call like the customer communication platform Intercom, and then today we call it like an AI-first customer service platform because it's kinda been 10 years and we've matured our thinking a lot. But yeah, that's how it started.

   11. HSHarry Stebbings4:08

       I mean, and also it has to be AI-first, right? I mean like...

   12. DTDes Traynor4:11

       Well, yeah, you'd be an idiot not to. Yeah, yeah.

   13. HSHarry Stebbings4:12

       Why, why would you not be doing that?

   14. DTDes Traynor4:13

       Yeah, I'm sure coffee shops are AI-first too as well.

3. 4:16 – 10:07

   The Rise and Integration of AI

   1. DTDes Traynor4:16

   2. HSHarry Stebbings4:16

      Hey, could I ask, man, how do you feel about everyone being like AI-first? Like, we see it as a media company obviously, every single kind of large scale up and growth company is like, "AI-first." And they come throwing on-

   3. DTDes Traynor4:25

      Yeah.

   4. HSHarry Stebbings4:25

      ... and you actually tweeted that brilliant-

   5. DTDes Traynor4:27

      Yeah, yeah, the, like-

   6. HSHarry Stebbings4:27

      ... picture of like-

   7. DTDes Traynor4:28

      Yeah, yeah.

   8. HSHarry Stebbings4:29

      Yeah. Yeah.

   9. DTDes Traynor4:29

      We can m- we're gonna end up talking about AI for a while now so I'll, I'll just start off, um... I think (clears throat) it's largely bullshit if the, if the f- basic tech stack hasn't actually changed that much. Um, so what I mean by that is, let's say you're a project management tool, all right? And like, hey, what, what is project management? It's like tasks, deadlines, files, approvals, decisions, communication, whatever. It's not like really obvious to me that AI is going to like take away a lot of that, 'cause like you're planning out projects and like oftentimes these aren't like, you know, um, these are like things that really have to happen, whether it's like you're building a house or you're, you're managing a rebrand for a client or something like that. There's a lot of like human decision and judgment made. It's not obvious to me where the AI kicks in and, uh, so if you're like an AI-first version of Basecamp, I still think that looks like Basecamp and maybe they just find areas to sprinkle in AI to speed up little bits and pieces here and there. Maybe, may- maybe not. I think there's, if you take say the area of customer support, everything about customer support is gonna be affected by AI. For example, summarizing an issue is a very comm- common workflow that has to happen. Replying to a customer based on the fact you found in a knowledge base is a very common thing that has to happen. Consolidating different articles to g- to give a conclusive answer to something, being conversational, these are all things that AI can do. So I think if you were to build a customer support solution today you'd build it totally different than say when Zendesk was started in 2007 or 2008 or whatever. Like, I think the world of like ticketing and structure and state and like, you know, "We'll get back to you in 72 hours" and all that, that's all gone. So I think like when I, to your question about AI, I just think it's, it's bullshit when actually the job is identical and you're just sprinkling on like salt and pepper a bit of AI here and there, and I think it's true when it's like...... shit, we need to reimagine this entire space.

   10. HSHarry Stebbings6:14

       Can I push back and just say that do you think not, you know, if you think about, like, an Asana or a monday.com-

   11. DTDes Traynor6:18

       Yeah. Yeah.

   12. HSHarry Stebbings6:18

       ... or whatever, they could actually connect to your Google Mail accounts, your, you know, calendars and actually create a huge amount of tasks or be more proactive in terms of how you spend your time, or actually be part of the creation of your to-do list, which is much more than just, like, a sprinkler. Like, I'm just-

   13. DTDes Traynor6:36

       Yeah.

   14. HSHarry Stebbings6:36

       ... pushing that saying-

   15. DTDes Traynor6:36

       Yeah. Yeah.

   16. HSHarry Stebbings6:37

       ... I think it's so foundational that it could actually be that instrumental to almost everything.

   17. DTDes Traynor6:41

       I, I think that's true and when that happens, I'll support them saying it's AI first. But that hasn't happened yet. Right now, it, we're, we're firmly into, like, salt and pepper category, in my opinion. But you are cor- correct, like, if the job of, like, you know, managing a project involves, like, reading email, checking tasks, looking at GitHub commits to see if something got approved or not, checking off tasks and corresponding with a calendar and seeing when things are on, on, but, you know, using all that information to take action or to request action, then yeah. If somebody builds, like, the, the project megabot that can do all that, I will celebrate them, I'll invest in their company. I'll, I'll call them, like, an, an AI-first project management solution. But I just, uh, we're not there yet.

   18. HSHarry Stebbings7:20

       I mean this respectfully, you mentioned Zendesk-

   19. DTDes Traynor7:22

       Yeah.

   20. HSHarry Stebbings7:22

       ... there in terms of, like, it's difficult to integrate when you have the stack that is like theirs. Intercom's is not like theirs, but you're not starting from nothing today where you have a complete clean slate.

   21. DTDes Traynor7:31

       Yeah.

   22. HSHarry Stebbings7:31

       Is it not still really fricking hard for you then?

   23. DTDes Traynor7:34

       One area where we have been fortunate is we were all in on messaging from the very start. So, eh, from the very start, we had this idea of, like, an actual messenger, like, you know, WhatsApp-style messenger-

   24. HSHarry Stebbings7:43

       Yeah.

   25. DTDes Traynor7:43

       ... embedded in the product. That's where chatbots live. They don't live in tickets. So, like, if you think about, like, the, the idea of a chatbot is actually the intersection of two mega trends. It's, like, AI and it's messaging. That's what makes a chatbot. So I think we were in a very fortuitous position to have a complete, sort of CS platform and have, as our core mode of interaction, actual messaging back and forth. We obviously support ticketing and all these other things, but, like, our core way people know of Intercom is the little floating bubble in the messenger.

   26. HSHarry Stebbings8:10

       Yep.

   27. DTDes Traynor8:10

       Uh, so I think we were, we were in pole position to do it. There wasn't a substantial rebuild 'cause th- you know, for a start, we had... Our first AI product was built in 2016 called Resolution Bot. So we, we have already been scratching at this, but a lot of those, like, bots kind of went through different eras, if you like.

   28. HSHarry Stebbings8:23

       I remember 2016 exactly well.

   29. DTDes Traynor8:25

       Yeah.

   30. HSHarry Stebbings8:25

       I was in the Valley and-
4. 10:07 – 17:12

   Business Strategies and Product Insights

   1. DTDes Traynor10:07

      I think it depends on what, what you're actually talking about. Like, there's a big difference between a feature and a product. If you wanted to, say, build a chatbot that does customer support, it is true that, like, and there's been a lot of these, like, sort of wrappers where give a URL, answer a question, right? That's pretty easy to build. It's, like, half a day's worth of work and you're sitting and you're wholly dependent on OpenAI or Anthropic or whoever it is you could go for it. There's a massive difference between that and, say, something that, like, a large company would actually deploy to its customers, which, you know, ingests knowledge bases, requests refreshes, reports on itself, reports on customer satisfaction. You know, like, uh, you know, can parse, uh, multiple conversations, can, uh, u- understand customer context and feed that in, knows the difference between a premium user and a free user and gives conte- context to where answers, whatever. Like, there's a wave upon wave of extra stuff you have to, to, to build in. Like, and that's before we get to, like, permissioning, controlling and targeting and making sure some users don't see bots and some users only see bots and... Like, all of that is, like, the extra stuff you have to build to turn what's, like, a thin, how would you say, like, science fair tech demo-

   2. HSHarry Stebbings11:06

      Yeah.

   3. DTDes Traynor11:07

      ... into an actual fully commercialized product. So I think, like, to the VC argument of, like, uh, a th- thin wrapper, I think if it is a thin wrapper, it's, it's a bullshit argument. Um, but if you take, say, like, um, some people are building on top of, say, DALL·E or something like that, right? Literally, it, it depends. You can build a Canva-style product on top of DALL·E where you've got, like, digital asset management, you've got controlled and approved images, you've got workflows that support all that, you've got embed, embed functions, you've got usage functions reporting and analytics, or you can build the thin wrapper, uh, and there's just a massive difference between those two. And I think that's, like, what people miss when, uh, when they make the argument is, like, you... There's also such a thing, I guess is a very thick wrapper. And the thick wrapper is, is like when you've actually solved the user's problem end-to-end fully in a way that OpenAI never will. You know?

   4. HSHarry Stebbings11:51

      What does that look like then? Sorry, everything about, like, the thick wrapper and we unpacked that. What is required to be a thick wrapper?

   5. DTDes Traynor11:58

      I think you have to solve, uh, all aspects of the user's workflow.

   6. HSHarry Stebbings12:04

      Mm-hmm.

   7. DTDes Traynor12:05

      Uh, you have to fit into their tech stack and integrate where you need to integrate. You need to, like, solve for, like, uh, when I say all aspects of the workflow, I mean, like...... in, like, the hypothetical, like, DALL·E wrapper. Like, you need to solve, uh, you know, feedback on, on various images or maybe there's a bit of, uh, collaboration on the client side, where they want to discuss things. Maybe there's, like, prompt, prompt version control to see what we can generate. Uh, maybe there's an approval workflow. Maybe there's, like, uh... And then, you're into, like, asset management after that and then maybe there's hosting, maybe there's embedding, uh, all that sort of stuff. And then there's the reporting on usage, uh, like, how many times was this displayed and, and like, uh, you know, does it, does it... You know, if it's a marketing product, maybe, like, does it, does it correlate or, like, does it test well with, like, conversions or does it not convert when we use it or whatever? Like, that to me is, like, all of the shit that you would want around a tool-

   8. HSHarry Stebbings12:51

      Hmm.

   9. DTDes Traynor12:51

      ... if you were gonna say, like, "This is Canva but AI first," if you know what I mean? I think if somebody goes and builds all that and makes it make sense, uh, the, the dude round the corner who's got, like, the thin wrapper, like, he's got a lot of shit to build.

   10. HSHarry Stebbings13:01

       Sure.

   11. DTDes Traynor13:01

       You know, so I just think you're then back into what I would call, like, maybe the, the classic B2B arms race, maybe, but, like, uh, it's those who really understand the user workflow are the ones who will actually win this, not those who kind of, like, you know, straighten their glasses and say, "Academically, that's been possible for seven months now."

   12. HSHarry Stebbings13:18

       (laughs)

   13. DTDes Traynor13:18

       It's like, "Great, mate, but, like, people actually need to buy real software and use it for real things."

   14. HSHarry Stebbings13:22

       I think the hard thing for me is, like, given the horizontal application of, like, what a lot of OpenAI is doing and able to-

   15. DTDes Traynor13:27

       Yeah.

   16. HSHarry Stebbings13:27

       ... when you look at the releases they made the other day, there are a lot of applications, to be fair, like, I don't know, prosumer wealth management. (laughs) Yeah? You would never have thought, like, "Oh, OpenAI is gonna build a deep set of tools for them."

   17. DTDes Traynor13:38

       Yeah.

   18. HSHarry Stebbings13:39

       Well, actually, like, a lot of their changes meant that most of those companies that have been innovating in the last year have just been killed overnight.

   19. DTDes Traynor13:45

       I think that's probably true but, uh, what I would say is, again, going back to the thin wrapper or thick wrapper, like, if what you've built is, like, the wealth management advisor bot, uh, yeah, you're dead, right? Like, as in, it's gotta be, like, an actual... One of the GPTs that's just gonna be available, right? If instead you've built something that, you know, connects to your Robinhood or your eToro, connects to your bank account, uh, monitors your wealth, your exposure, you, you know, issues you case-by-case advice, monitors the markets for you, like, again, thick wrapper. Like, does the actual things you have to do to manage somebody's wealth and can have a conversation with you about, like s-... And suggest actions and next best steps, that's still, I would say, substantially beyond, like, what OpenAI would ever do. You have to bear in mind, like, a parallel here might be, like, look at what iOS ships with, right? It ships with a basic Notes app. If you want a really fancy one, off you go and you can go and buy Reflector or Bear or something like that. Like, it ships with a basic camera. If you want Photo Room, you have to go and buy Photo Room. It's like, you have to understand, like, OpenAI will hit some sort of minimum viable, like, kind of, "What's good enough for everyone?" You're never gonna make money, uh, filling in any of the gaps in the platform. I think they haven't gotten round to you. Like, I describe that as, like, you're on a train tracks picking up, you know, euros or dollar coins or whatever, there's a train coming. It's gonna hit you at some stage. Doesn't matter how rich you get, that thing's hitting you, right? However, I think, um, if you find an area where they're not gonna go that deep... OpenAI is never gonna put, like, 500 engineers going hard on wealth management and, like, you know, banking integrations-

   20. HSHarry Stebbings15:06

       Yeah.

   21. DTDes Traynor15:06

       ... you know. So, like, that's the thing. Like, that's... Y- you pick an area and you'll say, "Let's do all of it." Let not, "Let's just do a little tech demo science fair," let's nail the use case, you know?

   22. HSHarry Stebbings15:16

       No, I, I totally agree. Can I ask, when we think about kind of, you know, that train hitting you, a lot of people suggest the commoditization of LLMs.

   23. DTDes Traynor15:23

       Mm-hmm.

   24. HSHarry Stebbings15:24

       Do you agree with that as an argument? I mean, when A, we're seeing more and more upstarts coming-

   25. DTDes Traynor15:29

       Yeah, yeah.

   26. HSHarry Stebbings15:29

       ... like your Mistral's from France.

   27. DTDes Traynor15:30

       Yeah, yeah. Yeah.

   28. HSHarry Stebbings15:31

       Um, do you think we will see the commoditization of LLMs?

   29. DTDes Traynor15:34

       I mean, it would suit me if we did. Um-

   30. HSHarry Stebbings15:37

       Why would it suit you if it did?
5. 17:12 – 24:17

   AI Innovations in Business

   1. DTDes Traynor17:12

      AI chatbot that you... Sits inside the Intercom messenger and it will have conversations with your users and help them, uh, answer their support questions so that your support team don't have to. We generally remove all of the undifferentiated heavy lifting that a support team might have to do. And we do a lot beyond that as well. Some of the more complex queries, uh, will, will also get fully resolved by Fin. Fin also sits in your inbox and helps your support team, so if a day one support rep doesn't know the answer, they can ask Fin and Fin will give them the answer too.

   2. HSHarry Stebbings17:36

      Yeah.

   3. DTDes Traynor17:37

      Um, so, and Fin is an, is, effectively, it's a product by Intercom that uses an LLM. It uses GPT-4 by OpenAI.

   4. HSHarry Stebbings17:44

      Mm-hmm.

   5. DTDes Traynor17:44

      When it comes to, like, picking, um, the model, the requirements we have are, one is, like, trustworthiness. So, like, we need to be able to prompt it in such a way that it will not hallucinate, knowing that that prompting will occasionally cost us a valid answer, but it's really important to our customers that, that all the answers Fin says are at high confidence. A second thing is that it has to stay on topic. So, we don't want Fin taking opinions on current events. Like, so if you go to, like, a bank website and you say, "Hey, who's the president and are they doing a good job?"... a bank's customer suppor- support has no official policy on it, and Fin better not either, right? So there's a lot of, like, staying on topic that is really, really important and then it's, like, just depth and power of quality of conversation. So we have a set of metrics, uh, that we kind of look at and we, and we configure and tweak our prompts and tune as much as we can to, to get it as good, uh, as we can. And right now, like, we are, we are, like, in with OpenAI and we, we, we sit on GPT-4 and it's against that logic I say that.

   6. HSHarry Stebbings18:40

      And when you compared OpenAI to the alternative providers-

   7. DTDes Traynor18:43

      Yeah.

   8. HSHarry Stebbings18:43

      ... what did the test show you?

   9. DTDes Traynor18:44

      There's no, like, uh, there's no hard criteria I could offer you, like, "63% in this and blah, blah." It's basically a quality of conversation. Does it fail any of our, like, uh, hallucination tests? Does it fail any of our, uh, of our, like, trustworthiness tests? Can it infer its own confidence? So Fin will sometimes say, "The answer is yes," and sometimes it'll say, "Based on reading this article, the answer looks like yes, but have a read yourself." And sometimes it'll say, "I don't know." Uh, we, we care about the ability to do each of those things.

   10. HSHarry Stebbings19:08

       Was it closer or was there a wide-

   11. DTDes Traynor19:09

       Yeah, uh, it... Close and narrowing and, and, like, it's also a work in progress. Like, uh, all of these things are moving targets, right? Um, there are better

   12. NANarrator19:18

       Sure.

   13. DTDes Traynor19:18

       ... capability. And obviously, you know, we have been moving very fast in this space. Fin launched, I think, in June, maybe, um, July? Uh, it was in a sort of beta for three months prior to that. A lot happened since then, so, like, we haven't even gotten around to maybe testing the latest and greatest of all the providers, which are increasing in number. You mentioned Mistral. There's also a GLaMa, there's Anthropic, there's a, you know, Cohere. Like, there's, like, a whole chunk of them and, uh, and it's a bit of work to go around and constantly be f- trying to find out has anyone got... We only really care about better, right? Right now, we're not in cost optimization mode. We're just like, "Who's got the best?"

   14. HSHarry Stebbings19:52

       Well, it's interesting you said there about that because (laughs) I'm kinda thinking of, like, an AI LLM testing team.

   15. DTDes Traynor19:57

       Mm-hmm. Yeah.

   16. HSHarry Stebbings19:58

       Um, but Alex, who I had on the show from Nabla in Paris-

   17. DTDes Traynor20:01

       Right.

   18. HSHarry Stebbings20:01

       ... he was like, "The companies that win will be those that are able to transition between LLMs most seamlessly."

   19. DTDes Traynor20:07

       Yeah, yeah. For sure. I mean, I, I think that's a... Well, I don't know if I agree with the sentence, but I think it's a really important attribute to be able to transition from one LLM to another. Uh, because if somebody does, uh, y- unlock new power, you'll wanna be able to use it very quickly. I just, I don't know. I think winning involves more than just simply, uh, being agnostic about your LLM, but, uh-

   20. HSHarry Stebbings20:27

       I, I, I, I, I, I feel more and more like kind of Joe Rogan in terms of not knowing the answer and asking quite base questions. (laughs) Um, my question to you is can you not just use, like, seven different types? And what I mean by that is you're, like-

   21. DTDes Traynor20:37

       No.

   22. HSHarry Stebbings20:37

       OpenAI is great for the high regulatory clients that need precise, perfect answers every time, but actually then A- Anthropic is better for-

   23. DTDes Traynor20:45

       Yeah. All the way down to your home-grown cheapest one or like even within OpenAI's ecosystem, you can use GPT-4 Turbo, GPT-4, GPT 3.5 Turbo. You know, you unlike... They cost different amounts and, like, they're good at their... Like, um, some of them are faster than others, some of them are more accurate than others. So, like, yeah.

   24. HSHarry Stebbings21:00

       What's been the hardest thing about kind of the integration with OpenAI?

   25. DTDes Traynor21:03

       Uh, just, uh, it's a moving target, uh, i- is like, and that's not a, that's not a bad thing. They, they move incredibly fast. We can grow a lot of confidence in a particular direction and then they can unlock new capabilities and we're like, "Oh, I guess we should jump on this instead." That's definitely one of the challenges. Like, we're, uh, we have a great partnership with them, but, like, the- their team moves, like, we're, uh, we- our team moves fast, but they're, like, trying out new capabilities regularly. So...

   26. HSHarry Stebbings21:25

       Does it feel like the kind of unnerving, bluntly, being moved in that way by an external team?

   27. DTDes Traynor21:32

       You know, we are all in on AI, so we have to. We just have to be the best providers of AI for customer service. So it's not really an option. Like, if and when the world stabilizes a bit, like, we, I, I suspect we're still in, we're still in this S-curve. We're still in the spike. There will be a plateau eventually, right? Where, hey, most of the main problem areas have been fleshed out and we're starting to get a bit more sober now. I think that's when we can move into, like, a different mode of work. For as long as, like, you know, r- you know, our whole team watched Tuesday's Dev Day by OpenAI, again, it was like, "Let's get to work." Like, there's, there's new stuff here. There's new things. Customers are gonna have questions. We need answers. We need to have opinions. We might have to... Might change the roadmap. I don't consider it so much dependent on OpenAI. I consider it dependent on, like, the, the, the, how you say, the boom time of AI that we're in full stop. I liken what we're going through right now to, like, the beginning of the internet. It's, it's, it's, it's very big what's happening to both our world of customer service, uh, technology in general, and society in general. We don't have the luxury of sitting back and watching. I think if we were, like, a mass- massive, like, government sanctioned monopoly, we could be like, "Let's just sit this one out for three years and see what happens," you know?

   28. HSHarry Stebbings22:39

       Yeah.

   29. DTDes Traynor22:39

       Whereas in practice, I think this is much more like, hey, there's gonna be a lot of, uh, a lot of winners and losers and there's gonna be a lot of market share, uh, up for grabs over the next two to three years, and people are gonna be looking for who's got the good shit, and that needs to be in our app.

   30. HSHarry Stebbings22:52

       I have so many questions to ask. First one I have to ask is, like, you mentioned bluntly, obviously the team watching, you know, the announcements. How does the structure of the team change when you move from Intercom V1 or kind of the previous gen of Intercom-
6. 24:17 – 35:09

   Resource Allocation and Investment in AI

   1. HSHarry Stebbings24:17

      resource is allocated to R&D in AI today?

   2. DTDes Traynor24:20

      R&D in Intercom is about, I... These numbers are always-... catch me up, but I, I wanna say it's between 350 and 400 people or somewhere in that space. And then how many are AI? I wanna say, like, uh-

   3. HSHarry Stebbings24:29

      What, what percent is that? I'm just intrigued.

   4. DTDes Traynor24:30

      Oh, of the company?

   5. HSHarry Stebbings24:31

      Yeah.

   6. DTDes Traynor24:31

      We're 900 people, so-

   7. HSHarry Stebbings24:33

      Wow.

   8. DTDes Traynor24:33

      Yeah.

   9. HSHarry Stebbings24:34

      Shit.

   10. DTDes Traynor24:35

       Yeah. Does that s- Is that shit big or small?

   11. HSHarry Stebbings24:37

       That's big.

   12. DTDes Traynor24:37

       Yeah, I mean, we're like a technology company, you know?

   13. HSHarry Stebbings24:39

       What was it a year ago?

   14. DTDes Traynor24:41

       Slightly less. I'd say we just... Yeah, slightly less. We probably, we'd just gone through a riff and, uh, and I guess AI hadn't happened yet.

   15. HSHarry Stebbings24:49

       (laughs)

   16. DTDes Traynor24:49

       Like, you know? Like, uh, AI literally like, we're coming around to the anniversary at time of recording, right? But like, we've always been... We've always believed that like the best product wins. We've always invested hard in product. And, and then on your AI question, I said like it's mi- you know, I think our... We anticipate growing the team to maybe 50 or so.

   17. HSHarry Stebbings25:03

       You said about, um, uh, kind of br- uh, being akin to the start of the internet.

   18. DTDes Traynor25:08

       Yeah.

   19. HSHarry Stebbings25:09

       Uh, to the start of the internet, a lot of people lost money and a lot of the projects didn't work out.

   20. DTDes Traynor25:13

       This is your famous tweet you're gonna defend again, isn't it? (laughs)

   21. HSHarry Stebbings25:16

       (laughs) No, no.

   22. DTDes Traynor25:17

       Go on, yeah.

   23. HSHarry Stebbings25:17

       I'm purely asking.

   24. DTDes Traynor25:19

       Yeah.

   25. HSHarry Stebbings25:19

       Do we think 90% of VC dollars going into AI startups today will go to zero?

   26. DTDes Traynor25:24

       Do you have a sense of what percentage normally goes to zero?

   27. HSHarry Stebbings25:27

       To zero? Um, I mean, it depends what data sets you take, but people think it's a lot more than they actually are. Uh, I think traditionally to zero, zero is about 55%.

   28. DTDes Traynor25:36

       Okay.

   29. HSHarry Stebbings25:36

       There's like, you know, 30% make between zero and 1X.

   30. DTDes Traynor25:39

       Yeah, yeah. Yeah, okay.
7. 35:09 – 39:00

   Economic Impacts of AI

   1. DTDes Traynor35:09

      is genuinely like, delivering real value, I think those budgets are gonna stick around and actually massively increase.

   2. HSHarry Stebbings35:16

      Yeah. No, I, I totally agree. How those budgets are spent is depending on pricing models. We've seen obviously per-seat model dominates that.

   3. DTDes Traynor35:24

      Yeah, yeah.

   4. HSHarry Stebbings35:25

      Um, I have Myles Grimshaw on from Benchmark and he said actually consumption model will absolutely use it up and overtake it.

   5. DTDes Traynor35:31

      Yeah.

   6. HSHarry Stebbings35:31

      How do we think about changing pricing models in the face of AI?

   7. DTDes Traynor35:34

      I think there's loads of obvious applications like, um, our world most directly, like if Fin is indeed doing like 50%, then like will businesses buy as many seats? What we're seeing a lot is like businesses are choosing to keep the seats and keep the people and like just turn the dial on quality, so they just get really good at support.

   8. HSHarry Stebbings35:50

      Do you think that's because they want to be really good at support or is it 'cause they don't quite trust it enough yet?

   9. DTDes Traynor35:55

      It's a good question. I mean, like, a, another way of saying it will be like if in a year's time there's another market downturn, will they still hold firm on that? But I think that comes down to the degree to which they see great service as a competitive advantage, uh, which isn't the case for all businesses, but is definitely the case for some.

   10. HSHarry Stebbings36:12

       Yeah.

   11. DTDes Traynor36:12

       But like in general, I think his point about like, uh, consumption pricing outpacing or like, you know, ultimately taking the place of seat pricing I think is going to be true. I think a lot of work is gonna get handed over to LLMs over the next five years. And I think, uh, we'll, gonna start trying to like price against the work that's being done, not price against the seats or the employees. But just say, "How much is it worth for you to have all of your digital assets created dynamically?" Or, "How much is it worth for you to have like your customers get sub-second replies to common questions?" Like that's the actual right way to think about pricing in the future. I think it, the wrong way to think about it is, you know, you, you were selling work, we're not selling seats and we're not selling like, you know, incremental, you know, tasks completed. We're just like, "What's the work and what does it cost to get done?" And, and you'll see a lot of, a lot of these, uh, folks will start... Like when somebody actually does a good job of an AI coach or personal trainer or financial advisor, the person who does it really well will be able to justify...... pricing it similar to what the human costs, uh, because people have already been paying humans for this for a long, long time. And I think p- people will be comfortable transferring that fee to a bot because it's 24/7, it's multi-language, it's blah, blah. And obviously, they, th- and they can bring the price down as they see fit as well.

   12. HSHarry Stebbings37:25

       The fees are expensive.

   13. DTDes Traynor37:26

       Yeah.

   14. HSHarry Stebbings37:26

       Like, OpenAI is not cheap.

   15. DTDes Traynor37:27

       No, not at all.

   16. HSHarry Stebbings37:28

       It's predicated largely on the cost of compute.

   17. DTDes Traynor37:30

       Yes.

   18. HSHarry Stebbings37:30

       To be fair on them. How will we see the cost of compute change? Will it, you know, uh, align to traditional Moore's Law? How do we think about that, and does this all value accrue back to NVIDIA?

   19. DTDes Traynor37:40

       I don't have any, like, concrete answer here, but I can say that generally happens is, like, there is no tech really in history that has not gotten cheaper. There is no, like, piece of hardware or software that hasn't ultimately been ripped off by a cheaper candidate to offer eh- uh, lower prices to more of the market. No matter what way we go, we will see a kind of a fleshing out of the spectrum. So right now, I think we've got, like, high end NVIDIA powering high end LLMs, and we should assume on the fullness of time that they're gonna, like, continue to produce and they're gonna find more, how would you say, more market points that they can sell potentially simpler or ba- more basic chipset. And same with the LLMs. There'll be the fast LLM, and there'll be the accurate LLM, and people will make choices between them, and there'll be the, the c- cheaper and then the other one, and all that sort of stuff. We'll see the solution space fully explored, and in that exploration, uh, I think what you're gonna get is, like, yeah, I don't, I don't think NVIDIA in five years' time will be sitting here proud as the only people who can do a GPU, and I don't think that OpenAI will be able to, like, be the only people who can do LLMs properly.

   20. HSHarry Stebbings38:42

       Mm-hmm.

   21. DTDes Traynor38:42

       I think, uh, y- you know, there'll be winners and losers. There'll be, like, the best, but the best isn't necessarily the biggest market share, you know?

   22. HSHarry Stebbings38:50

       100%.

   23. DTDes Traynor38:50

       So, like, uh, y- you just have to realize, this is all gonna be a spectrum, and, like, people will find, people are basically gonna be finding where is the most dollars to be made, right?

   24. HSHarry Stebbings38:58

       Would you be a buyer of OpenAI

8. 39:00 – 49:53

   Corporate Perspectives on AI

   1. HSHarry Stebbings39:00

      at 90 billion?

   2. DTDes Traynor39:01

      I know that they have a funky thing where you can only make a certain amount of return back on your investment, as far as I'm, I'm, I'm aware. Like, I think it's, they have some sort of policy, like, after 100X, you have to sell back or something like that.

   3. HSHarry Stebbings39:11

      Wow, I didn't know that.

   4. DTDes Traynor39:12

      Uh, yeah.

   5. HSHarry Stebbings39:12

      But as a net new investor, would you buy at 90?

   6. DTDes Traynor39:14

      Yeah, just, just straight up on clean terms would I buy at 90, and I think they're doing a billion in revenue, or something like that, is what I've heard. I probably would buy a bit to be along for the ride, but I don't think I'd be, like, putting millions into it because I don't, uh, you know, it would, it would be a small phone check for me. I think on the whole, I'd make the bet that they'll pass 90 in value. I think that's doable.

   7. HSHarry Stebbings39:33

      But you gotta see multiples on that. You need to see a five X on those. So the question is-

   8. DTDes Traynor39:36

      Well, you need to see it boom.

   9. HSHarry Stebbings39:37

      ... is it worth 500 billion?

   10. DTDes Traynor39:38

       Yeah, yeah, to where it's... So, so you need to see multiples. I don't. I'm just throwing in fun money, right? Like, uh...

   11. HSHarry Stebbings39:42

       (laughs)

   12. DTDes Traynor39:44

       If I was an investor trying to, like, you know, make my name, would I go all in, like, on putting, like, I don't know, 10 billion into OpenAI for n- uh, 90 or whatever? I don't think I would, and the reason, uh, is that, um, whilst I think they'll pass 90, your question was could they go to 500 in, in the near term or, like, within the lifespan stage of the fund. The areas I'd be wary of is Amazon. I just think they're very... Amazon play this game well. So, like, I- I could see Amazon just, like, flat out, like, buying Anthropic and being like, "Let's just make this part of the EC2 cluster," and that's just a very easy route to market. And I think if OpenAI run out of, um, new vectors of differentiation, uh, and the commoditization starts to kick in, even for basic stuff, I think it'll just become an easier, "Why wouldn't you just use Amazon? It's already in the cloud. It's already virtually private." It's, you know, you- you can, like, leapfrog a lot of other adoption concerns. So I think they're one risk. Um, I think Microsoft are probably less of a risk 'cause of the pretty close partnership they have. I don't really know what Google has up their sleeve. I don't even know if they have sleeves at this point, um, but I suspect that, like, th- they won't be totally napping on this.

   13. HSHarry Stebbings40:46

       Have you not been impressed by Bard? When Bard came out, everyone was like, "Oh, open AI." It meant-

   14. DTDes Traynor40:51

       Um, I- I played with Bard a bit, and I wasn't, I wasn't unimpressed with it. It felt like another version of a ChatGPT. I'm more, um, underwhelmed by, like, their, like, their leadership in this area. Like, they don't... It feels like Google, who kind of owned all this at the start, bear in mind, with the original paper on Transformers, feels like they have, like, rescued defeat from the jaw of victories, in a sense, uh, from the jaws of victory. What we saw on Tuesday of this week, I don't know when this will go out, but, like, uh, from my mind was, like, Sam and to some degree even Satya, uh, out there, like, talking from a leadership position, from an authoritative position about where we are, where we're going, what's happening, uh, what the future holds, et cetera. And I don't think we've had that yet from Google. I think that's the sort of, to take a leading position, I think you need to, like, do that thing where you walk out on stage and say like, "Here's our mission, here's where we're going, here's where the whole world is." Like, "And we have a big wide-eyed vision." Like, think of, like, Elon Musk saying he wants to terraform Mars ultimately. Like, you know, like, it's, like, that is like leading a company in a certain direction, saying, "Here's where we're... Here's the end goal." I feel like Bard unfortunately felt like, "We, uh, we have to release this 'cause ChatGPT was getting a lot of traction."

   15. HSHarry Stebbings42:08

       (laughs)

   16. DTDes Traynor42:08

       You know, it didn't feel like, "We've actually cracked search again. We've reinvented ourselves all over again." You know, they need to have that sort of a, um, Jay-Z, like, allow me to reintroduce myself moment, right?

   17. HSHarry Stebbings42:17

       (laughs)

   18. DTDes Traynor42:17

       Where they come back and they say like, "Google Two is here." You know, that's what we need to have here.

   19. HSHarry Stebbings42:23

       I totally get you, but they don't want Google Two because the cost of doing it actually cannibalizes the core business, and so it's like-

   20. DTDes Traynor42:29

       Well, yeah, no.

   21. HSHarry Stebbings42:29

       ... going, it's Netflix and, you know, Blockbuster go.

   22. DTDes Traynor42:32

       N- now- now you're pulling onto the real potential problem, which is are they willing to risk it all to win at all, right? Like, are they willing to disrupt themselves or are they happy to take the, like, long slow decline into obsolescence or irrelevance or whatever, right? The long slow profitable decline with, like, predictable shareholder outcome, right? Versus saying, "No, we're gonna, like, take a massive hit to our own revenue, um, and we're gonna go, like, LLM first and the future of Google is actually gonna be a Q&A style thing, even though we haven't yet worked out how we're gonna monetize that." And, like, that's... It's a massive existential risk, and people have thought about, "Can you-"

   23. HSHarry Stebbings43:04

       What would you do?

   24. DTDes Traynor43:05

       Yeah. Yeah.

   25. HSHarry Stebbings43:05

       Genuinely. Like, it's so hard. I mean, it's easy to say it here, but if you are CEO of Google and you have shareholders on Wall Street-

   26. DTDes Traynor43:12

       (laughs) ... I'd be scrambling to find ways that... I kind of hate what I'm about to say, but I'm gonna say it because I think it's what I, it's what I would feel compelled to do. I'd be scrambling to find ways that, like, uh, companies can sponsor injections into the LLM. Who is the best footballer in the world? And the answer is clearly, like, Lionel Messi. But you could say something like, "According to Transfermarkt, the answer is Lionel Messi," right? Like, and that's what I mean by a sponsored injection, right? And then, like, you could, uh, pot- potentially, like, augment the Bard-style answer with such injections. So it's presenting you facts that it, it is kind of disowning, right? Like, it's kind of saying, "Hey, like, this isn't the LLM deducing this. This is just what we think. Uh, or what we've..." Sorry. "This is what we've been paid to say whenever we talk about this type of thing." And, like, I think that's the angle I'd... That's kind of the attack vector I'd go on. And I'd try and float that with all my big ad buyers and sort of say, "Hey, look. Let's be honest. The world is going to go to LLMs. So, you know, even if this doesn't work, we have to give it a lash."

   27. HSHarry Stebbings44:07

       (laughs)

   28. DTDes Traynor44:07

       And then, uh, and then I'd try and, like, get some sort of traction going for that type of ad model. Then I'd explain to the investors, "We have to, we have to move to this because the alternative is the company basically starts to be on a ticking clock," right?

   29. HSHarry Stebbings44:19

       I just don't know if you can move a Titanic-sized ship like Google in that direction so radically-

   30. DTDes Traynor44:26

       Yeah, I-
9. 49:53 – 58:01

   Tech Giants and AI

   1. HSHarry Stebbings49:53

      landscape assessment, but it's just, uh, Adobe. Everyone says, like, "Oh, Adobe've moved very fast," and, like, speed of movement. How do you think about Adobe's movement?

   2. DTDes Traynor50:00

      I think they've produced some really cool stuff as it relates to visuals, and I think if they play their cards right, they can actually expand beyond creatives, if you know what I mean. Like, Adobe previously or heretofore has been, like, the creative's tool. Like, as in, you have to be good at something. You have to be good at design or whatever to use Adobe. And, and that's even kind of true for their Figma acquisition, with the exception of FigJam. Like, you know, if you're not good at, like, drawing or whatever, the tool's not that rewarding to use. I think if they play this correctly, loads of people want to draw things but can't, right? Like, if I said, "Draw me a picture of a penguin," right, like, most people will make a ball of smut, right? So, uh, what... I guess, people who want to draw and then a much smaller subset of people who can draw, and then a much, much smaller subset of people who are very good at drawing is like... It's an example of the total addressable market and the serviceable market and, like, you know, how much, how much, how much of them can we actually charge money for? With the idea of, like, some of Adobe's features, where you just select a bit of text and you say, "Make the sky darker," or you select an image and say, "Take this person out of the photo." Uh, or you say, "Give me a rendering of, like, a leprechaun walking over a rainbow." And it does all this, (snaps fingers) like, instantly, to a pretty high degree of fidelity and, like, just nice polish. Like, certainly far beyond what an average illustrator could do, in terms of both out- li- you know, sorry, maybe not an average illustrator, far beyond what an average person could do, and potentially beyond what some illustrators could do. And it can do it instantly or, like, in four seconds or whatever. So I think we should see a world where more people are able to, like, weaponize the ideas they have in their head in, in a visual form than previously could. Previously, you'd just be like, "Oh, I have this vision in my head, and then whenever I, you know, go to pick up my mouse and open up, like, Photoshop, it falls apart." Uh, now you don't need that. Now, like, you know, thinking and, and rendering become, like, this new sort of skill set that you can have. So I think Adobe could see a significant expansion of people who care about Adobe tech if they play it well. 'Cause it's obviously gonna be like, oh, if you have an idea and you're struggling to visualize it, just use Adobe. Like, and you're like, "Oh, well here's what I'm trying to draw, and here's, I wanted a diagram, and I want this, and I want that." "Oh, you got that nearly right. Let me select this piece. All right, edit that so it's not like this, it's like that, then have her face in the other direction." Okay. It can do all that shit, where you're, like, the, at least the Tech Demos I've seen, the stuff that, like, Scott Belsky shares and all that, it's, like, phenomenal-looking stuff. They could have a really significant expansion of people who could use Adobe software, far beyond just creative geniuses.

   3. HSHarry Stebbings52:13

      Where does value accrue most? Is it the application layer or is it the infrastructure layer, and how does that change over the short and long term?

   4. DTDes Traynor52:21

      The only parallels we have to go by here are what's happened in other areas, right? Generally speaking, I think right now, a lot of value is going straight into the infra. Uh, like, as in we're handing it all out the back door to OpenAI, in, in this case, right? In your first question about commoditization, if that happens, then the value starts to, to reduce there, right? Like, as, you know, if there's more competitors offering the same thing, that's the beginning of commoditization. Or at least, you go from a monopoly, to an oligopoly, to ultimately perfect competition where, you, you know, and price goes down on each step. Um, e- excluding collusion or price fixing, which are usually illegal. Tha- that's how I expect that will play out. Unless, uh, unless OpenAI can continue to find mass market differentiation, which is always a hard thing to find, right? Like, as in the differentiation that all the customers care about, not just specifics, right? So let's assume over time, I s- I believe that, like, just, you know, the amount of people investing in this space means that I think the infrastructure layer will gradually get a lot more AWS-like, right? Price, uh, mixed with strength of, uh, product as a, as a competitive, uh, weapon, if you know what I mean, right? Amazon make good money off Intercom, you know? Uh, but they're also like, we're heavily... We're all in on them and we're very, uh, you know, we're very committed to them. Um, so I think, you know, that's what the competitive battleground... There'll probably be three or four big providers is my guess. It'll be probably GCP, AWS, maybe OpenAI directly, and I'm sure Azure. That's how, how I think the, that layer will play out. And they'll end up in pretty direct price competition with each other. Where does the value accrue beyond that? I think just differentiation. Value just generally follows differentiation around a stack. So, whoever has the stuff you can't get anywhere else can charge a margin state no one else can charge. Sometimes it means quality of software products. Sometimes it means quality of network or social network. Sometimes it means, like, we've got the most, uh, distributors in our marketplace. But, like, whoever has the differentiation that no one else can get, so, uh, is the person who can actually, uh, you know, basically charge the highest price.

   5. HSHarry Stebbings54:16

      The thing that worries me almost is, like, the best quality product does not always win. "Well, it's bundled into our existing package. We don't need a new vendor relation. We don't need new security and compliance." It's 85 of us. Fuck it, we're not technologists anyway. So product doesn't always win.

   6. DTDes Traynor54:31

      Correct. Uh, I generally think if you're playing in somebody else's playground and you're bringing one little thing, uh, even if you do have a, a reason why your thing will be definitely better, persistently and sustainably, it needs to overcome a lot just to be worth the cost of an extra tool in the stack.

   7. HSHarry Stebbings54:48

      Yeah, I totally agree. Well, with that in mind, if you project out now 10 years, will, will more value accrue to startups or incumbents?

   8. DTDes Traynor54:54

      With AI, a lot, um... A lot was made at the start of like, "Ooh, everyone's in trouble." And I remember when we first released our first wave of AI features, which was just AI in the inbox. It was things like summarization, expansion, change tone, like ex- you know, take this blunt reply and make it polite. Somebody, I can't remember who, somebody like one of the big dogs, uh-... you know, use it as a showcase of like, uh, "This is why there won't be new startups built on this. All the value will go to the incumbents." And I think the hypothesis there is that if the value is available on top from OpenAI, it's on top ... In our earlier example, it's on top for Mailchimp and it's on top for the new startup too, right? So the thin integration where you're shelling out for your differentiation to a third party, it does not create a wave of disruption. The disruption comes fr- as we were saying earlier, like it's wh- it's when you would build the entire thing differently in a world post-AI. If you put a gun to my head, I'd say, I'd say incumbents will probably do better out of AI than, like, startups thinking they can displace incumbents. But that's not the same thing as, uh, saying that, like, startups have no business in AI. I think there'll be loads of new stuff in new areas that no one's ever seen before 'cause of AI.

   9. HSHarry Stebbings55:59

      And my whole thing is like, as a, uh, general capital allocator and capitalist (laughs) who wants to make money, bluntly-

   10. DTDes Traynor56:05

       Yeah.

   11. HSHarry Stebbings56:05

       ... I think the upside is already priced into Microsoft or NVIDIA-

   12. DTDes Traynor56:09

       Yeah.

   13. HSHarry Stebbings56:09

       ... where I'm not looking at them like, "Ooh, there's a buying opportunity now-"

   14. DTDes Traynor56:12

       Correct. Yeah.

   15. HSHarry Stebbings56:13

       ... because they're so richly priced already, I think.

   16. DTDes Traynor56:15

       Yeah. I think that's, that's totally correct. I think, like, the areas where AI will deliver most value in startups will be things that just weren't possible before, are now possible. And I think, like, you'll see, um, companies like Synthesia will, like, will have, like ... They'll produce massive new things that just literally weren't being done before and they're, like, new categories and new jobs, new workflows, new capabilities. That's where, like, I think you'll have the kind of I'm sorry. Um, bluntly, I'm really gonna trouble for this, but, like, there's so many people going after Synthesia's space- Yeah.

   17. HSHarry Stebbings56:42

       ... that, th- they don't have a huge amount ahead of anyone else.

   18. DTDes Traynor56:46

       I, I guess, and like disclosure is I'm an investor at Synthesia, and I'm guessing you're not. (laughs)

   19. HSHarry Stebbings56:50

       Disclo- disclosure, I met them, and I think I was one of the first VC meetings they ever had.

   20. DTDes Traynor56:54

       Oh, wow. Okay.

   21. HSHarry Stebbings56:55

       Like, when I first started at Stride.

   22. DTDes Traynor56:56

       Yeah. Yeah.

   23. HSHarry Stebbings56:57

       Honestly, it was just like, "This is batshit crazy."

   24. DTDes Traynor57:00

       Yeah.

   25. HSHarry Stebbings57:00

       Terrible for me. Bad investment.

   26. DTDes Traynor57:01

       Yeah, yeah, yeah. Uh, you probably should've invested.

   27. HSHarry Stebbings57:03

       Thanks, man. (laughs)

   28. DTDes Traynor57:04

       Um-

   29. HSHarry Stebbings57:05

       Should've been 10 previously. (laughs)

   30. DTDes Traynor57:07

       (laughs) Um, independent of the actual company, I think what we're seeing there is a growth in, how would you say, AI-powered video customization to do lots of, like, things that people pre, pre- like to ... Bear in mind, the cost of video editing and all that sort of shit is so high. So let's say, uh, you have a video that explains how your product works and then you change the name of one of the features. Now, like, in the current world, that's a reshoot. In the new world, that's a, like, you edit a JSON file. You know, it's, it's we're, we're moving into a different fucking world here. I will defend Synthesia all day long, but like, uh ... And, and we'll, yeah, we'll come back in five years and we'll see. But I think what you'll see is, uh, the area of, like, AI-powered, say, video creation in that case, didn't exist three, four years ago, really. And it's gonna be a big area, is my guess. I think you're gonna see Adobe in there, you're gonna see, yeah.
10. 58:01 – 1:15:20

    AI in the World of Investing and Marketing

    1. HSHarry Stebbings58:01

       right? For me this, like-

    2. DTDes Traynor58:01

       Yeah, yeah. Well, and, and, uh-

    3. HSHarry Stebbings58:01

       ... sits natively in an edit platform where it's like, "Hey, Des actually said the wrong thing here."

    4. DTDes Traynor58:06

       Yeah. For sure.

    5. HSHarry Stebbings58:06

       Cut it in here.

    6. DTDes Traynor58:08

       An interesting thing for you to scratch on then is you're, you're simultaneously saying you don't see that they're a, an advantage over everyone else, and you also don't know why anyone else isn't doing it. So it ... I'd posit that maybe they are doing things that, uh, are hard to do.

    7. HSHarry Stebbings58:21

       Well, no, I'm not. I'm saying that I think it belongs inside an existing editing application.

    8. DTDes Traynor58:25

       Oh, right. Yeah. Fair enough. Like, I mean, everything at a- (laughs) If you're building anything in creativity at all, it belongs inside Adobe, like it's the cr-

    9. HSHarry Stebbings58:31

       (laughs)

    10. DTDes Traynor58:31

        It's the terminal buyer of creativity, right? Like ...

    11. HSHarry Stebbings58:34

        Or in tot- totally, or Descript or any kind of consumerized application you could-

    12. DTDes Traynor58:37

        Yeah, yeah, yeah. I mean, yeah. Descript to me, I think is, is much more focused on a very specific use case than, than say Synthesia is. I think they're gonna nail that use case. Descript's awesome.

    13. HSHarry Stebbings58:45

        Descript just fucking killed it.

    14. DTDes Traynor58:46

        Yeah, it's awesome.

    15. HSHarry Stebbings58:47

        I turn that down too.

    16. DTDes Traynor58:48

        Awesome. Yeah, yeah, yeah. A little mistake-

    17. HSHarry Stebbings58:49

        If you would like, I can send you all my turn downs.

    18. DTDes Traynor58:51

        Yeah, yeah, yeah. Totally.

    19. HSHarry Stebbings58:52

        And tell you, "You should just invest." Can I ask you, you're also a fantastic angel investor. Given what we've seen over the last year with AI, has that changed how you think about angel investing? Like, is it like, "Why on earth would you ... I'm investing in an enterprise software company now with no AI."

    20. DTDes Traynor59:06

        Um-

    21. HSHarry Stebbings59:06

        Is it like, "What are you doing?"

    22. DTDes Traynor59:08

        If I was investing in like a classic commoner gardener B2B SaaS enterprise company in this day and age, I would need to be quite convinced that AI has no role, or if it does have a role, the founders are all over it and going to lead. Otherwise, I just don't see why I'd take the risk.

    23. HSHarry Stebbings59:24

        Do you think you can accurately assess whether it has a role? And what I mean by that is just, like, open AI and ChatGPT and, and all the other places are s- are so good, moving so fast.

    24. DTDes Traynor59:32

        Yeah. Yeah.

    25. HSHarry Stebbings59:32

        I don't rate my knowledge that highly.

    26. DTDes Traynor59:34

        If you're building a web analytics product today, I'm just picking, like, a space. I don't really understand it and I don't think I have a dog in the fight in the web analytic space at all. If someone was trying to pitch me like, "We're gonna redo GA. Everyone hates GA. Go Google Analytics, it's gotten too complicated. So we're gonna do a better version of Google Analytics." I'm like, "Okay, I'm listening," and I can kinda see my first few questions, 'cause, like, this is an area that isn't obviously affected by, uh, AI in that it's reporting on data points that, uh, like, that I gathered on websites about did the user visit the checkout page or not, right? What I wanna see is, "Show me your reporting engine. Show me, have you considered how, like, chat UI could be relevant here?" Like, how, like basically instead of building, like, 150 million dropdowns to, like, you know, query by browser, type, by day of week, by this, by that, "Can I just type the query I wanna see and can you generate it? Can, in fact, can you go one step ahead of me and suggest new thing? Can you use AI to, like, power your reporting engine to suggest new, uh, reports I should be looking at that I haven't seen before?" I'd wanna be confident that the founders have thought through that and come to a conclusion. And it might be, "We're not gonna build any of that." Or it might be, "You know what? We should really go hard on, on, uh, like, natural language as a, as a query engine for reporting." That's, like, where I'd be like, "Yeah, okay." I'd wanna see that. And I, I'm kinda making this up on the fly, but, like, if they said to me, "No, Des, we don't think AI is gonna change anything to do with how people query, uh, analytic skills." At that point, I'm probably leaning out. I'm probably saying, right, "I don't know how you can be so confident in that." And, and I'm hon- if I'm honest, I actually think it will.

    27. HSHarry Stebbings1:00:53

        Yeah. Okay. And so when you look at your deployment today, has it changed much? Like, have you been investing more? Have you been investing less?

    28. DTDes Traynor1:00:59

        Uh, not more. Uh, not too much more. I would say I've probably, like, the last five or six checks have been, have written, have all been into companies that are either, either pure AI or heavily AI infused.

    29. HSHarry Stebbings1:01:12

        Of your deal flow state, how much is AI in the first tagline?... I would say for me it's 80%.

    30. DTDes Traynor1:01:18

        I'm probably more B2B SaaS exposed so maybe, uh, maybe I, I'm probably more like 60%. But yeah, if the question was like how much of it mentions AI in like sort of bulletry, or whatever, then yeah, 100%.

Episode duration: 1:32:32