Devin Gaffney 0:00 I'll be, I'll be, I'm still on note, come on, come on, let's be real. 0:16 Ah, I see, I've allowed it. 0:26 Do you need to go to System Preferences? Speaker B 0:28 Oh no, yeah. 0:29 You can do this right here really quick. 0:33 Yeah, yeah, go for it. Devin Gaffney 0:51 Great. 0:55 Alright, can you guys hear me okay? 0:57 My voice is a little hoarse after being out late every single night with everybody and going on about @proto, so I apologize, I sound a little rough here. 1:05 I have some, I'm still on note a little bit, so I have some prepared remarks and then we can kind of do some questions if that's helpful. 1:12 So, let's see here. 1:16 A handful of oligarchs control transformative technologies that have reshaped the world in a generation. 1:22 They foist negative externalities upon us, vertically integrate us into servitude, and entrap us into network effects that we cannot escape. 1:30 I'm talking about the Gilded Age, but that also describes our moment right now. 1:35 Some Gilded Age problems even look the same. 1:38 These goblins representing the diminished information landscape of the 1880s echo our own current degraded moment of AI slop, rage bait, and corporate capture. 1:51 The Gilded Age ended though. 1:52 Back then the transformative network was rail. 1:55 Early companies started with unique track gauges, then relied on that as a competitive moat. 2:00 Entire, eventually entire trainloads would have to unload and reload at the edges of proprietary networks just to continue a journey. 2:09 People accepted this absurdity until they did not. 2:12 The standard gauge followed, rail became commoditized and interoperable and companies that were dependent on that network lock-in collapsed. 2:20 In our Gilded Age, the transformative network is the attention economy, and in the attention economy, algorithms are the engines that convert our time into their revenue. 2:30 Incumbents built empires on proprietary algorithms trapped in walled gardens, and like the train passengers, we're realizing the absurdity of algorithmic entrapment, that we don't have to accept it, and that the legacy platforms are now starting to pivot and panic. 2:47 In 16 months of being online, we've hit some good numbers. 2:50 We're personally thrilled by the traction, but we think it's much more about a real unmet demand about the basic functioning of the attention economy. 2:59 People want way more control over their experience than legacy platforms can provide. 3:06 The way that we enable user choice in algorithmic feeds is simple. 3:10 We provide tools for people to use that are simple no-code systems that can be combined to curate and shape any vibe for any sort of content that they want to see in a feed. 3:20 And when those people end up creating feeds that attract a large community, we think that we can make that sustainable with the same exact monetization techniques that power the rest of the existing attention economy, but place those tools directly in the hands of the stakeholders responsible for those communities. 3:37 One of the best examples of this is Andra's newsfeed. 3:39 Is Andra in here yet? 3:41 Where's Andra? 3:41 There's Andra! 3:47 If you essentialize the value of a microblogging platform to a single thing, it would be to get vetted news right now. 3:55 Andra's also grown her newsfeed into a news ecosystem with different tailored and niched versions of the core feed to serve distinct audiences and their needs. 4:04 No wonder that one of the most popular feeds is a direct carve-out of this affordance. 4:09 The fact that no other platform thought to create this as a specific vertical in the last 20 years also speaks to the dynamic nature of an open market of algorithms. 4:19 In an ecosystem where people can freely build and grow an audience around their own algorithms, individuals can design the informational flows they and their peers actually want, and readers can pick and choose those experiences freely. 4:35 I believe this creates the conditions for a more equitable, more responsive, and less exploitative attention economy. 4:43 This isn't just about the long tail of the Reddit moderator/Wikipedia editor class of the internet, though. 4:50 One of the best examples of how we can use this technology to start unwinding some of the harm social media has wrought upon journalism was during the New York City mayoral election. 5:00 We worked with WNYC to build a feed that contained vetted information sourced from newsrooms to cover the election results in real time with editorial hands on the controls, which stood in stark contrast to the breaking news experience on X, which had stale, cheap jokes and racist caricatures as the breaking news experience. 5:22 By recentering news media, we can use custom feeds as a vehicle to begin to re-enfranchise their voices in these systems. 5:30 And all of this stuff is going to be sort of aspirational and moot unless we can build the big business case for this technology as well. 5:37 That's why we're working with folks like the NBA to design custom feeds that exactly match the league's ideal user-generated content voice. 5:44 Because they actually have ownership over their algorithm and they know what their version of ideal fan engagement looks like, they can create a feed that showcases the best of the fan community while drowning out the poly-market the slop screenshots that have polluted other platforms. 6:01 And this is not just a story for Bluesky. 6:03 This is a story for every emerging platform in this ecosystem. 6:07 As they grow and challenge their respective incumbents or create new experiences altogether, one of the greatest challenges and most powerful weapons is the radical composability of custom feeds. 6:18 Just as Bluesky is sowing the green shoots of feeds that outcompete experiences on X, So too can the teams taking on every other legacy platform that is predicated on knowing what's best for users. 6:31 In the 16 months of running this company, we've learned a few lessons and we've picked up a few things along the way that may be of use to folks in the room building out their own companies on the protocol. 6:41 The first and most important is that fundraising is exceptionally difficult in the current ecosystem in general and profoundly difficult with social media. 6:51 There's significant headwinds in every direction. 6:53 Most important, and maybe hardest to hear for some in the room, is that @proto is not growing exponentially, which means functionally it is a non-starter for VC investment, full stop. 7:04 Also, the AI hype cycle has all but consumed available oxygen that would otherwise remain. 7:10 On top of that, there's a common misconception that social is solved. 7:13 We're sort of at this end of history moment where the idea that large platforms could ever be replaced is not an imaginable outcome. 7:21 In compounding/reifying all of this is the VC herd mentality. 7:26 No one wants to be last to an investment trend, but no one wants to be first either. 7:32 VC funding is harder in atproto as well because you have to tick all of the classic boxes against all those headwinds while also being in many ways the inversion of the classic VC bet. 7:44 One way I explain the Platonic ideal of a VC investment is, what if federal crime, but with a computer? 7:59 The idea being that with companies like Uber, Airbnb, Polymarket, OpenAI, Bird, Lime, DraftKings, Calci, etc., the wager is that inserting digital infrastructure into what would otherwise be the illegal privatization of a regulated commons, or illegal removal of a protection from privation, is in many ways the best investment a VC can make. 8:22 Atproto is a complete inversion of opening up a commons while also having to tick all the other boxes that makes a company traditionally investible, which is a much more sophisticated argument that is much rarer in its chance of success. 8:35 Everyone's journey through this is going to be so different that there's no real cheat sheet on the way through funding. 8:41 The basic north stars are sort of the same as always though. 8:43 You have to have an extremely clear theory of action. 8:45 For why your product will become a huge success. 8:49 You need a thesis for why you'll be able to get to that successful outcome despite the existence of incumbents. 8:55 And you need an initial wedge for how you can work from a starting point to success to test that theory of action and thesis. 9:02 And honestly, do not mention @proto, at least until it goes exponential again. 9:06 Treat it as a vehicle rather than the story itself. 9:11 Finally, I want to share my own personal belief of where I think the ecosystem is headed, at least insofar as our work at Graze is concerned. 9:20 At a high level, we're at the intersection of two tectonic tendencies in technology. 9:25 First, technology is sedimentary. 9:27 This isn't a new idea. 9:28 You can find this in media theory or information science or sociology, or very literally embedded in the OSI model itself. 9:35 Technological systems layer upward in increasingly abstracted goals. 9:39 The lower layer concerns itself, uh, with the flow of electrical signal. 9:44 A higher layer concerns itself with the liking of a mutual's skeet. 9:48 The highest layer, uh, the second network technologies have a way of being forced open, just as brake gauge crises of early rail networks gave way to the open standards because the friction of incompatibility outweighed the value of the network. 10:04 Bundled proprietary companies like AOL, who also stood at their own end of history, eventually get outmoded by open standards like the open web. 10:13 We believe @proto is the protocol that forces this distinction for social. 10:18 Trains don't choose your stop. 10:19 Power companies don't choose your computer. 10:22 Apple doesn't choose your browser. 10:23 Chrome doesn't choose your bookmarks. 10:25 At every layer of this stack, the carrier eventually loses control of the layer above it. 10:31 And yet legacy social has treated the user account as inextricably linked to what you see when you log in with it. 10:38 That's never obtained for long at any other level, and we don't think it will obtain here now that the distinction is being made visible. 10:45 The best that we can do is accelerate this process, but this is a contingent moment. 10:50 If we don't build products that exploit this distinction, we end up back in the same structures, and we don't stand to be able to beat platforms with 1,000x the staff, 1,000x the investment, and a 20-year head start by competing on their legacy bundled experience terms. 11:07 The only way to outcompete them is to change the rules of competition in exactly the way that history has proven works against entrenched proprietary networks. 11:18 Network technologies are not inherently inevitable. 11:21 They start as these idiosyncratic cobblings solving a problem only slightly better than before. 11:26 They etch organically atop some sort of agar until they become so ingrained in the world that regularity and order must be imposed upon it. 11:36 We've been at this end of history online before. 11:39 The open web of the early '90s was a weird pain in the ass to use. 11:43 It was hard to see how it would outcompete the bundled, slick experience of systems like America Online, until it did. 11:52 We're in that fragile, weird point right now with the protocolization of the social layer. 11:58 We get to collectively decide if we're at the end of history at this layer. 12:02 We get to decide if the everything apps get displaced by an anything protocol. 12:08 There are still headwinds, but the experimentation in this community still makes me optimistic about the odds. 12:13 Thank you. Speaker B 12:33 Questions? Devin Gaffney 12:40 Come on. Speaker C 12:48 I just want to congratulate you on what you built. 12:51 I don't think this community would be what it is without Graze. Speaker B 12:55 Thank you. 12:57 Thank you. Speaker C 13:02 I would love to hear, you know, you shared some of the kind of biggest like banner successes of like the NBA feed, the New York election feed. 13:11 I would love to hear like some of the smaller ones or quirkier ones, you know, some of the feeds or communities that were really exciting to you. Devin Gaffney 13:24 Sure, okay, so I've been like in this milieu for 20 years in my work as half my career was in academia, half of it's been in practice. 13:34 I started with doing research about the Iran election protests in 2009 and then the Arab Spring in 2011. 13:41 So that's sort of my, you know, SICO origin story. 13:47 And one of the things that like breathes life for me in this stuff is the work that Rich Farrow and Zombie Gomone, I don't know his real name, they are both I know his real name, I forget his real name right now. 13:59 He's known as Zombie Goemon. 14:01 You should follow both of these people. 14:02 They're amazing people. 14:03 They are part of this editor class of the internet. 14:06 They built a feed collectively last summer while ICE was occupying LA because they saw that they needed to stand up some sort of ad hoc infrastructure to understand what was going on in their community in real time. 14:19 That is exactly what we should be doing. 14:20 It is a gain of function above just using fucking hashtags that it has to be an opt-in experience. 14:27 And that is the sort of community resilience type work that is what I talk about every time I get an opportunity to talk about. 14:33 And then there's also just a lot of weird feeds that one particular person, Roland Crosby, builds. 14:38 I don't know if he's watching from New York, but he's a total weirdo and he builds feeds with poorly cropped images, that sort of stuff. Speaker D 14:49 That's hurtful for my sponsor. 14:51 Posts that I made. 14:55 Other questions? Speaker E 15:03 So you had some comments on the investment landscape, and there's this AI elephant in the room, or in taking over the universe. 15:09 And then yesterday we heard all of a sudden an AI addition on the blue sky side of things. 15:15 What's— do you have a reading as to how AI and the social world from the perspective investment are going to, or ought to, or not to relate going forward to make this ecosystem here more investable than it is? Devin Gaffney 15:33 Sure, the story that I tell about that sort of thing is platforms are being— there is a cat-and-mouse game for hacking the attention economy. 15:42 When you have a centralized algorithm, the target is bigger, the value of hacking it is, you know, it's more valuable, right? 15:48 So a centralized platform is always going to be at a disadvantage in terms of being able to prevent this sort of stuff. 15:55 Federating elections makes elections safe or unhackable in some sense because there's a huge surface area. 16:02 Federating algorithms has a similar effect to be able to prevent this sort of stuff. 16:06 Also, as AI-generated slop will infect any platform, The ability for individual curators to be reactive to that in their local community, I think, is going to be a much more efficient system than just having some sort of de-slop button. 16:25 In general, the sort of shift is going to be from the value of creators to the value of curators that are able to filter this stuff out actively for their communities. 16:35 All of that is going to be, you know, that is the effect of AI, and I think it will make this ecosystem grow. Speaker F 16:46 To that same kind of point, do you think it's— do you think investment will eventually just catch up, or the ecosystem has to, like, change to fit the mold that investment sees? 17:07 As warranted? Devin Gaffney 17:10 Yes, and I think that there will be some teams that are able to come up with a totally novel theory that is, you know, not hitting the walls that a lot of teams that have, you know, came up the beginning of last year within the ecosystem. 17:24 There's going to be some sort of new theory, I'm sure, that's going to emerge. 17:27 Otherwise, you know, I think the biggest boon is going to be some VCs taking a contrarian bet. 17:35 The first word in the acronym VC is venture, right? 17:40 There is some sort of risk involved with these things, and someone will eventually have to take a chance on this space if they want to see a return on that investment, right? 17:50 So I think it's going to be both. 17:52 Somebody— VCs need to get off the sidelines and understand what the theory of the case here and what the opportunity is. 17:58 and then other people are going to have to shift strategies. Speaker G 18:04 I wanted to ask you, talk about AI-generated slop, and I wanted to ask how you contrast that to the way that Graves uses LLMs to help folks generate feeds, and how you sort of make that distinction when you think like an LLM is being used effectively or not effectively. Devin Gaffney 18:26 Sure. 18:27 LLMs should be fundamentally in service of a curator. 18:31 They should not replace a curator. 18:32 These should not be systems that are simple drop-in, just make this thing work for me sort of experiences. 18:39 That is not a durable way to build software with LLMs because, again, this is a dynamic adversarial environment where systems will adapt accordingly. 18:48 Our theory is that in what we built a year ago, was an AI generation tool where you can explain, I want this feed into the system, and then it will design a sort of initial spec of like, this is where you should take it. 19:03 And one of the other benefits of it ultimately being a JSON spec, right, is that it is an interpretable document, right? 19:12 You can share this document and say, this is how— this is the logic of this space, right? 19:17 That opens itself up to other adversarial things, but is also a function of transparency. Speaker D 19:23 View source. Devin Gaffney 19:24 Yeah. Speaker D 19:25 Another question. 19:29 Devin, again, thank you very much for being a returning sponsor, round 2. 19:34 Thank you very much. 19:36 You've done so much.