#132 The impact of AI on digital experience (with Yali Sassoon at Snowplow)

[00:00:00] Lizzie: Hello and welcome to the Measure Pod by Measurelab podcast dedicated to the ever-changing world of data and analytics with your hosts, Dara Fitzgerald and Matthew Hooson. Between them, they’ve spent more years and they’d like to admit wrestling with dashboards, data quality, and the occasional Google Curve ball.

[00:00:32] Lizzie: So join us as we share stories about how analytics really works today and where it might be headed tomorrow. Let’s get into it. 

[00:00:41] Dara: Hello, and welcome back to the Measure Pod. I’m Dara. I’m joined, as always by Matthew. Hi, Matthew. How are you doing today?

[00:00:47] Matthew: Yeah, I’m all right.

[00:00:48] Dara: What level of existential dread are you feeling today on a scale of one to 27?

[00:00:55] Matthew: I’ve been doing a bit of. You might not know this about me, Dara, but I like doing a little bit of woodworking to just take me back to the old times and just put me in a state of flow. And I think that’s helped. It’s like meditation. I’m not sure 3D printing is actually woodworking.

[00:01:12] Matthew: No, I used, I did actually, I 3D printed a jig, which I used in my woodworking. So it all connects. That’s all it does. Yeah. Thematic. But yeah, I’m all right. How about you? Yeah, I’m good. 

[00:01:23] Dara: Yeah, I’m good, thank you. That’s it. I’m good. That’s all you’re getting.

[00:01:26] Matthew: That’s fine. It’s fine. I wasn’t actually interested.

[00:01:28] Dara: It’s just for the content. Yeah. You know, fe interest. Just to pretend we actually get on. Yeah. Yeah. We hate each other in real life. No, nobody’s, nobody’s buying that. No. Alright. Should we go on, get on with the news? Yeah. There’s a lot. 

[00:01:43] Matthew: Again, there is, I feel like we say that every bit of a theme now. Yeah. But there are tons and tons and tons.

[00:01:49] Matthew: And Dunno if we’ll be able to get through everything we’ve got written down here. But we can give it a try, we can give it the old college try as we say. 

[00:01:55] Dara: Yeah. I haven’t checked this, but I have just a feeling that our news section is getting longer and longer every episode. 

[00:02:03] Matthew: It is. And at the end I’ll just state why I think that is. But I think it’s ’cause more and more stuff is appearing, coming out. That it’s funny. That is relevant. Yeah. So, yeah, we can check about that at the end. But the top piece of news, I guess, has to be the Atlas browser. 

[00:02:17] Dara: Mm-hmm. 

[00:02:18] Matthew: That OpenAI released yesterday, 21st of October as we record this. Yep. There were rumors for a long time that it was coming.

[00:02:27] Matthew: Everyone kind of assumed it would appear on dev day, but it didn’t. Yeah. 

[00:02:30] Dara: Weird to have it announced like two weeks later. It must’ve been so close. I wonder if someone got Yeah a telling off for that. 

[00:02:38] Matthew: Maybe. Yeah. They were just about to announce it and it tried to, oh, sorry. 

[00:02:41] Dara: It’s not quite ready. 

[00:02:43] Matthew: And yeah, tried to murder everyone and they had to sort of just roll it back a little bit. But yeah, it’s a whole standalone browser and I did download it last night. I used it, it was available to me, a lowly British person to use and it kind of specified it as having three big features. One of which is just the fact that every time you, so every new tab is just a new instance of chat, GPT, essentially a new chat instance.

[00:03:10] Matthew: and you can type into it to sort of search and it will return very much like Google’s AI mode, but then at the top it’s got just news or just web search and images. So it feels like Google essentially, that they’re, they’re going directly up against there. And the idea is that it kind of, yeah, kind of has memory and context of everything you’re doing around the web that it can pull in and, and use to be useful to you throughout.

[00:03:36] Matthew: And you’ve got like a little sidebar if you want it at all times. Having church PT five there that can, you can talk about the page you’re on or you can do all sorts of things with it. I haven’t quite figured it out yet. I, I, every time I get these new, new things, I, I download it and I look at it and I’m like, right, what do I do? Yeah, what do I do? 

[00:03:53] Dara: Yeah, ask it. Ask it. I dunno. 

[00:03:57] Matthew: I’m trying to think what, help me, you know, help me use you. But then the big feature, I guess, is the agent. So agent mode can just sort of. Wander around your browser and do things for you, you can use it in logged in or log to out mode. So in logged in mode, it can go inside your accounts and do all these things for you.

[00:04:16] Matthew: So the example they used was they were making some just, just so wet, the examples they used on these, these things. But it was a Halloween pumpkin pepper recipe they ‘re going to do for the office part in like, I’m so excited for this. Shut up. Anyway, they looked at that recipe and it figured out all the ingredients it needed.

[00:04:34] Matthew: It went off to some supermarket and added all the bits to the basket ready for them, then came back and purchased it. So it kind of can just go and perform actions for you. And another example was they were planning a party, which again, they were so excited about, and it looked at who had replied and who hadn’t, and sent an email and added notes to a doc and things like that. I assume it was a pumpkin themed party? It was a haunted house party. 

[00:04:59] Dara: Yeah. Loosely, loosely related, yeah. 

[00:05:01] Matthew: Loosely Halloween related. 

[00:05:03] Dara: Yeah. I, yeah. Well, I need to. We need to come back to agent mode. ’cause I just wanted to mention a couple of the simple use cases of the just, you know, just the, the browser in general.

[00:05:14] Dara: And it is sometimes the simple things, isn’t it? ’cause I, I, I’m not sure why, but when I kind of heard it came out, I was like, well, okay, great. But you know, okay, I, I get it. You can use chat g bt within the browser, but it’s sometimes the kind of simple things that you don’t realize are necessarily a problem until they’re resolved.

[00:05:31] Dara: And one of the examples they gave was, you know, if you, this is a pretty common thing if you have a whole lot of tabs open. not something I would ever do. Obviously I’m very highly organized if you’ve got a lot of tabs open and then you can basically just say to it like, open up all those tabs that I looked at last week with all those different holidays.

[00:05:50] Dara: I was thinking of booking your shoes, I was thinking of buying. And they’ll do it. Or you can, you know, that. Certainly defeats the purpose. You just end up with a lot of tabs open together. But you can get it to kind of summarize or say like, tell me all those things I was thinking about and then tell me where I could find them and tell me which ones are the best reviews or whatever.

[00:06:07] Dara: So that research thing, I think we talked about this few weeks ago as well, and we were, you know, that laborious, slightly frenzied research process when you’re trying to find a product online and you get analysis paralysis and you end up with a thousand tabs open. This should, you can just offload some of that cognitive load.

[00:06:25] Dara: You can just say, you know, and even it doesn’t have to be what you would like to look, you know, summarize every pair of shoes I’ve looked in the last month or whatever. and it will just do that for you and you don’t have to keep going back and finding them, remembering what side it was on. So it is stuff like that.

[00:06:41] Dara: And then the other one, which you know, is a bit of a time saver, but it’s still, these things do count. Is that the moment where you’re having to, if you want to take something from the web and then ask, chat to, to question, you know, you’re either. You’re like getting the URL and pasting it in or you know, if it’s an image or something, you’re copying it or downloading it and then putting it into chat GBT.

[00:07:00] Dara: Whereas it can just see everything you’re looking at in the browser and you can directly ask it questions about that. So it’s going to, it will quite, quite dramatically change that browsing experience, I think. 

[00:07:12] Matthew: Yeah, I think it’s getting used to those workflows, I suppose, isn’t it? Yeah. ’cause it, it’s so much muscle memory and years and years of copy paste.

[00:07:19] Matthew: Copy paste, yeah. Copy paste and just moving around and, and we talk, we do, we do kind of cover it a bit in today’s podcast about what we think the future of sort of the web and, and browsing is going to look like. But I definitely feel if you can get into the swing of it and, and, and use it in the right way, it will, it will be a massive sort of time saver.

[00:07:38] Matthew: I think. I think part of that as well, just to go back to the ENT thing, is trying to get it to do things and not sit and watch it. Which is the tendency you kind of want to do, to just watch this thing move around the screen and do things for you. But I think the overall idea is you can just say, right, I’m going to set this over here, go and buy me a thing I need for this shop in list.

[00:07:55] Matthew: And then you can wander off on other tabs and start doing other things and it, and it kind of keeps running in that tab and it’s still doing whatever it’s doing in the background. And you can have sort of three or four of these things doing various things in different sites for you while you’re all over here doing something else.

[00:08:08] Matthew: And it’s, that’s a strange thing to get used to as well. Just, just having something to wander around and perform tasks for you while you do all the tasks. So, I don’t know, stare, stare at a wall blankly. 

[00:08:20] Dara: I think that’s it. It’s maybe, it’s unusual because we don’t know what to do. It’s like, oh, but I was going to do that task now I don’t have to.

[00:08:26] Matthew: I better find something else to do while the agent is doing my task for me. Yeah, there, there was actually an article, and I think we talked about this. It was Mark Edson who was, who mentioned it, but there was a BBC article that said there may be more burnout approaching because of air. ’cause you could imagine, even though you’ve handed these tasks off to sort of three or four different agents are off doing them and then you are off doing something else, there’s still a little bit of context locked away in all these tasks that you are having it perform and do.

[00:08:52] Matthew: And it’s just that tendency for people just to fill up any gap with more productivity. That could, could be a problem. 

[00:09:01] Dara: Yeah. We might be just spreading ourselves thinner and thinner. You think you’re moving on to something else, but as you say, you’re actually keeping a little bit of, because you need to remember, there’s no point in going off and doing this thing and then you’re not remembering that it’s done.

[00:09:12] Dara: So you’ve got to kind of keep mentally, keep track of that and be like, right, I better go back and just check that I haven’t ordered 8,000 pumpkins by mistake for my, for my pumpkin latte or whatever it is. Yeah. So you do have to still have some sense of what you’re getting these agents do and that does loop back into.

[00:09:30] Dara: That of the point, which is, and we’ve talked about this so much lately, but you know, this double-edged sword of having this capability and you know, if you just let it loose and say off you go, do all these things for me, there’s a lot that can go wrong. Yeah. Both in terms of, you know, your own kind of mistakes or the mistakes that the agent makes where it misunderstands you, it does buy 8,000 pumpkins or, or then the more malicious stuff like this kind of prompt, ejection and, you know, hidden code on sites or in emails that then get acted on by these agents.

[00:10:02] Dara: So it will be so easy to just think, and this is going to happen, people will just say, yep, I want an easy life, so I’m going to just turn everything on and let the agent go off and do everything for me. And then those same people, and I’ll probably be one of them, will then be complaining, saying, I can’t believe it went off.

[00:10:17] Dara: And yeah. You did that. 

[00:10:20] Matthew: So there’s going to be so much pumpkin spice lattes and I’ll be drinking, drinking pumpkin juice for the rest of my life. 

[00:10:26] Dara: It’s going to be really hard to get that balance right between being able to delegate to agents and, and, and having that take away some of the misery of daily life.

[00:10:35] Dara: Yeah. versus, you know, like what, what the risks of that are. and even again, a bit like, you know, reading the release, the release article, whatever from OpenAI, it was very similar to the one that Anthropic released around their Chrome dev. I forgot what it was called last week. You have to remind me again.

[00:10:52] Dara: Copilot. Two episodes in a row. Yeah. So a couple or Chrome autopilot, something like that. Yeah. But it was the same, a lot of the same language. They’re talking about red teaming all the different problems and they’re talking about how they can only do their best to put these measures in place, but there’s always going to be a risk.

[00:11:08] Dara: And it’s kind, that language is becoming normal now. It’s like all of the big AI companies are basically saying, you know, you are, you know, it’s like leaving your car. You know, it’s like, well what’s that thing they say in car parks? It’s like cars are left at the owner’s risk. Yeah. Or whatever. It’s a bit like that.

[00:11:23] Dara: It’s like, here’s this, you know, here’s this thing we’re giving you. But it could go really hard. It might kill every chance person. Yeah. But, you’ll be 10 x more powerful. Yeah, yeah, yeah. More say. Okay. Well that sounds good. I’ll go for that. That’s fine. Yeah. So it’s going to be really, really interesting to see how that like evolves in terms of like the capability versus what you’re giving away.

[00:11:45] Dara: And then there’s the other aspect of, you know, talking about giving away, you know, you’re also sharing so much more with this one platform or, or you know, multiple ones if you’re using different LLMs and you’re just giving all of this sensitive context about your life. And you know, if you’re starting to give access to bank accounts and Amazon accounts and you know all the different logged in accounts you have. The risk is getting more and more vigorous in proportion to the utility of it. 

[00:12:13] Matthew: It is, I mean that, that, one of the, one of the features of this is that you can, I don’t know, I know how they were controlling it, but in the demo it kind of stopped at purchasing then and just presented him with the basket.

[00:12:25] Matthew: Whether he just set it up for it to do that or Yeah. Whether it would, if, if you just ask it to go and purchase it would just go all the way through. You would assume so. ’cause you have the AgTech commerce protocol that they fully seal the other month. The other month. It was like two weeks ago.

[00:12:39] Matthew: And you, and you can take over from it and say like, I’m going to carry on going. I, I played with it a little bit this morning and it wasn’t, didn’t blow me away, but I was asking if to do weird things like I was on LinkedIn trying to get it to cheat on a game for me and it just kept on asking for hints over and over again for like two minutes until I went, until I took over and I tried to get it to go and collect data layer information from a site which had.

[00:13:00] Matthew: Didn’t seem to be able to do, I just don’t, I just don’t think you had access to the, to the dev tools and things, which potentially is a rea is because of the things you’ve been talking about there, about the risk not to have access to those things. 

[00:13:10] Dara: I don’t know, but dev tools was something mentioned at the bottom of that, article saying that that’s what the, I don’t, I don’t know exactly what they meant, but they did say in terms of like what’s on the roadmap.

[00:13:20] Dara: They did talk about better dev tools. So yeah, maybe that is, but it, you’re probably right. The reason why it’s probably further down the line is because of the safety concerns. 

[00:13:29] Matthew: Yeah. It’s fascinating. It doesn’t feel very far away now that you could with the, with combinations of MCP servers or the computer use stuff from moving around UXs and it just being a copilot across things that you could just, some of the things that may have been a mainstay sort of three years ago of working through e-commerce journeys and designing data layer specs and testing that they’re in place and working correctly as you move through journeys and also like that.

[00:13:56] Matthew: Pretty much just be tasked and handed off, in a completely automated way. It feels, if not already possible, pretty much there, which is amazing really to think about. 

[00:14:06] Dara: And even whether though this is maybe a, a light spoiler, but I, I’m not going to spoil anything, but this comes up or something to do with that comes up in today’s, with today’s guest around whether these journeys are actually going to become more personalized or individualized as well.

[00:14:21] Dara: Yeah. So like I will too, if, you know, if you go and buy something through chat, GG chat, GPT browser or whatever, is that going to be the same user journey that I will have? Because if it knows about us, we might not have to go through that same journey. So, and then they have a knock on effect of what that does in terms of data collection and, and, and analysis and trying to optimize. And, it’s all going to change. Basically. Everything’s going to change completely. 

[00:14:46] Matthew: Yeah. And on that front, so in, we’ve all just had our heads turned by the Atlas browser. Like sort of looking around the Atlas browser and doing a bit more reading about it, I realized two massive other pieces of browser AI related browser news had happened that we kind of passed us by.

[00:15:04] Matthew: So the first is like about three weeks ago. Complexity released a browser called Comet, so another AI player is getting into the browser game. I’ve not seen a huge amount about it, but they released one. That’s as far as I’ve got. And Atlassian the makers of Jira and Confluence. So the, you know, the people who are really in the enterprise work world or the browser company.

[00:15:27] Matthew: The browser company makes the browser I use called art, which is supposed to be sort of an AI first type browser thing. They bought them along with, they’ve just released and they just released, vo dev, which is their ai, CLI dev tools and things. It seems like Atlassian are making moves to, to try and.

[00:15:46] Matthew: Carve out space in browser and CLI. Seems like this, those two, Steve internally said today, those two seem to be pretty common battlefields across the AI players browser and CLI. 

[00:15:56] Dara: Well, you should be quite happy with the perplexing news because they’re tackling lop, they’re, they’re going to break, they’re going to bring down slop, bring down slot.

[00:16:05] Dara: This was there. At least that’s part of their kind of marketing spiel. But there, this is in response to the kind of low quality content that’s being, I dunno that’s being prioritized, but you know, that, that’s being kind of pumped through some of the, some of the browsers. So there, ’cause I think the comm browser was $200 a month or something and they’ve decided to just roll it out for free to everybody.

[00:16:28] Dara: ’cause I think the thinking is for it to, yeah, for it to truly get people on board, it needs to be accessible. But this is their angle on it, is that they’re tackling the, tackling the slope. 

[00:16:38] Matthew: You think you’ve got a feel now that in the same way that Google was sort of jolted into action with just gen LLMs in the first place when OpenAI first came about, you can imagine a loss of people are going to go and download that.

[00:16:51] Matthew: Yeah. That browser. They’ve got a big audience. Google has been, you know, poking around the edges of co-pilot and, and they had this Jarvis copilot that was supposed to be in Chrome and things like that. That hasn’t really been realized yet. I wonder if this is again, going to have to just push them into action before they start to, I mean, they are ridiculously dominant in the browser space.

[00:17:13] Matthew: Mm-hmm. But so is Netscape, so Internet Explorer, once upon a time, that’s it. Exactly. You know, you can’t rest on your laurels, can you? Nope. What else did I have on here? Claude skills. Yeah, that is, it’s like these ways of giving Claude context about specific operations and things you do in the form of like these YAML files, I believe.

[00:17:34] Matthew: So you can be kind. You can almost sort of get it to, it’s almost like prompting a little agent that will follow either your branding guidelines or write things in a particular way in templates. And people were, some people in the corners of the web were hailing this as bigger than cps. It certainly seems like we’ve been using something internally called the Bmad method, which is a, like a, a series of different agents that you can use with Claco or other CLI or other, a LMS that will run through systematically various templates and checklists and, and prompts.

[00:18:07] Matthew: It’s got, and then it’ll hand one task off to the next task to the next agent and another one to the next agent. And it proved very powerful. So I think this is a way of being able to define and chain together these things. I’ve not had a huge play other than in Bmad, but it 

[00:18:22] Dara: Do you, where do you do the ’cause I haven’t at all. I mean, I saw them, I was, I was away. and I saw that, I saw you post actually that the skills have been released, but I didn’t, I haven’t played around with it yet. But where do you actually go? So it’s this, it’s not a Claude Code thing, it’s in the, it’s, it’s Claude Desktop claude main cloud core claude.

[00:18:43] Matthew: Apparently, basically you can use this, you can use it anywhere. It doesn’t have to be Claude. Okay. So yeah, you, within, within Claude, you can upload these files and sort of set them as skills, but the files themselves are essentially very structured prompts, I believe. So you could design one of these files and then take it to open AI and just okay, give it that file and then, then it’ll start acting in that way.

[00:19:05] Matthew: So I don’t, I think what Claude has done is sort of integrated that into Claude and it’s got the capabilities and skills, and then you’ve got all these different skills that you can turn on and off. So the examples they give are like brand guidelines, internal comms, canvas design, something different.

[00:19:21] Matthew: Different little agents with skills that can perform tasks and it has like a chain of thought thing and it can move through different skills and pass one thing off to the other. So yeah, it, it, it is a superpower sort of taking inspiration from Bmad to make our own sort of marketing team almost within Claude Code and Claude using sort of similar stuff.

[00:19:40] Matthew: And it is pretty cool, pretty powerful what it can do. Yeah, there’s cloud code on web, which again is, you can use cloud code in the web and set it off to do multiple tasks in parallel and it can work when you’re not there and it can perform tasks on actual git repos and all these sorts of stuff, which is, I think, pretty similar to what Codex is for, from OpenAI.

[00:20:02] Matthew: Okay. And what, what JUULs is for Google? So I think they’re, they’re similar products, but I think Claude has now entered that space, which would be interesting to have a deeper look at, ’cause I’m a bit of a cloud code fanboy. And then this is a bit of a. Is this lighter or is this high? Dystopian still?

[00:20:18] Matthew: Channel four, channel four in the uk have, think it said the world’s first. They presented a pretty much entire Channel four program on dispatches with an AI presenter. And it was kind of the gimmick of the thing. I think the dispatchers were about AI and AI taking jobs. At the end they revealed that this presenter wasn’t a real person and they kinda showed all the shots and then the presenter being sort of generated into them just to sort of prove the facts that AI is pretty good. Hang on, hang on. It was quite, quite interesting from when I saw it. 

[00:20:47] Dara: Too many questions about this. So is it, is this kind of like an AI slap, I think, do they actually generate an AI person? Yeah, person and voice. Oh. So we are getting a step closer to replacing ourselves on the measure pods. 

[00:21:01] Matthew: Yeah, it was pretty convincing. I don’t know how long, how long it took, but there’s a video, we’ll add the video that’s, there’s a video somewhere that I found at a Dropbox of the thing, or go and watch Shuttle Falls dispatches. That will probably show it. But yeah, it was pretty convincing. We’ll add it to the show notes. Interesting. You can imagine more and more of these, especially now, they get realer and realer. Yeah. The cost of doing that compared to the cost of paying for a presenter for however long it is. Significantly different. Oh, that’s good. It’s not nice is it? But ultimately everything comes down to money. That’s right. But why? Why is it capitalism baby, that’s the one. 

[00:21:37] Dara: Yeah, that’s it. That is interesting. Channel four. I mean, that’s going to set the cat among the pigeons. I think it’s going to, and they’ve done it. 

[00:21:43] Matthew: They’ve done it to point it out. I think it is. It’s quite a matter of using it to say like, yeah. AI’s taking jobs.

[00:21:49] Matthew: Even the presenter who says could even take my job because I’m not real. And then it flickered in and then showed that he wasn’t a real person. Yeah. but AI can’t take AI’s job, can it? Maybe Claude can take open AI or open AI can take em. Ai, yeah. Yeah. And then they’ll start, maybe there’ll be AI unions of different agents with different skills before long striking.

[00:22:09] Dara: That’s interesting. Yeah. We, yeah. Include that video in the show notes. I think that would be nice too, we need to stop terrifying people though. 

[00:22:16] Matthew: Yeah. I’m trying to look down on this, see if there’s anything good. Anything positive. I was trying to find it, I was. I went googling positive AI names earlier, but I couldn’t find anything that was recent.

[00:22:28] Matthew: I mean, you could argue all this stuff has positive spins on it, you know, information, things like that. Yeah, 

[00:22:33] Dara: I think, I think especially, yeah, the, the persons, I think, yeah, the, the, the browser related news there is always, I mean, it’s like, well, it’s like with any new technologies, isn’t it? It’s powerful and therefore it can be dangerous.

[00:22:43] Matthew: Yes, absolutely. There is also, lemme do two more, I’ll do two more quick AI ones, and then there’s a couple of just good old fashioned Google analytics BigQuery pieces, Braze. So last week we were talking about open AI. PT and Braze has created an OpenAI SDK, so you can allow developers to track customer engagement with OpenAI apps.

[00:23:11] Matthew: So that’s the first kind of first company I’ve heard of getting into that space of sort of tracking things within the AI OpenAI interface and seeing how companies are interacting with it. Interesting, interesting thoughts around what data do they get? Mm-hmm. From the context of the chat and or not and yeah, lots of questions there.

[00:23:27] Matthew: And similarly, Salesforce have partnered with OpenAI to surface, like customer query, sales records and customer conversations and build Tableau visualizations, things within ChatGPT. So they’re kind of entering that game as well. And then the two hall cleansers at the end, as it were, and this came out, this was a little while ago, but Google GA four data is now available to pull in from BigQuery.

[00:23:51] Matthew: So you don’t just have to sync from GA four into BigQuery, which is just sort of the raw data and is, you can’t backfill it, but with. This, you can, it’s essentially using the API, so you can pull out specific reports, design your own extractions that get pulled into BigQuery and, and do backfilling and things like that with just using the BigQuery data transfer service.

[00:24:12] Matthew: Mm-hmm. so some of the questions we had were like, what does it do in terms of the, the attribution settings and all the rest, all those settings you can sort of turn on and off. Is instantly reflected within GA four. Where does it get that information from? ’cause that’s not necessarily available in the API, how does that work?

[00:24:30] Matthew: So we’re kind of doing some investigation internally to figure out exactly what goes on there. 

[00:24:35] Dara: So I guess it’s limited by the API tokens and limits. 

[00:24:40] Matthew: Yeah. I wonder, I I do wonder if they’re doing, I mean we, we’ve designed it, we’ve built a couple of custom extractors in our past for UA and for GA four, and we, we’ve had to sort of stack the tasks and do, to, to get around the query, the API query limits and batch them so we don’t end up with sampling and things like that. So I’d assume it’s all kind of built in to Yeah. The transfer service. But we’ll see. 

[00:25:02] Dara: I would like to include a bit of trustee GA four news. Yeah. It’s been a while. 

[00:25:06] Matthew: Yeah, exactly. I think that’s, that’s a lot. Think that’s it. That’s a lot. Yeah, on the subject of a lot, I just wanted to churn through very quickly in three seconds why we think these, these new segments getting longer and longer.

[00:25:18] Matthew: But in since October, what has happened, and we’ve covered most of this, if not all of it, on this pod open AI agent kit, open AI apps in chat, saw a two or slot talk five Chan, DBT merge clocks on it for 2.5. Claude Haiku 2.5 a Gent commerce platform. Nano banana, Chrome dev tools. MCP GA four MCP BigQuery, MCP called Chrome Pilot BigQuery Data Engineering Agent Gemini Enterprise Friend AI device Meta AI glasses.

[00:25:44] Matthew: Claude Agent SDK Gemini computer use, new Claude features and updated GA analytics, Google Analytics agent within Google Analytics for itself. And that’s not good in anything we’ve covered today. That’s not that much really. Well, we can up our game. 

[00:26:00] Dara: Yeah. Yeah. I think we need to, I wonder what the longest podcasts are on Spotify. We need to go for that record. 

[00:26:04] Matthew: Yeah, I think we might be getting there this week. 

[00:26:07] Dara: We are entering Joe Rogan territory. We just need to up the saltiness of the news as well. Yeah, no, that is an incredible amount and it’s probably going to continue to, to increase. So, and even that we’re missing things, so.

[00:26:20] Matthew: Yeah. Yeah. Even including my hope, in my hope is that there’s dev days and things like that, that tend to be in and around the autumn, that maybe we calm a bit, but yeah, God knows. 

[00:26:30] Dara: Alright. that’s it, that’s it for our four hour news section. So, our guest today is Yali Sasoon, who’s one of the co-founders and is the CTO of Snowplow. So we had a really good chat with Yali. I started with a bit of a confession saying I don’t know why we didn’t try and line him up sooner. and it took Mark Edmondson to recommend him recently. But Yi’s great. He’s a really smart guy, really interesting. Did a lot of good. Views. We asked a lot of our kind of, you know, trademark kind of future looking questions.

[00:27:05] Dara: and he had some interesting takes and I think one of my key takeaways was how optimistic his view was. ’cause we do tend to lead towards the, you know, dystopian existential threat kind of views. I thought Yali had a really optimistic, but also quite practical view on how things are taking shape.

[00:27:27] Dara: And it was interesting to hear kind of how they’ve evolved at, at Snowplow and how they’re taking advantage of the rapid pace of change and the, in the space we’re in. 

[00:27:36] Matthew: Yeah, no, I think you’ve captured everything there. It was a really good chat. And, yeah, another, another eyeopening one. Got lots of interesting takeaways for sure. Enjoy.

[00:27:48] Dara: A very warm welcome to today’s guest who is Yali Sassoon, co-founder and CTO of Snowplow. So Yali, firstly, welcome to the Measure Pods and thanks for agreeing to join us. Thanks for my pleasure. Thanks for having me. Really good to have you. I’m actually surprised and, and, and a bit ashamed that we haven’t got you on before, and you were recommended recently by Mark Edmondson, who was a recent guest of ours.

[00:28:12] Dara: And I said I was going to, yeah. So damaged. I can’t believe, I haven’t thought of, of yali up to this point, so it’s good to, good to have you on now. but for the benefit of our listeners who maybe are less familiar with Snow flower or, or don’t know You, we always get our guests to introduce themselves.

[00:28:28] Dara: So you can take as little or as long as you like and to give a whistle stop or a more detailed background. but just a little bit about who you are, your journey leading up to what you’re doing 

[00:28:39] Yali: today. Alright, let me, let me try and not take too long to introduce myself, so thanks for having me on.

[00:28:45] Yali: Dari Matthews really appreciated it. I am, I’ve always worked in. Data all through my career. I started my career as a consultant working with companies in the shipping and logistics and energy industries, using data to optimize operations. And I went to PWC as a strategy consultant, saw how data could be used, strategic leads to drive organizations, and at a kind of a higher level.

[00:29:13] Yali: And then I got my sort of break in digital. I always wanted to work in tech, working in a startup that was at the time called Open Ads and renamed to OpenX. And at the time we were the fourth real time ad exchange. This was way back in sort of the two thousands when ad tech was still cool. It’s hard to remember now.

[00:29:35] Yali: Now it’s a, that’s a, a little bit awkward talking about, talking about ad tech. But at the time it was, one of the early VC backed companies with, that was headquartered in London. Skype was the big one that everybody wanted. To work out. A number of the ex Skype team moved to open ads since I was lucky to be one of the early teams there looking after data and, and getting to work with digital data and digital data at scale.

[00:30:03] Yali: So that was very exciting. I left there. So that’s where I met my co-founder, Alex. He was on the engineering side and I was on the kind of commercial side. And we spent a few years, really between 2008 and 2012 having too much fun running our own consultancy. So it’s just the two of us. We built a roster of quite interesting companies, mostly based around London, working at the intersection of data and tech and strategy.

[00:30:36] Yali: And that gave us the freedom to sort of experiment with building things on the sides. And, one of the things that we were really sort of excited about. Was this kind of explosion in digital experience that happened with the advent of the iPhone. You know, this is going way, way back, absolutely in the early 2010s and that period where there were apps, everything.

[00:31:04] Yali: And so suddenly there were all these things that we used to do without an app or a website that suddenly you could do with them. And as data people, that was really exciting, exciting ’cause you could download from these applications very, very detailed data describing how people collaborate with their workers and productivity tools.

[00:31:24] Yali: All slur or data sites all manage their health and fitness apps. And the data was amazing. It was so broad. It was so deep. You could collect it at a crazy scale. and with our kind of consulting hassle, we really wanted to. Help our customers use this really new exciting data to much, much better understand their customers and deliver them a much, much better service.

[00:31:51] Yali: So let’s give a really simple example. We were working with retailers who’d had lots of success doing the kind of Tesco club card things. So they, seeing what people bought and profile them based on that, and then give them personalized offers based on their purchase history. But actually it is really interesting though, not just what they purchased for, they tried to purchase and didn’t.

[00:32:09] Yali: That’s an obvious marketing opportunity. And that was the kind of thing that was in the behavioral data that you could collect on the digital platform. You can’t necessarily collect physically, you don’t get to see when someone walks around the store, what they looked at and thought about buying but didn’t buy.

[00:32:24] Yali: But you can do it in an online store. And when we tried to do that with our customers, we found it really difficult. That data was locked in Google Analytics or Omnisure, ’cause it was called at the time. and so it was hard to get at. And so. We, Craig, this is taking a lot longer than I was expecting.

[00:32:44] Yali: and so the idea hit us to take these amazing big data open source platforms like Hadoop and this, these new cloud web services things, Amazon web services thing that let you spin up these really complicated, tools really easily and only paid for the servers while you were running them. And build a platform that would let any company build effectively a clickstream data warehouse.

[00:33:11] Yali: And at the time there were few organizations with clickstream data warehouses, but they cost huge sums of money. ’cause you needed to teaser or vertical these other very extensive technologies, big networks and lots of expertise. And so Snowplow was this open source project that was formed to make that, to kind of democratize access to that and make that really easy.

[00:33:31] Yali: And so we created that for ourselves and we implemented that for some of our customers in 2012 and around that, a big. Open source community sort of sprung up, other people around the world were like, we want to do that too. And they set us up as a big community in Sao Paulo and in Berlin and in Sydney and, all these different places, users started popping out.

[00:33:57] Yali: So we carried on running the consultancy, but we spent more and more of our time working on the tech until we both sort of went full time at Snowplow in 2015. We’re like, oh, that was just going to grow this, this, this company. And so the, the, the focus for Snowplow as a company was really helping people build a great data foundation with their clickstream data, with their, what we, what we now call their behavioral data.

[00:34:23] Yali: and that journey has continued until today. So the technology’s evolved a lot. At the beginning we were very focused on analytics use cases, so marketing attribution, cohort analysis, customer journey analytics, and so forth. And then more recently we’ve been very focused on machine learning use cases and actually helping organizations take this data and use it operationally in their build actual data products that personalize their search results or do other valuable things with it.

[00:34:58] Yali: And most recently we’ve been looking at, how we can help organizations that are building customer facing newgen experiences, so chat bots on their websites, but also generative UIs and other things, what they need to do with the data, how this data can power those sorts of applications. So in that time between 2012 and 2025, I guess a long time, the way organizations have used that data has changed very, very dramatically.

[00:35:30] Yali: And our software and our product sets had to keep updating to enable our customers to. To deliver all those different, different use cases, which means haven’t been born at all on that, on, on that journey. It’s been a very fun space to be in. But yeah, that’s, sorry, that’s me in a, in a nut. Not, not really in a nutshell.

[00:35:52] Dara: It’s always the hardest question. It’s, it’s almost like we’re cruel, we say to people, come on, introduce yourself. And then it’s like, oh, right, okay. How do I do this? But, no, look, that’s really, really interesting and, and does tee us up for what we’re, I guess we’re going to focus more on the, the, well both the current, you know, the kind of present day, but also more of the forward looking, both for Snowplow and just in general, your, your kind of thoughts on where everything’s going.

[00:36:14] Dara: But before we do get into that, I’m just kind of keen to ask maybe, one or two questions about the earlier days. I suppose the first question I had was, did you think when you conceived the idea that you would end up focusing on it full-time? 

[00:36:28] Yali: You didn’t. No, no, no. We thought we’d, the first version of Snowplow we built, less than a day.

[00:36:36] Yali: So we thought we were, sort of. We were kind of surprised we got something that worked so quickly and our intention was to use it, but it was, it was just a tool for us. So we were sort of surprised that as many other people wanted to use it as they did. And then right at the beginning, we very naively thought, well, it’ll take us probably six months to turn this very, very rough data pipeline into something that’s more polished.

[00:37:05] Yali: And then we’ll be finished and then we’ll go back to doing the scene that we loved doing, which is actually using the data that the, that it, that it, that it delivers. And what, what we had factored was how fast the space was evolving and how much, every capability you build, the kind of the demand for additional capability just grows and grows and grows.

[00:37:29] Yali: So that was a big learning. And there’s still a big, there’s still a part of me that is a little bit south that I don’t get to. To, to do my, to do any data analysis really. That’s, that, that, that’s prior to Snowplow, which I really enjoyed. But I’m, I’m not complaining 

[00:37:42] Dara: Again, do lots of other things.

[00:37:44] Dara: And my, my other question, which you, you kind of answered in a way at least. So you’ve said about how from the very early days there was a kind of key user base and there were popping up all over the, in, you know, different parts of the world and it was kind of an open source community around it. But from a maybe more of an enterprise perspective, did you find it hard in the early days?

[00:38:04] Dara: Was there a lot of people who were just tied in with GA and, or Omniture and just thought, oh, we, you know, we, we see what it could do, but we’re just going to stick with what we know. or did you actually find that there was a lot of frustration even then, and people wanted a little bit more control over their data?

[00:38:22] Yali: It’s a really good question. So the landscape changed a lot in the time over that period. So at the beginning. Most people I talk to about Snowplows were like, why would I do that? What’s the thing that lets me do that I can’t do in ga? I had a lot of that and I was going to a lot of conferences talking about what we were doing.

[00:38:50] Yali: I went to like all the measure camps and, and others and and it was always the question that I got, but the indus, the industry changed. And so when GA came along and had the integration with big queries, that was a big moment. So quite a few people pinged me and were like, oh, now the snow cloud’s finished.

[00:39:13] Yali: There’s no, there’s no need for you guys. We can get the event level data ourselves. But actually that was a real blessing because it was Google coming out and saying, Hey, there’s this whole world of stuff you can do if you have access to the underlying data, and we are going to let you do that. So that was a big, big.

[00:39:28] Yali: Shopped in the arm. That sort of changed the way people thought about things. The other thing that really changed, a couple of other developments that really changed the way people saw things were the growth of cloud data warehouses, Redshift, Snowflake, and then BigQuery. Before that, we were really writing the data to files in S3, and you’d have to spin up an E EMR cluster.

[00:39:53] Yali: It was just a horrible experience, and one that most analysts would never abide by. As soon as you had it in a to load, to load it in a cloud data warehouse, the data was really easy, accessible. You could stick Tableau, any familiar analytics tool on, on top of it. and so the birth of cloud data warehouses has meant suddenly you had organizations with data platform teams going around and systematically trying to centralize their data and let people do, kind of combine all their customer data.

[00:40:23] Yali: Sets all in one place. The product data sets in one place. And so that was a big thing and that sort of led to the modern data stack, which has since kind of ODed. So there’s quite an interesting arc or cycle we can explore there. And then the other thing, the other sort of technological change that changed the way people were thinking about this was the whole Kafka real time data processing, stream processing.

[00:40:49] Yali: Can we use this data in, in real time systems? And it, I feel sort of, the cloud data warehouse stuff happens really, really quickly. The real time data processing stuff is happening much, much slower. So Kafka and came out all all, I feel like all the real time technologies, we saw all, all this innovation in 2015, but even today they’ve not been widely adopted for like stream processing, kind of analytical and machine learning workloads is still like very rarely.

[00:41:22] Yali: Done. So the whole, the whole adoption of that has been much, much slower. And a lot of my time at Snowplow, especially with our new product signals, has been helping organizations get those kinds of real time personalization, use cases, a kind of in product, lives, which is surprising ’cause all of that technology has been available for a long time.

[00:41:45] Yali: But I think it’s a lot less accessible and a lot harder to use than all the tooling around the cloud data warehouses. It’s another area where you just can’t, you can’t do that stuff with Adobe. You can’t do that stuff with Google Analytics. So, to come, sorry, I’m really not being very brief today at all.

[00:42:04] Yali: It’s come back to your question. It, the, the, the objections have changed over time. They’ve always been organizations that have been happy with those tools and that’s fine. But the number of organizations that have been looking outside those tools has grown and grown and grown over this series. And that’s really been driven by some of them.

[00:42:21] Yali: Factors, some of the technological factors that, that these new possibilities that the technology enables an organization is looking to realize that, and them not being well supported by, by that tool set.

[00:42:33] Matthew: It amazes me still to this day how many people are still on the per bit as in getting the data into the warehouse and centralizing it and joining it up before we even begin to get to the real time processing.

[00:42:44] Matthew: It feels like there’s a lot of people out there that are still at that stage. I dunno how much you see that maybe that’s part of the, the lag of you having to go out and help people figure out that real time piece because often it’s marketing teams and people who are almost creating their own warehouse off on the side to, to get around central bureaucracy.

[00:43:03] Yali: There is a, a lot of that, and I actually often forget it because those aren’t the, the, the, the, the people who are struggling with it are not the ones that we’re typically talking to. Yeah. so there’s a whole part of the market that we don’t really address. It is probably much bigger than the part of the market that we address.

[00:43:22] Yali: But, more and more people are coming over the, in, in defense of those organizations. There’s a lot of legwork to do before you see a return on that investment. Mm-hmm. So cost conscious companies are quite rightly wary of taking it. And I, I think one of the things that we haven’t been good at as an industry is, and we’re certainly trying to do with all our customers that it’s, it feels like we’re swimming against a Tide, is show how you can take a big infrastructure project like a data replatforming and look to, deliver some real return on that investment.

[00:44:04] Yali: Really, really early on, long before you’re finished with the replatforming. There are no, there are no enterprises that have got the patience to wait for. Suits are nuts, replatforming from start to finish, you’ve got to find use cases that you can unlock really, really early in that journey, that surprise and delight the stakeholders to give the project momentum.

[00:44:26] Yali: And I see too many, too many organizations sort of rolling up their sleeves and going like, right, we’re going to redo our eventing infrastructure. We’re going to, our tracking’s all over the place, our data’s all over the place. We’re going to rethink this from the ground up and get everything right the first time.

[00:44:44] Yali: And I, I love the enthusiasm, but it’s, it’s, it’s, it’s not good, it’s not a good approach. Yeah. It normally ends badly. And, I don’t think we, we probably spent enough time talking or 

[00:44:56] Dara: thinking about that. So has your, your typical customer, and I mean the, the person within the, within the organization or the team department within the organization, has that changed over time?

[00:45:06] Yali: It’s only started to change quite recently actually. So. In the very early days, it was a real hodgepodge of difference, we used to call them internally, data MacGyvers. They’re kind of mavericks within an organization that had the autonomy to go off and spin up infrastructure and get stuff working.

[00:45:29] Yali: And they asked for forgiveness, not permission. and they often got amazing things accomplished in incredibly short periods of time, and they’re often running ahead of their organization. And we’d go and meet them and realize that, look, the colleagues didn’t realize really what they were up to and what they were, what they were doing.

[00:45:46] Yali: So that was very much 20 12, 20 13, 20 14. I still know some of those people. They’re brilliant, brilliant minds. As the, the cloud data warehouse, seeing sort of, we, there, there’s a sort of like, head of data platform that became the primary, our primary stakeholder in most of the organizations that we, that we worked in.

[00:46:09] Yali: And they really were. They had a remit within their organizations of building shared data infrastructure and assets and data products and so on that could power the business. And they were similar, they were either the same people or they worked very closely with the team that were trying to drive, self-service analytics and business reporting.

[00:46:28] Yali: And then they also collaborated with the kind of machine learning teams, it started springing up to try and try and build valuable models and deploy them around the business. The change, the real, the real change that we’re seeing is, is, is only just happening now. So a couple of weeks ago we launched signals, which is really the second to only.

[00:46:48] Yali: Our second product is our core slow plow customer data infrastructure, like the, the pipeline from collecting, collecting the data from all these different platforms and these SDKs and doing that in a, a scalable and robust way and streaming that data into, into the data platform, into cloud data warehouse and stuff.

[00:47:09] Yali: That was the, if that was the first product, signals is the second product. And Signals is really geared towards product and engineering teams that are starting to deliver different experiences to different users. So there’s this shift that’s happened in product engineering. Your average product engineering team has always built one product for all the customers.

[00:47:33] Yali: And if you’ve got two customers in the product that needs to do two different things, that maybe one of them is there to buy shoes and another one’s there to return the shoes to the wrong size. It’s on the customer to navigate around the product, to figure out where the, where they need, where they need to go to do the thing they want to do.

[00:47:51] Yali: But everybody gets the same experience. And that’s been the paradigm since. digital products sort of were, were born, but now AI is making product managers and products engineering teams more generally think about, we’ve got these different users, we’ve got different. Needs and expectations and jobs to be done when they’re in our tech.

[00:48:12] Yali: How do we deliver all of this, the right experience? Like why don’t, why don’t we try to meet them where they are instead of making them do, do all the, all the work? And it’s AI that’s kind of driving that shift in mentality and that’s really exciting. But it means that suddenly these teams need infrastructure they didn’t have before.

[00:48:35] Yali: And, some of that infrastructure is kind of all the AIOps stuff and they’re great solutions in markets, that do all of that, you know, sync Databricks, think AWS SageMaker, think Google Vertex stuff. They really are a brilliant and growing number of providers in that space. And, but around that, they need real time customer signals.

[00:48:59] Yali: They need to know. Right now, who is this customer? What are they doing? What do we know about them based on the historical data? What can we see about what they’re doing now and what are the best guesses we can make about what they’re trying to accomplish? And then across, across that signal, or we can use that signal to deliver really great experiences.

[00:49:19] Yali: And if those experiences are agentic, that kind of context engineering, can we get real time customer context and give it to the agents, the agent can be that much more effective? Or if it’s a machine learning use case, can we compute features that summarize what this user’s doing right now so that the machine learning model can, can return the best predictions right now?

[00:49:43] Yali: And, and that’s what signals is, is built on top of the core pipes that we’ve had forever. That is a total change in the So map.

[00:49:54] Matthew: This is kind of related to the first part of what you were saying there, but. Having that, that shift to the sort of cloud owner, to someone who sort of owns that centrally or, that new cloud infrastructure.

[00:50:05] Matthew: Do, do you see any shift in that in the, in the fact that because of the techno, the stuff that’s being released, so I’m talking about AI and LLMs and the speed with which new features and capabilities seem to be coming aligned, is that central ownership that, that everyone has to feed through potentially going to cause bottlenecks and there’s going to be, need more, need to sort of democratize the ability to, to handle that, that sort of cloud data and, and, and put it in the hands of everyone.

[00:50:31] Yali: It’s a really, really good question. So I think even before the AI stuff came along particularly, there were, people were recognizing that there was a cost to having a central data platform team, that that team ended up becoming a bottleneck. And that was very much the motivation for, a max data mesh and the idea of you could have a centralized team that they’re really.

[00:50:58] Yali: About policy and building reusable systems. And then you want the actual ownership and management of the data to be distributed in the team. So even before the AI stuff came along, that was a ship, a ship we saw. And even before sorting out the Mac Roher paper on the data mesh, we observed this journey that our customers would go on where they deduct Snowplows as part of an effort to centralize their data stacks.

[00:51:24] Yali: And often they’d say things like, our data’s a mess. We’ve got like a hundred different teams. Each of them use Google Analytics inconsistently, you know, we go into BigQuery. No one can make head or tail when in any of this means, and none of it’s documented to da, da da. so they do that, but then they’d find that the centralized system was too rigid and so the pendulum would swing, swing back.

[00:51:45] Yali: and we spent a lot of time building, we call it our data product studio, so our tooling to help organizations manage either the centralized governance, so they decentralized knowing that they probably moved between the two. Over time. Then when you lay the AI stuff on the top, you find that that infrastructure has some buses and some cons.

[00:52:08] Yali: So the plus is you’ve got this well governed data set. If you’re building applications on top of it, that can be a great foundation, especially with the longer term data. So if you’ve got your customer data, if you’ve reconciled it across different sources and you’ve got really key customer tables, when we implement signals for you, we can reach into the data warehouse and go and fetch those records and pull them, pull them back.

[00:52:35] Yali: And it’s really fast to be able to pull really highly relevant, highly actionable information. Two challenges though are often that aren’t. The case and the time to onboard new data into that environment is too slow. Then also, the computation of that data is typically happening on some kind of cross.

[00:52:59] Yali: It’s no good for real time and real time is real. Time is really important because we are all, so also the, the one, the one thing we all, we all have a finite amount is seconds in the day and whatever my wife tells me, we can only concentrate on one thing at a time. When we think we’re multitasking, we’re just switching between tasks.

[00:53:22] Yali: You can’t really do three things at once, so if you are an organization and you have someone’s attention, that’s gold. You could lose that to any second. You have no idea if this is like what notifications are appearing on this. Users like handsets. Where they are, you know, what, what could happen the next, next minute and the costs of driving them back into your website or your app could be really, really high.

[00:53:48] Yali: So you’ve got to make the most of it. And so you really got to, you’ve really quickly got to figure out what is this person doing and what, what’s the best experience I can deliver to meet them where they are. And that realtime flow is not it, it’s just not po the, like the modern data stack. It’s just too slow and it’s too slow to onboard the data.

[00:54:09] Yali: It’s too slow to process the data. So all of that is perfect for that kind of stream analytics, whether technology’s been around since 2015, that organizations are still, are still sort of struggling to do that. And that’s a lot of what we’re, we’re, we’re sort of solving for with the, with signals is, is, is the real time piece.

[00:54:27] Yali: So helping, come out with a way that organizations can hook into their deep store of information on their customers. But also providing an alternative path to really quickly figuring out who this is and what they are doing now so I can take action in milliseconds, not in minutes or hours. How 

[00:54:45] Dara: has the, I’m just thinking about what, you know, the, the point in time when you were, well not a single point in time, but when you started building the core product and right up until probably signals came out or you started working on signals, the pace of change with the technology that you were working with will have been a lot more manageable.

[00:55:07] Dara: So how are your internal, how have you adapted to work on signals, which is going to be something that’s just going to be, you know, breakneck speed compared to what it would’ve been like. So how have you kind of changed as a company in terms of your, you know, how you’re actually managing and building that product as such?

[00:55:25] Yali: That’s such a good question. It’s, we’re, we’re on our own journey, so the, you’ve seen this, there’s, there’s never. It’s never felt like the data space has moved slowly. We’ve had the whole real time thing happen. We had the cloud data warehouse stuff happen and now we’ve got the, we had all the machine learning stuff and now we’ve got all the AI and agent stuff happening.

[00:55:48] Yali: So it’s always, we’ve always been in an industry that’s moved really quickly, but nothing has moved as fast as AI, especially this sort of wave of agen AI that has since 2022 when ChatGPT launched. And it’s been interesting. Snow plows 140 people I think at the moment. That feels like a big company to me ’cause I’ve only ever worked at the list.

[00:56:15] Yali: and I sort of remember the early days. I know it’s a small company relative to a lot of others and it still feels like it’s been a struggle for us to adjust to. That pace of change. It’s, it’s like just, just keeping on top of all the, the developments in the tech is more than an all-time job.

[00:56:36] Yali: So no one really, no one really manages that. I mean, the amazing thing about AI is obviously that it’s an incredible productivity multiplier for us internally. And so we’ve been learning how to use Claude Code and agents to help us develop, develop our products faster. And that’s been a, that’s been a shot in the arm, but it’s, it actually, it’s not been a shot in the arm.

[00:56:59] Yali: It’s been something we’ve, it’s been a new set of tools we’ve had to learn how to use to drive, to drive us more effectively. And we’re still very much on that journey. We’ve seen a ton of benefits, but not as many benefits as I, I think the hype suggests, I think we, there’s just more learning. There’s more learning to do and it’s, it’s hard to learn from other people ’cause you’ve got to cut through a lot of them.

[00:57:26] Yali: The hype. So that’s definitely been part of it, the delivery side. The other part of it is I think a lot of us at Snow Plan, myself included, spend a lot of time in, in market with our customers and prospects, but also with the next generation of startups because they’re where you see the new patterns and the new approaches because they’re smaller, they’re more agile, they get to experiment with the stuff the rest of us will only experiment with later.

[00:57:55] Yali: And sort of building for the future that they’re already living in is a much safer place to be. ’cause the pace of change is just so, just so enormous. I feel like we’ve got one foot in, in the presence of our current customer base, and you’ve got to keep them happy and help them solve the problems they want to solve, you know, next month to next year.

[00:58:14] Yali: But we’ve got another foot with these other guys who it feels like living on another planet. And we’re seeing what’s coming over and, and that means we can see what’s coming over the horizon, the horizon and build to that. And it’s, it’s uncomfortable, but it’s really fun. It’s, there’s so much creativity right now, so much creativity and innovation.

[00:58:32] Matthew: Speaking of that, I guess this, this is going to tie together a couple of points, but you, you mentioned right at the top that the battle old days of like 2007 to early 2000 tens where there was an app for everything and, and you know, you were trying to figure out that, and obviously the pace of change of what’s coming up now, I’m interested in what your reaction is to like the open AI integration and surfacing all of those, the, the e-commerce and the all of the rich sort of modal pieces just directly within a chat interface.

[00:58:58] Matthew: Just generally what your thoughts were when you saw that and what, what popped into your head?

[00:59:02] Yali: Oh, that’s something I spent a lot of time thinking about, so I really recommend everybody who reads, I think Brian Balfour from Reforge has written some really good, good, good posts on it. and I’m going to try and summarize some, I’ll try and summarize.

[00:59:16] Yali: I think they’re the sort of foundation that, that have shaped a lot of my thinking, and I probably won’t do them justice, but there’s this sort of, there’s several stages that platform companies, go through when they’re growing, where they kind of, open, open the gates and play nice with all their partners and everybody rushes to work with the hot new thing.

[00:59:38] Yali: And it’s a new channel. It’s a new way of doing business and it’s the future. And it’s, it’s this kind of, and you see, so when it’s successful, you see this whole ecosystem of companies grow up around it and that carries on while the company is building a kind of a competitor. Mo it’s a competitive moat.

[00:59:58] Yali: It figures out what its motive is and it’s driving that moat. And then when it sinks, the mode is high enough, it will bring the, bring the, the doors down on that ecosystem. It will start competing with its most successful partners. It starts ratcheting up. The cost to play at that point. They become an incredibly valuable company as open AI is going through like that, that kind of, that’s the journey that they went on and Twitter and Facebook and LinkedIn and Google search and all these different platforms.

[01:00:31] Yali: And what we’re seeing with open AI is that whole journey happening in a much, much, shorter way. So ChatGPT is an amazing technology. Like these ages or specifically the, the LLN is an amazing technology that can use it. There’s so many opportunities to use it, to rethink so many things that we take for granted fundamentally changes.

[01:00:58] Yali: What a, what a computer can do so fundamentally can change everything. And in ChatGPT you have an application where with, without integration, which lets you do. Huge amounts. I know, you know, people who get all their medical advice, their legal advice, their, use as a brainstorming tool like that, it’s just in, it’s just sort of incredible.

[01:01:16] Yali: You look at what they’re doing with shopping, for example, and the deal with Walmart like, last week and letting, letting people, it’s great for them. People can do more in chat. GPT is actually a great place for a variety of shopping related things. So it’s like researching what product you want to buy and then so now you can actually buy that product in, in, in ChatGPT without ever leaving it.

[01:01:43] Yali: And for OpenAI their modes, the ChatGPT Moat is memory. So if you are one of those people and there’s more and more of them around the world who kind of live in ChatGPT and you’re doing your, it’s your therapist and it’s your financial advisor and it’s your lawyer, and it’s helping you with your work, it has an amazing knowledge of you.

[01:02:06] Yali: That probably no other organization in the world has, and it can use that knowledge to do all kinds of things. And we like them, so the obvious example is sort of recommend, suggest products that you might want to buy. But even by saying that I’m sort of stuck in the old world because in the old world that was like the most valuable thing you could do with the information that Facebook gleaned and you could figure out what or someone might buy and what they might vote for.

[01:02:33] Yali: And, and both those two pieces of data are very, very valuable and to be monetized very, very easily. I strongly suspect that there’s going to be a lot more high value stuff that you can do with the ChatGPT data. So open AI are, are, are building this amazing competitive moat and they’re really turning ChatGPT into a super app.

[01:03:02] Yali: it’s, it’s conceivable that you might spend more time in ChatGPT than you might do in the browser as it becomes more and more powerful. And then even if you are in the browser, open ai think planning to release a browser with ChatGPT and there’s a chat, GPT will still be there, see everything you’re doing and help you every, every step on the, on the way.

[01:03:24] Yali: So it’s very clear what the benefits to open AI are of bringing all this activity and all these partners into their ecosystem. And what the partners have to think about really long and hard is what does that mean? so that they can go and they could partner with OpenAI and they can launch their own applications and their ChatGPT marketplace and they can.

[01:03:53] Yali: Try and carve out the space for themselves in this new platform where everybody knows that door’s going to come down and either they’ll be checked out or they’ll be made to pay a very high price to stay there. And so they have to think, no, they, they have to do that. I’d argue they don’t have a choice because that’s where the whole world is moving.

[01:04:14] Yali: But they also have to build a strategy that gives their customers a reason to engage with them outside of chat, GBT. Otherwise they’re going to be like Zinga with Facebook, all their eggs in one basket. And, it’s just too high risk, a business strategy. So I think there are important implications for them.

[01:04:35] Yali: And then from a customer perspective, a kind of consumer perspective, I think there are lots of benefits to a, a really powerful agent or set of agents that can do all of this stuff. That it’s really fun being a chat GP. Customer and seeing all this new capability come on online and find new, new uses for it.

[01:04:55] Yali: But it’s, it’s, we’ve been worried about what data, big organizations that don’t necessarily have our best interests at heart have done with that data in the past. We should just be terrified about the possibilities. We, it’s open air.

[01:05:10] Matthew: That’s kind of the argument that’s making. If we had a Brian Clifton a while ago and he was talking about the change in privacy and how, how, how individuals were starting to get a bit more clued up and were taking the power back and were pushing back on.

[01:05:25] Matthew: Privacy and, you know, the newer generation are saying no to this and more ad blockers and all that sort of stuff. My argument pretty much echoes what you just said is that most of what you were getting out of allowing these com companies to have more of your data was added to personalization. Pretty much.

[01:05:42] Matthew: But the, the idea with these LLMs is the more data they have on you, especially if they end up working in like assistant and agent capacities, they get better. Like you’re getting a better product by them knowing more, more about you. So it’s, it’s, it’s a really hard thing to sort of see all of this unbelievable tech that looks like magic and not engage with it, and not give it more and more information outta you to unlock all of these, all of these new features.

[01:06:07] Yali: YY yeah, it’s, in some ways it’s, it’s really similar to what’s, what’s happened in the past, but just, 10 x if you think about how, like the therapy example is terrifying, is that people who are sharing their innermost, deepest, deepest thoughts. I don’t, I, I, I’ve, I’m fundamentally an optimist and I think the open AI guys, are, are doing what they’re doing for at worse commercial reasons, which is, I, I have no problem with, and at best, there’s lots of idealism and, wanting to solve like lots of the world’s problems with this, this, this new text.

[01:06:46] Yali: I don’t think that, people, but you, a bad actor, could absolutely use that information to do some real damage, to manipulate people in all different kinds of ways. So the stakes are much, much higher than they were in the previous paradigm. Much, much higher.

[01:07:03] Matthew: Yeah. So Sora itself was, I, that’s where I tipped over the edge of, so it got everyone’s best interest in heart’s heart, and when they created Slop Talk in this form of Soah, that was indistinguishable to me.

[01:07:14] Matthew: I was like, Hmm, that, that’s a, a very social, a bit of societal danger there for profit, but. 

[01:07:21] Dara: Yeah. Anyway, so shift. No, I was just going to say, shifting a little bit from like, what the end users should or might be concerned about. What about you, YALI and what at, at Snowplow, when you see open AI trying to bring more and more into ChatGPT, how concerned are you as a company that that’s going to lead to less and less data available for clients to capture and therefore maybe need a product like Snowplow?

[01:07:51] Yali: I’m not, I’m not too worried about that. I think we and a whole bunch of other organizations have a huge interest in making sure there’s a thriving and open web outside of, ChatGPT. And I don’t think, and I, I, I don’t think that’s going to, that always has to be the case, that even open AI needs that to keep training models.

[01:08:17] Yali: There has to be some, mm-hmm. That sits outside chat, GPT or chat, keep eating, eating itself. So, and then the other thing is, at least at the moment, is chat. GPT is just chat. So what we’ve learned is that chat is very powerful, it’s very flexible. It’s a great interface for a whole bunch of user experiences, but it’s, it’s just one, it, it’s probably, it’s, it’s just one type of user experience pattern and they’re probably others that are much better suited to solving other, other problems.

[01:09:00] Yali: So I always, I always like to give the example of travel ’cause it’s, travel industry is an example of one where once upon a time you walked into a travel agent and they would give you this kind of white glass where, I dunno if it’s white glove, but they’ve really sort of stepped you through the options and.

[01:09:18] Yali: You rate your, your, your trip with them end to end. And that was very nice that you didn’t have to do much work. And it was quite fun and enjoyable, but it was really expensive and you, there was no real competition in, in, in terms of you had to take or leave what the agents gave, what the travel agents, the human agent gave you.

[01:09:39] Yali: And then we went and we learned with the early internet how to build our own itineraries, our own trips. And you buy your flight here and you buy your hotel there and you look at this travel guide and this block to figure out how you get from one place to another. But that was a huge amount of work. So agents could bring us back to that original model that you probably want an agent that can generate some kind of in, like when you’re thinking about inspiration, it’d be nice to have some kind of brochure that you can flick that like a magazine experience is really nice when you want inspiration.

[01:10:12] Yali: You can maybe talk to it. And as you’re talking to it, it can adapt in real time and you can keep flicking through it. And at some point you might say, well, I think it’s going to be this destination, this destination. And then maybe the UX pattern changes. And it’s less about a brochure that you’d scroll. And it’s more about diving into the destinations and what the similarities are, the differences on what you really want to get out of the trip.

[01:10:33] Yali: And then you might be comparing hotels. So that kind of much richer UX experience feels really powerful if you narrow your domain down to a particular customer journey. And OpenAI is always going to be building all journeys in, in chat GVT, which is just a really big task. So there’s, there’s a huge amount of space for most companies and most industries to really think, think, think about how to use this technology to build the next generation of domain specific experiences.

[01:11:12] Yali: If they’re good people will go, people might start in chat GBT, but they might then hop over ’cause they want that brochure experience. They want to keep having to talk to the chat agent to just see little boxes of destinations. They want to be immersed in the experience. Maybe they’re on their, their iPad or what or whatever.

[01:11:30] Yali: So I think there’s so much, this is such an exciting time to be thinking imaginatively about what those digital experiences can be and be investing in them. And I think that’s the kind of thing that we see a lot of startups doing. Whereas a lot of the more established organizations at the moment are still trying to get on site agents working for their customers effectively.

[01:11:51] Yali: So it’s, it’ll, it’ll come and I think it’ll be big, but it’s a, it’s one of those things that I think will take a while, a while to happen, but I think it will happen. So I think there will be a thriving web side of open ai. It just might be fundamentally different. Experience, like you say, you can have, you can see like even playing with Claude code, like I know you guys say you’ve been doing that internally.

[01:12:13] Yali: Some of the stuff that that can just produce, especially if you give it underlying frameworks to adhere to and things like that, it can just sort of create this page. You can imagine really just completely dynamic pages fitting, almost like you saying with signals, like fitting to the customer journey of like, they’ve done this, I need to generate a page that has these features on it because this is what they need next.

[01:12:33] Yali: It is, it just might be a different way of web developing. And fundamentally, I think that’s exactly right. I think, just so almost like just in time developments, the agents figuring out right at this, at this minute what the right experience is. And it’s more like we have to, we have to see how it goes.

[01:12:51] Yali: We have to see what our expectations are as users. It’ll take us a while to learn how to use this. ’cause it’s a bit, yeah, I think, I think we need to learn, like we had to, we’re still learning how to use track GPT effectively, and we’ll have to learn how to use these. Yeah, new generative experiences.

[01:13:06] Yali: You can’t follow up your ground bar and say, click the button at the top. It’s just like the rest of the button at the top. Yeah. This is it. It’s, yeah, it’s going to be interesting. We’re all going to have to change a lot. It’s going to be, it’s going to be tough for grandma. It’s going to be tough for all of us. Well, it is.

[01:13:20] Yali: And 

[01:13:20] Dara: there’s a lot of talk as well about how the, you know, the, at the AI tech, it’s limited by how we’re using hardware at the moment, how we’re using devices, because obviously devices were developed for a technology that was significantly older. So it’s like this, and then there’s rumors about this Johnny Ives, Sam Altman collab, and, you know, all these different types of pendants you can wear around your neck or whether it’s the, the meta RayBan glasses.

[01:13:46] Dara: But we, you know, we are still thinking in terms of this being something that we interact with by typing something into a box. And, and that might not be how we’re interacting with these chat features, these LLMs in, you know, six months time, 

[01:14:01] Yali: two years, whatever. I’m sure, I’m sure you are right. Type typing is going to seem very archaic.

[01:14:06] Yali: I can’t work out what to tell my kids to learn. I think it might be a complete waste of time. 

[01:14:12] Dara: It’s definitely going to be one of those things where, where they look back and say, I can’t believe you used to use your fingers to type messages. 

[01:14:20] Matthew: Our interest. W with all this new technology and, and I know you said earlier, you’ve mentioned startups a couple of times and that, that being like a bit of a signal for you of what’s coming over the horizon.

[01:14:29] Matthew: Have you seen an explosion in, in startups with, with this all the sort of new LLM technology or has it remained static pretty much throughout the lifetime of when you’ve been looking at them?

[01:14:40] Yali: No. There’s been an explosion of startups and the turnover in those startups is, yeah, was very, very high.

[01:14:46] Yali: There’ve been startups that started two years ago that folded or pivoted. Yeah. But the technology’s changing. It’s changing so fast. It’s a really exciting time to build a new, new business. It’s a, it’s a really tough time in the sense that it’s not clear you are building on shifting sand. Mm-hmm. You the expression.

[01:15:04] Yali: Yeah. The expression is, so I’ve got a huge amount of respect for the founders that are toughing, toughing it out now. It’s definitely harder for them than it was for us when we, when we started.

[01:15:14] Matthew: Yeah. I, I can think back to say 2023 or 2024, early 2024, where you see all these new startups appearing on, say, LinkedIn or EL elsewhere, that maybe were like connecting up your BigQuery data to here and analyzing your BigQuery data in it with an LLM or passing a PDF.

[01:15:32] Matthew: And they play so much within the potential tooling that will appear with these frontier models as they mature. Like we’re seeing now that, yeah, one, one off the cuff art by some Altman on stage can bury 60 companies overnight. I shouldn’t laugh, but it’s just, yeah. It’s crazy.

[01:15:54] Yali: It’s brutal, but it’s part of that open platform, closed platform. Shifts. The AI tools for development are like the, the probably the most mature, the engineering productivity is probably the most mature, use case for this, this technology and both open AI and floors now have really compelling products it market that compete with a lot of their best customers in terms of cursor wi Search and dev and others.

[01:16:24] Yali: So it’s, I think the cursor guys have done phenomenally to keep, building great new features and, they keep people using cursor alongside Codex and core code. But, it’s like what a market to compete in. What a, what a terrifying prospect to be relying on these. Partnering provides us while they’re competing with you. You’ve got to, you’ve got to be made to really tough stuff.

[01:16:49] Yali: On the optimistic bent of it, it does feel like anecdotally just looking, looking at GitHub and looking at what people are creating, the open source world is booming because of all of this new tooling. It’s kind of putting, facilitating people to be solving very niche and specific problems for themselves, and then releasing that out into the world, which feels really good and positive to see, just to turn it around to a positive.

[01:17:12] Yali: Yes. Yeah. Yeah, yeah. And I, I, I shouldn’t, I, I should say I’m hugely optimistic about all of this. I don’t want to, I don’t want to come across as a, as it’s down eye 11, it’s, they’re just, there’s a huge amount of opportunity. There’s a huge amount of, every time there’s a lot of opportunity, there are also a lot of dangers to navigate.

[01:17:35] Yali: And when things are moving past, it’s, it’s, it’s, it’s really exciting that, that anyone can, can now, Write applications to publish the shares. It’s interesting, the, I don’t know to what extent I was, the model with open source is you make the source codes available. I wonder if actually what we should be making available are the prompts that we use to generate the source so people can take them and play with the prompts and, but maybe it doesn’t matter.

[01:18:07] Yali: Maybe take the source codes and the, the agents are not so good at figuring out how they’re built and you can edit them that way.

[01:18:14] Yali: We have done something kind of, not, not exactly like that internally, but when we’ve been using Claude Code, we’ve created a GitHub template. So you, so it’s got a, it’s got a Claude MD file in there and it’s got various pieces of information about the stack to use.

[01:18:31] Yali: So an individual can just pull that down and then just start typing away. And it looks at the Claude MD file pulls a lot of contacts out of what it needs to do and how it needs to structure things. It really takes even further away all of that initial jump. And, and that’s been quite effective in getting really nice looking things out quite quickly.

[01:18:47] Yali: So suppose it’s akin to a prompt, just a very big complex one. Yeah, yeah. Next year, two years. What do you see on the horizon? Is there any big thing, big change that you, you are confident or have a prediction that you would plant your flag in that’s going to be fundamental to the industry or just to the, if you want to be so bold to the wider world?

[01:19:08] Yali: I think the, I, I, I’d go back to the prediction I made earlier. I think UX is going to look totally different to how it does today. And so that’s going to be really, really interesting for us as consumers are going to have to learn how to work with these new UXs. They’re going to kind of evolve with us as we figure out how to do that.

[01:19:27] Yali: I think, data people, we’ve got a really important role in helping to figure this out. What are the signals that matter when you’re trying to guess what somebody is trying to accomplish? I think everything we’re, we’re going to need to build systems that really effectively give people what they need. And so how good you are at figuring out what your customers need is going to be super, super important.

[01:19:55] Yali: And obviously it’s going to be AI figuring that out with humans. But the people that are good at that, the data folk that are good at that, I think have a very bright future and matter. In that paradigm. 

[01:20:07] Dara: Alright, YALI, thank you, again for joining us. That was a really good chat and I think we might have to get you on again at some point in the future if you’re willing.

[01:20:16] Yali: My pleasure. That was really, really enjoyable. Love talking to you both. So thank, thanks so much for having me. 

[01:20:20] Dara: Thank you. That’s it for this week’s episode of the Measure Pod. We hope you enjoyed it and picked up something useful along the way. If you haven’t already, make sure to subscribe on whatever platform you’re listening on so you don’t miss future episodes.

[01:20:34] Matthew: And if you’re enjoying the show, we’d really appreciate it if you left us a quick review. It really helps more people discover the pod and keeps us motivated to bring back more. So thanks for listening and we’ll catch you next time.

• Spotify
• YouTube
• RSS Feed