AI, Infra vs Apps & Spotting Red Herrings with Akash Bajwa, Principal @ Earlybird

0:00Hello and welcome to another episode of Riding Unicorns. Today we're joined by Akash Badjwa, Principal at Early Bird. Akash, thank you so much for joining us. It would be great if you could tell us a little bit about your journey into venture and how you ended up at Early Bird. Thanks, James. And thanks, Hector. Super delighted to be here. Yeah, I'd say my journey into venture was somewhat accidental and a coincidence. So I actually started out at Barclays in their private bank. That was my first job after uni. and I was allowed to rotate into their corporate venture arm for a little while. And a lot of my colleagues at the time were former VCs or founders and operators.

0:46I got to know the fintech ecosystem in London quite well. And once I'd spent some time there, I kind of caught the bug, I guess you could say, and wanted to work in tech. I started to consume a lot more content, podcasts, blogs, et cetera. I did a series of internships after that. And then my first real full-time role in venture was at Augmentum Fintech, which was a series A, series B fintech fund based in London, which was really rewarding in several ways. Firstly, the firm had only been listed for several years the time I joined. So it really had the hunger of an emerging manager, which also meant much more rope and freedom to add value to the firm.

1:33But on the other hand, I also found myself gravitating more and more towards seed or the very inception stage of investing. I did a couple of angel investments, very tiny nominal amounts, but still gave me a taste of what it's like to partner with founders really on day one. And I was building conviction that that's where I saw myself in the long term. And then I got to know the Early Bird team just as peers in the industry. And they were looking to open an office in London. And so it was just really good timing that that relationship deepened at that very moment. And then so I joined Early Bird in 2022, beginning of 2022, as we set up our London office.

2:16And now we've grown the team here considerably. We're five investors in London and continuing to add to the team. And now it's going to be the biggest office for us in terms of investment team headcount. And so, yeah, fast forward, it's been three and a half years and looking forward to, yeah, lots more to come here at Early Bird. And Akash, tell us a bit more about Early Bird. I mean, you guys have built an amazing platform, you know, many investors on the team now, lots of well-known team members. And tell us a bit about the focus areas and what makes EarlyBird, EarlyBird. So the best place to start is we take a lot of pride in being the longest standing early stage fund in Europe.

2:59I think I can say that without any caveats. We were founded in 1997. Of course, we have some peers who were also set up around a similar period, whereas some of those folks have gone into growth investing, other strategies. We've just stuck to what we know best, which is early stage investing, partnering with founders as the first money in usually or first institutional investor. And so now we have 28 years of just doing that. And we really believe that has allowed us to hone the art of early stage investing. Obviously, the asset class has institutionalized. European venture is completely different to the first 20, 10, or 15 years that early birds existed.

3:40But we've seen different cycles. We've seen different technology waves. We can analogize better when we look at AI with mobile, with cloud, with AI. And uniquely, at least in Europe, that's a strength of ours. And so that's still the mission we carry out today. We invest out of our eighth early stage fund. We're typically a lead investor at Pre-Seat Seed and Series A. Continue investing out of the early stage fund. And then we have an opportunities fund that then can continue being a partner to companies all the way to the exit. And as a lead investor, we take a lot of pride in being very hands-on.

4:14So we tend to think of the ways we strategically help companies as really boiling down to two key dimensions, product and go-to-market. And on product, we think all great software companies go from single product companies to multi-product companies by the time they IPO. Equally, your first channel has to obviously scale on go-to-market, but you also have to go multi-channel over time. And we think there are some tried and tested playbooks and patterns that we've seen the best companies execute on that we try to bring to bear for our companies. And then lastly, since we've been investing in Europe so long, we also have a deep and wide network that we activate for early hires and early design partners, customers, as you expand not only within Europe, but also to the East Coast usually.

5:00And then, of course, eventually to the West Coast. So that is how I tend to characterize the way we partner with founders. And you talked about the art of early stage. What does that mean to you guys? How do you hone an art? Is there a science behind the art? Is there an early bird process? How do you think about the art of early stage investing? Yeah, I think one thing we acknowledge as a team is that we have a diverse range of opinions on this. I think you'll both probably know well, my colleague Andre sort of spearheads in many ways a lot of the data driven initiatives are happening in venture in Europe.

5:42and we're very proud of the strength that gives us in sourcing and in screening and all your support whilst at the same time we also have other team members and a general view that there's a lot to this which is relationship-based which is qualitative which can't be really boiled down into just numbers and data and for example I'd say founder assessment is something that empirical data could tell you, you know, unicorn founders exhibit these kinds of traits, but that's obviously backward looking. And if you look at even recently, the last two, three years, the cohort of breakout AI companies, arguably a lot of them break some of the patterns that you might have had based on data that looks into the past.

6:28So we tend to see that as a strength, that there's a range of opinions in the team on how to best marry the qualitative and more artisanal way of investing and sort of forming mentor models and frameworks, different kinds of questions that reveal certain traits that we think are generally indicative of great outlier potential with the rigor that a data-driven way of assessing companies, assessing markets, allows us to make better, higher quality decisions. And I think we see that as a strength also in our investment committee. All of our five GPs have different verticals they spend time in. But when they look at a deal in a certain sector, they bring that expertise from all those different verticals, whether they're looking typically at capital-intensive hardware businesses, or they're looking at e-commerce businesses that have a completely different margin profile, different types of defensibility, or fintech, and then applied AI.

7:26I think it's the combination of all of that that makes us a better firm and capable of making better decisions when we're looking at companies. You mentioned the sort of techniques to finding outlier founders, and it's something we all have to do and trying to do better all the time. But, you know, the nature of outliers is that they are outliers. So looking for patterns to identify outliers is for all with challenges. Although I think, you know, I think there are some sort of that you almost have to look on like meta level, you know, like what are the underlying? and traits as you mentioned this new wave of ai companies and founders do generally look quite different to the sort of previous generation but what what are the things have you guys looked at any of the found any traits characteristics that are common between these two generations of founders i think uh the conventional wisdom before ai in the sass cloud sass era was that actually more experience is a positive.

8:24So some of the data would have suggested that if you look at companies like ServiceNow, Salesforce, others, like those founders were in their 30s, 40s, often a lot of the incumbent systems of record who sit on like some of the most lucrative IT budgets and ERP, CRM, ECM, and so on. Those were founded by founders who were actually not just two years out of university weren't maybe either calling strong files either or coming from certain kind of privileged backgrounds. And obviously today we see a very different profile of young scrappy builder who is capable of just having that neuroplasticity where they can learn what it is to be AI native faster than someone who is experienced can learn how to adapt to becoming AI native.

9:17That's not only playing out at the level of the individual, that's obviously also playing out at the company level where you could say maybe in some cases an intercom has done a better job at becoming AI native in their vertical than a lot of incumbents have in their other respective verticals. And so it's not to say that those who are experienced founders that we all were gravitating towards in the cloud SaaS era are not capable of becoming or being humble enough to relearn a lot of things or unlearn certain things and then learn a lot of new things when it comes to being AI native. But it's definitely like a, it causes a bit of cognitive dissonance, I think, that you have to change your business model, for example.

9:59You have to change your go-to-market motion. You have to be forward deployed. And a lot of the other things that have only recently become really popular. And the founders that are very young today, obviously, they only know how to build AI native companies. So they have no priors that they have to unlearn. I think that's one of the biggest contrasts that we're seeing. And then if you combine that with the speed of how this is all unfolding, like the data is obvious, but if ChatGPT becomes a 1 billion monthly active user company within three years, that's just astonishing when you compare that to all the other big consumer apps of the last 20, 30 years.

10:40And I think that is reflective of the way that AI is diffusing across society and the economy generally. And so we could have had in previous platform shifts more time to adapt, but now there's less time. In this one, there's way less time than mobile, cloud, and so forth. So I think that places even more of an onus on the experienced people to adapt quicker, because the window of opportunity isn't going to be there forever. There's a really interesting thing where the first generation of venture in maybe the 90s was often find great technology and then bring in a management team who experienced CEO who built an organization and knew how to build processes and run companies as they grew through scale.

11:28and I think Akash what I what I wanted to understand a bit more is whether you guys are willing to underwrite these kinds of founders who are super hacky super young don't have experience scaling teams and businesses but perhaps you know you guys are able to underwrite those kinds of founders because you have the right playbooks you can equip them with you know we've now got in Europe countless cases of founders who've done that and we know how to scale businesses in a way that we didn't know 10 years ago. And so are you happy backing those founders or do the founders you back need to have that innate ability to scale teams, which some people are gifted with more than others?

12:09I'd say the premium on a second-time founder or repeat founder is still going to be very high as it would have been pre-AI as well because arguably a lot of things do become de-risked in terms of hiring, especially if it wasn't successful. it, building a culture of excellence and execution, fundraising, even for that matter. But I wouldn't hold it against first-time founders that we expect other traits now than we did before. And obviously, that's nuanced based on where they sit in the stack. So when we're looking at vertical AI companies, the importance placed on vertical domain knowledge is still as high as it would have been if you were looking just at vertical SaaS companies in the 2010s.

12:54arguably the element that matters more now than building the product is that go-to-market insight. Like what is your unique insight on selling to tradesmen that unlocks that market? And it may well be that there isn't a particularly unique insight, but again, there's only so many founders that can speak like the language of that buyer that know what resonates with them, that know how to build a field sales team and so on. All those things, especially when it comes to vertical plays where you're bringing an offline industry online wouldn't look that different in a post-AI world. If you're building infrastructure, I think that also, in my view, hasn't fundamentally changed in terms of the challenges of doing bottoms-up PLG, building developer communities, doing developer relations.

13:42If anything, maybe the importance on good developer relations and good developer docs is starting to matter more as we think about agents and how traffic is going to get redistributed from humans to agents. So there's some meta trends unfolding there. But by and large, I think that's still quite similar to before. And so and then if we don't touch on foundation models, at least, then looking at infra and apps, I think a lot of the same principles apply. And if anything else, I think it's just that you have to place another higher premium again on speed. And it's not just speed where it's directionless.

14:22You want velocity. Like it has to be speed with an aim. It has to be aimed at demand. And the other kind of heuristic I use for that is the teams that have the fastest learning rate. So what I mean by that is, yes, you have a certain assumption today, and then you'll speak to customers, you'll speak to prospects, and you'll validate if, for example, the product you're building today is maybe your wedge product, but then actually where you'll make four of your money is on the second and third products you'll sell them as you build out like a suite and you're willing to revise your assumptions constantly but what's more important is that you're uncovering those insights faster than your competition because for the reasons we know you'll have way more competition than SaaS companies did in the early days of cloud.

15:08That I think is something that isn't super easy to measure a precedence seed. You're obviously looking at like prior experiences and what they achieved in those companies. You might look at the period they spent validating their idea, how resourceful they were in speaking to potential design partners and how quickly they iterated on different assumptions. But I think that is probably going to be, for me, it's one of the traits I place the most emphasis on above and beyond things on their CV or their prior experience. Yeah, really interesting. And you've invested in both infrastructure and application layer AI companies.

15:45Can you tell us a little bit more about your sort of thought process behind those so the infrastructure it feels like there's a sort of a more sort of a stickier you know less competition but you know really hard to pick which companies are going to be core infrastructure players and then on the application there it feels like you're going to be in a mix of yeah lots of competition as you just mentioned but you can with the speed of adoption of some AI businesses now and maybe the markets are so big that it doesn't matter about competition, you can still build amazing companies. So can you just talk us through a little bit your view on how infrastructure and application are playing out when you're looking at AI?

16:26Yeah, I think the very tricky thing is to make really generalizable comments about both layers. So if I start with infrastructure, one category that was very hot two, three years ago was vector databases. And at the time, it was obvious because of where we were in the cycle that all of the pure play vector databases were going to see incredible growth. And it would have been easy at the time to extrapolate that growth into the future with really bullish assumptions. And the detractors at the time were saying, look, this is very easy for MongoDB to add for existing multimodal databases to support vector search.

17:06But purists would have argued that, oh, actually, if you want higher quality and higher performance on vector search, you need a pure vector database. And I'd say looking at the recent news about Pinecone and then also how MongoDB and other incumbents have had a lot of traction of your vector search operate, it seems like that was a case of too much hype around a specific infra thing. And the way I would maybe generalize that case study is to say, I think there's a lot of primitives in infra, whether that's observability, databases, and security, and so forth, where there's a class of companies emerging that are claiming they're like AI native, like they're built from the ground up to serve an agent economy, or to serve AI workloads, which look very different in nature to SaaS workloads.

17:58I generally see that argument, but I just think it applies to differing levels based on what that primitive is. Like, for example, in AI security, we've seen there's been a few recent transactions. Palo Alto Networks acquired a company recently called Protect AI, and you would have seen Latera was acquired by Checkpoint. And I think that speaks to the fact that maybe each primitive won't support a large standalone company. And like in vector databases, actually, a lot of these primitives can be bundled into an incumbent offering. But there will be some cases where there should be a completely new type of infra player.

18:41So that's a tricky thing around generalizing. On the app layer, I think the reason for a new swath of companies is more compelling. Just because I think application layer, ultimately, you are not only changing the back end, you're also really changing the front end. So the way human-computer interaction looks is being also completely reinvented. And if you then trace a nice narrative of how we've gone from the days of Oracle databases towards sort of cloud host success to today, I think there's just one clear and variable trend, which is augmenting human workers and their productivity and finding better ways to do that.

19:25And arguably, there needs to be a redesign of products, how they enable humans in the age of AI, how maybe you could, you know, spin up five cloud code instances or have cursor doing a bunch of asynchronous jobs or have granola, you know, potentially synthesize insights from Aureo calls. and surface those to you at the right time. That is what I think is unique at the application layer is that the way you even interact with the software, it looks different. It's not only that in the backend, there's a transformer model that's powering them, but we have a chance now to also rebuild the way we use these tools.

20:03That's what makes the innovator's dilemma really acute because if you've already been building with one form factor and one UI for a long time, you have a customer base that you're reluctant to migrate onto a completely new way of doing things and obviously this is a very consensus argument but i really think it's true it's simple but true yeah it's really interesting just before you ask the next question can i just ask that because i think that that's such an interesting way to think about application layer changes and disruption because i do you remember when like tinder came out and it was all about that like swipe interface and then everyone tried to build like almost like a tinder for property a tinder for fashion whatever and it was just like this new interface created by other people trying to create it it didn't really translate as well as the dating one but now we have prompt interfaces we have voice we have all these other angles and i think a lot of the innovation is a lot of the innovations dilemma is that thing of like, well, actually, we have the data and the customer base, but this new way of doing things would actually be a lot better.

21:12But it's hard to make that call and make that migration. And so does that then leave the opportunity open to the new player coming through? Yeah, that's what I would say. I just recently read some really interesting data which just was tracking how Chatty PT sessions were increasing over time. So if you trace it back to 2023 and look at every cohort for every quarter since then, you can see how session lengths have been increasing and there would be inflection points around key new features being added, such as advanced voice mode, such as memory, and so on, which is to your point, James, that we're reaching a point where the models and their capabilities are maturing so much so that really entirely new experiences can be built around them and people use them for a wider range of things so if previously people only did research with these models we can see entirely new verticals being unlocked like financial well-being health and wellness travel etc because these new capabilities that we're adding make them more human-like or make them more empathetic and and so forth uh the of course the modality of the input matters i think if we increasingly move away from typing and just to dictating.

22:31I think I'm a heavy Whisperflow user. I think it's incredible to imagine what kind of automation potential you can have if you can just say, hey, using Whisperflow, I dictate to my OpenAI operator that I want certain things to be done and it's doing those jobs for me overnight. If you just imagine where this goes in a year or two with continued rates of advancement, I think there's entirely new verticals that get unlocked by this and and yeah to your point someone has to rethink these experiences well I'd love to know just in your portfolio in your experience using different tools like whisper flow or whatever else what's the experience that you're just like wow oh my god like that product is amazing this experience is incredible you know we're all seeing so much across the board I'd love to hear anything that that sort of really blown you away?

23:23Yeah, I'd say the metric that was always seen as a North Star for a lot of PLG companies that I think is still very appropriate for AI companies, it's just time to value, but time to value in one thing and doing that exceptionally well. So in the case of Whisper Flow, I think what people appreciated was just, it's just blazing fast and accurate. And I think it's a bit like if you're an early stage company and you're doing design partnerships, you want to have a certain scope and not anything bigger than that, because otherwise no one will work with you. But in that scope, you really want to nail it.

24:05Just nail one thing really well. And I think that speaks to a lot of prosumer AI companies. I think Granola, when people first experienced it, the time to value was super short because on your first call, you just saw, First of all, I think it sort of inculcated this new behavior where people read the transcript as they're listening. And that didn't exist before. And also this really nice aesthetic of the notepad was also something that in your first call, you immediately realized value. And it was just that one other difference beyond the other incumbent note takers that they nailed really well.

24:42I think in the past, there was a lot of debate, such as in the case of Superhuman, on this topic of should you stay in stealth for a while, for a few years, and just release something that is like super feature complete, super comprehensive. And like the way that they kind of describe these two ends of the spectrum was like Hollywood production, blockbuster style, you know, release, or you really go with the lean startup methodology. I really think in the age of AI, I'm more in the camp of the lead startup methodology, like iterate really fast, but I think really nail one thing really quickly, and then you earn the right to quickly expand from there as long as you have that speed.

25:21But I think if you're building for a long time today with this kind of internal conviction that isn't driven by feedback from the market of what kind of reception your product will have among users, you don't have a community or a wait list that you can get a pulse from of what actually drives value for them. you might actually be far less aware of where to build than the company that maybe is better at getting some distribution through one viral product and then following the demand signal from there and then eventually building to keep their customers happy. So I think you see that a lot lately.

25:59Clueli obviously released a pretty viral demo, a controversial one too, but at the end of the day, what they had was... Who was that, Akash? Who did you say? Clulee. Clulee is the company in the US that's somewhat similar to Granola, actually, but it's more of an ambient assistant that sits on your desktop. It was the company where he was on a date and he was sort of cheating during the date because it was reminding him of the questions to ask that Andreessen funded. But the lesson a lot of people took away from Clulee was that if nothing else, he built a massive community of waitlists that if he now has to build and have some direction of where to build, at least he's doing that with that type feedback loop from the community, as opposed to someone who doesn't have that distribution and reach and is kind of building in a silo.

26:58And I think the argument again for that former case is that the models are going to improve. So the things you're promising that might not be possible today they will be possible probably in three months or six months. And so you'd rather maybe over-promise a little bit just to get that attention. And again, nail time to value and one initial thing, do it really, really well. And then at least you have that constant low latency feedback from the community of where you should build next. Then if you were building, again, in stealth, without any feedback, and just waiting for the models to improve, but without showing anything to the world.

27:33at least that's that's a pattern that we're seeing really interesting i think we've discussed previously on the podcast is that trying to invest you know ahead of models and and what they can actually do so if stuff is kind of doing it now it will be able to do it in like six 12 months but then that raises a question which i'd love to get your thoughts on which are what are the biggest red herrings in ai investing at the moment the things that seeing like they would be you know obviously successful but actually you're sort of thinking well no it might not play out that way yeah i i think um the the insight that a lot of pain points felt today are like ephemeral or short-term in nature is very true in ai today i think that was definitely the case with a lot of developer tooling an example of this would be at least, I hope I don't have said a lot of founders I know and have spoken to, but in the MCP world, MCP, for I guess the audience who won't know, is this protocol developed by Anthropic.

28:42The main purpose is to allow clients, which are usually tools like Cloud or Cursor, where people do their work, to connect the tools. It's just a kind of communication protocol. And people People analogized with the early days of APIs and concluded that there would be a need for a lot of middleware, basically, to discover MCPs, aggregate them, host them, and so on. I think that is just that I'm calling out that market to say that I think that's an example of where, yes, there might be a business to be built for 6, 12, 18 months. But I do see Anthropic, and I think this is true of other cases too, but whoever the main actor is behind that pain point, for example, obviously OpenAI initially didn't support guardrails.

29:34Then they started adding guardrails too. And so a lot of people build for these tiny white spaces that are at that moment in time underserved by the foundation model labs, but that quickly become definitely on their roadmap. And when you have well-funded labs that have every reason to solve those problems for their customers, it seems, again, going back to the infrastructure pain point especially, it seems challenging to be building for pain points that are by nature like very short term. And if that doesn't give you a wedge into an enterprise who would then give you the trust to continue solving other problems for them, then after six, 12 months, you could be looking at a business that's kind of disappeared overnight, which is this proverbial red wedding that happens with every demo day or dev day that OpenAI has.

30:29or it's on big tasks. So at least these last three years, that's been happening every year. Obviously, I hope I'm wrong in a lot of these cases with these companies and the founders who I really admire who are building these businesses. But I think that's the tricky thing. And obviously we're all wondering that, which is where are the labs going to focus and what will be neglected, both in terms of verticals that they'll neglect or that are like non-core or too dilutive to their core business. And similarly on the infra side, you could obviously make a case for routing because routing is always going to be model agnostic.