Proud pod parents (Friends)

37:59So one of the years, and you, of course. I mean, you were able to poke and prod him about Rust and get him to sort of geek out. So, yeah. What ended up happening was one of the years we were running Unboxed, I won't go into too many details, but it's kind of chaotic getting set up for a large event like that. And so a lot of things changed last minute. And so originally we were supposed to be like filming something else the day or two before unboxed with some of our partners, some of the DevRel team. And then tons of things started falling through. And so we had to completely change what we were filming.

38:40And so it turned into me just doing interviews with several of our like community developer friends. And like, I think we called these, I'm going to butcher this, like dev chat sessions or something like that. if somebody goes to square dev on YouTube, you can find the dev chat sessions. That to me was the first step towards like doing a podcast. Because when I sat down and did the interview, I actually just had like very loose questions. And basically, like, it was in a podcast like format. But we just had like a full film crew there filming it like it's, you know, some sort of partner interview.

39:19And I think like when we saw how well that was received, even within square and from the partners who participated um that to me was like the okay i'm 100 convinced like we can actually just do a podcast because the podcast is actually like i don't want to like say it's easier but it's easier in terms of like hey i'm in my home office studio type thing they can be anywhere we're not having to fly everybody into a single like location um and so yeah i that that is like actually what totally convinced me um that we should be doing the podcast and like this is what the content should be um so that was like kind of like my my revelation gotcha so now you're on the other side you have a season in the can i think episodes are coming out weekly it's four years later or whatever and you've successfully created and branded and shipped a podcast so being where you are now can you speak to anybody who might be back where you were a few years ago thinking maybe we should have a podcast like what were the hard parts what was expected unexpectedly easy is it is it all worth it for like another platform company to go through the building process and the creation process what are the hard parts what are the easy parts etc i would say like the hard part i mean adam adam and you became familiar with some of the hard parts of you know i would say that like the my number one recommendation is like just start small like and i and when i say small i mean like with a small team like don't try to get too many cooks in the kitchen yeah just too many cooks in the kitchen everybody wants to go in a different direction and i think that like really just start doing it even if so like even when we first started doing the podcast here i think one thing that helped a lot and i wanted i'm nearly positive this is that like adam's recommendation is that we actually just interviewed like brought in somebody internally to like run like a kind of like test episode or two and i think like it's just if you go and you do that like you can just build at least the tiny bit of confidence that you need that like this can work um and you can hammer out all like those little things that you might be worried about ahead of time.

41:35The other thing that I would just say that is, it's actually a lot easier than anybody might be making it out to be. Like, you just got to start, like, just sit down, have a conversation and record it. It might not be great the first time, but like, when you just sit down, you have the conversation, you record it, you can have a sense of like, what went well and what didn't go well. And it's weird that I'm saying this now, because we ran into the same exact thing when we launched our YouTube channel is that, you know, when we first recorded episodes of the YouTube, we like, you know, overproduced way too much.

42:10And we realized like, okay, we can like throw out like half of the work here. Like we were, we were way overthinking it. And I think even before doing this podcast, we were way overthinking it. So like, like the thing that I just can't help but like continually think is like, should have just like listened to Adam and trusted him more about it that, you know, just start it. you know the format will start to like make sense and i think like the only part that i think i would still i still got hung up on was like what what like the content is going to look like um and i think in retrospect like you just need to start inviting like a few guests and start recording because the content will start to make more sense as you start making it because you might think oh this is like what i want the content to be about you start having the conversations and the interviews with people and then you realize like it keeps going in this other direction and this other direction is actually like not what I thought originally but it's working and so like you I always think of this analogy um a lot of the time I've said it to my team I don't know if it works well um but I always say like the map is not the territory so like the map is an abstraction of where it is so like there's a very big difference between reading on a map of where you're going and actually walking the path.

43:28And so like in actually walking the path, you can see like, Hey, there's a tree fall in here or Hey, there's like, you know, uh, not a lot of cover through your maps. Don't tell you these things. They can kind of just tell you an idea of like, Oh, I can get from here to here, but it's very different actually going and walking that path. And I think like when you actually go and make the podcast, that is very different than like when somebody tells you like how to make a podcast. A lot of analogs to software too there. It's like you can't really know until you're in there what actually it's doing, what's not going to work.

44:00You have to get real just to go back and borrow some 37 Signals, Getting Real, which was a good book title. The territory is the reality, right? The map is the abstraction. And so it's a helpful tool. But until you're actually in the real world, out there where that tree is, you know, the map didn't have the tree on it. Well, I missed something. You don't actually know. And so you have to just get going. It's not full fidelity. it's just right that's right and then you actually create the thing and then you're like oh well that wasn't very good but i got some ideas for my next try and you just iterate and shine and you know don't go back and listen to our early episodes please all the people do that and i'm like what are you a glutton for punishment you know yeah i think too with with you richard that we had a so our audience is aware of this our ad styles are uniquely different than i think other ad styles are where we really go inside of a company and have deep, authentic conversations and pull clips from that and produce them around the story.

44:58So that's how we do it. And you and I've had many conversations, you know, across, you know, we don't just sit down for 20 minutes and hammer out the content. We'll obviously have various conversations throughout. And sometimes we'll have an hour long session and 20 minutes recording, but I had learned enough about you and learned enough about how you operate to have confidence and what I think your potential was. And so my idea wasn't just, oh, Square should have a podcast. It was like, no, Square needs you to produce the podcast because you have all these unique insights into DevRel and what Square is and this evolution from Square to Block.

45:38And, you know, I don't know what your title is on the inside around APIs. I know that your title is mainly around DevRel. But I think you've also been in charge of the API or APIs generally for many, many years now. So you kind of don't title yourself. So you had this depth and background and ability to talk to people and share stories that I just was like, man, you are an underutilized asset. They can do some cool stuff for Square. Let's make it happen. And so I had complete confidence in you from back in the day. Well, it's just so very warm to hear that. It's so warm. I feel like I spend so much time just like steeped into these like, you know, ridiculous conversations.

46:22I mean, they're not ridiculous. They're important, but they feel ridiculous. The conversations around like API design, API. I just so the yeah, that dual title is like I lead our it's called API working group. I kind of like externally just say it's like an API design team. They kind of like set the standards, guide teams on like how to build APIs in a cohesive, like consistent manner. And yeah, I think that actually spurred from like we, our DevRel team just kind of got frustrated at times. I guess the nicest way of putting that, where we would have an API make it all the way out to being shipped into, put into the developers hands.

47:03And then like the developers, like this doesn't work or like this isn't like well designed. And, you know, we were just like, this doesn't make sense. Like, why are we putting this in their hands? And so then the design group at that time would consult with us pretty regularly. And it eventually got to a point where like, OK, one of us needs to actually be a part of this group and helping you make these decisions rather than just, you know, being consulted every once in a while. And so we ended up, I mean, eventually I ended up leading it as things changed over. um and so we've kind of like formalized this into like a company-wide group um in terms of like api design because cache actually has like their own apis now and they decided to adopt the standard and so yeah it's something that i don't talk about up until recently more publicly and i'm hoping to actually talk a lot more about like api design um especially after like uh i participated in like We had this sort of hack week internally where we started building.

48:04Everybody just went crazy on building MCP servers. And the light bulb kind of clicked for me when we built an MCP server for some of Square's stuff. And I was like, oh, this can be so powerful. But you kind of have to have that foundational layer in order for it to be capitalized on. Well, let's unpack that. So model, context, protocol. It's all the rage. everyone's shipping servers i saw github's official mcp server is hitting the top of you know github trending i saw shopify mp mcp amazon mcp servers you're on linkedin talking about mcp so yeah what's all the hype and excitement about is it seems like a new style api i haven't actually looked into it i know it's for llms to get context but why do we need something new why can't we just use the apis that are already out there you know that is like the the great question that i feel like um there's many people asking the same question like why can't you just you know hand like a bunch like say here's an open api spec hand it to your llm and say like go like build this thing for me like and and look at it and um part of it is because the the llm clients need some way to like know how to call tools and so like this is kind of trying to standardize what the tool calling is um so like the way that i sort of describe mcp is like it's like the sdk for robots and so like you have these well-formed ways of saying like hey this is how you call the payments endpoint or this is how you call the catalog endpoint and rather than because you could i've tried this before you can sort of just like hand an open API spec to the LLM, usually it's like too large or unwieldy for it to like really kind of parse out the right things for the particular context.

50:02And so MCB kind of like formalizes the way that you want the LLM to interact with your platform. Okay. And so like, and, and it also is trying to like formalize like things beyond like, say just REST APIs. So I've seen some folks who actually just use it for like tapping into like snapping like visual elements out of a computer and sort of using that to interact i saw a demo of somebody actually like having it work with an android app using some sort of indirect visual like reasoning where it's basically taking a snapshot of the ui and then figuring out where to like oh press this button and then like gives an instruction to like press a different button so like it can go well beyond just like you know traditional like rest apis yeah it seems like on the face of it except for that beyond stuff it's kind of delivering on what the promise of rest was at least with the hadios stuff where it's like you should be able to give a single endpoint to a programmer and the programmer can through uh tooling discover the entire entirety of their abilities and it never really worked out it always ended up being like a spec or go read the docs.

51:14I'll make the docs really good, which was going to step up from bad docs. But still, it was never, like you're still just basically reading URLs and finding nouns and like, okay, now I got to add this and nest that and then I get what I want and then I can post, et cetera. And that was never what we wanted it to be, but it was better than, I guess, an RPC endpoint with a list of methods you could call. And this seems like it's maybe taking that concept and saying, well, when the LLMs need their contacts, we're going to give them everything they need to actually do that without a human reading the docs or without an LLM even having to read the docs or scrape the site.

51:51Yeah, I think what you touched on there is I've been thinking about Hadeos a lot recently that for people who built their APIs in that way, I feel like they're definitely advantaged when trying to integrate or adopt AI because it's self-documenting. It can figure out how things are linked together. I feel like they just have an easier time with integrating this type of stuff. One thing I would actually, we had this argument, well, not really an argument, slight disagreement in how to view APIs going forward was somebody within our company said, oh, we should design our APIs to work better for robots.

52:35And I was just like, well, I, I, I, here's where I would disagree on that. And mainly that all of the AI LLMs have been trained on like APIs, SDKs, documentation, that's all designed for people. So like it is trained to try to understand it in the human like way. And so I feel like trying to design your API to work in a more robotic way makes it less human, less usable, less useful overall. And I think that like it just means that you should build better, more well-defined and like clearer APIs going forward. Because if we eventually hit this dream of AGI, like you don't need to have a robot specific API.

53:21You can have like a human API. So either a human or a robot can use it. interesting so what does an mcp server look like i said describe it to us sure um so at the base level um an mcp server will first like sort of define all the available tools that you want to give to it um and so like you know in the case of like say square like an example like you might say like oh there's a catalog customer a payment tool um and then sort of describes like how these can be used in different contexts. And so when you ask your LLM, hey, can you go build me a menu for a taqueria in my Square account? It's going to go to the MCP server and say, list all the tools and then figure out, okay, of these tools, which one of these matches with something that they're asking me.

54:13And then it'll go call the tool and then the tool will give an additional description of what else can you do with this? And then it figures out from there or how to actually accomplish the task that you want. So the thing that I always think about MCP servers is it's actually ridiculously simple in what they actually enable. Because you're just sort of creating this tiny little abstraction layer of saying, oh, here's all the things we're giving you access to. It's all just JSON at the end of the day? Or in terms of actual interchange formats, it's like typical stuff, there's no special sauce?

54:52Uh, essentially, I mean, like, so like, I think there's two, there probably is three. So like, this is the part where I'm like, I've only sort of worked with like server sent events is one format. So that's when you can have like a remote MCP server. Um, and then you also have like command line ones. So like you might have like a command line tool that it's, you know, I've seen a lot of them are just like calling NPX and then whatever, like this, basically it's like a CLI app that's like sort of packaged together for it to call in and get all those different tools and so those are kind of like the two main ways but um i think the most common interchange format is probably json that people are you know sending stuff back to the llm with and how do you plug these into your clients then does it depend on the client maybe uh yeah it does depend on the client so like when we were building an mcp server for square during this hack week Like Goose is our main open source AI client that we've created here at Block.

55:50And it has basically like sort of this standard config. And I think it was originally YAML. It might be in JSON now. That basically just sort of says like, this is the name of your extension. This is, you know, the interchange format. Is it going to be, you know, server sent events or is it going to be command line? And then you just kind of like say, here's the command. I've seen versions of this where like people actually are just like running a Docker container locally on your system. And like, that's actually what it's communicating with. And so I think Claude has like a, its own standard config file.

56:27But for the goose goose desktop client, we have like a way for you to just like do a deep link URL. So you just like click that button and it immediately starts pre-configuring that into your, your desktop client and installs the extension for you. Well, friends, I am here with a new friend of mine, Scott Dietzen, CEO of Augment Code. I'm excited about this. Augment taps into your team's collective knowledge, your code base, your documentation, your dependencies. It is the most context aware developer AI. So you won't just code faster. You also build smarter. It's an ask me anything for your code.

57:01It's your deep thinking buddy. It's your stand flow antidote. OK, Scott, so for the foreseeable future, AI assisted is here to stay. It's just a matter of getting the AI to be a better assistant. And in particular, I want help on the thinking part, not necessarily the coding part. Can you speak to the thinking problem versus the coding problem and the potential false dichotomy there? A couple of different points to make. You know, AIs have gotten good at making incremental changes, at least when they understand customer software. So first, and the biggest limitation that these AIs have today, they really don't understand anything about your code base.

57:38If you take GitHub Copilot, for example, it's like a fresh college graduate, understands some programming languages and algorithms, but doesn't understand what you're trying to do. And as a result of that, something like two thirds of the community on average drops off of the product, especially the expert developers. Augment is different. We use retrieval augmented generation to deeply mine the knowledge that's inherent inside your code base. So we are a co-pilot that is an expert and that can help you navigate the code base, help you find issues and fix them and resolve them over time much more quickly than you can trying to tutor up a novice on your software.

58:16So you're often compared to GitHub Copilot. I got to imagine that you have a hot take. What's your hot take on GitHub Copilot? I think it was a great 1.0 product. And I think they've done a huge service in promoting AI. But I think the game has changed. We have moved from AIs that are new college graduates to, in effect, AIs that are now among the best developers in your code base. And that difference is a profound one for software engineering in particular. You know, if you're writing a new application from scratch, you want a web page that'll play tic-tac-toe, piece of cake to crank that out.

58:51But if you're looking at, you know, a tens of millions of line code base, like many of our customers, Lemonade is one of them. I mean, 10 million line monorepo as they move engineers inside and around that code base and hire new engineers. Just the workload on senior developers to mentor people into areas of the code base they're not familiar with is hugely painful. An AI that knows the answer and is available 7 by 24. You don't have to interrupt anybody and can help coach you through whatever you're trying to work on is hugely empowering to an engineer working in unfamiliar code. Very cool. Well, friends, Augment Code is developer AI that uses deep understanding of your large code base and how you build software to deliver personalized code suggestions and insights.

59:37A good next step is to go to augmentcode.com. That's A-U-G-M-E-N-T-C-O-D-E.com. Request a free trial, contact sales, or if you're an open source project, Augment is free to you to use. Learn more at augmentcode.com. That's A-U-G-M-E-N-T-C-O-D-E.com. Augmentcode.com.

1:00:06I'm thinking back to the early days of Web 2.0 and REST APIs, or let's just call them Web APIs. And the spirit of innovation and discovery and creativity with open, not open API to spec, but like open web APIs. uh, specifically around like Twitter and Flickr and Foursquare and that group of social cool tech companies. it was really the heyday for mashups and trying new things and it was really fun and i feel like some of that excitement right now is i'm getting flashbacks to that with mcp servers because there's a lot of cool demos right now you can plug it into darn near anything you got mcp server uh you know i saw a guy who just like plugged it into postgres and of course his llm can speak sql but it's just better it seems like you can just it can just do better with not having to write SQL just to hit the MCP server and you can just talk to your Postgres database.

1:01:11And then I'm curious about what happened the last time was the good old days got old and then the companies grew up and they needed to make money and they shut things off and they locked out their party developer and it just got very hostile eventually. Sometimes out of necessity, sometimes you don't know, but it happened. And I'm curious your thoughts on that parallel because it seems like mcp and like letting your llm talk to x where x is not the social network but x is all of the things that have been mcp servers is like really cool right now and could produce some really interesting use cases and like let some awesome hacks out and then i'm wondering if is there a plan for when that spirit of openness changes what are your thoughts no i think it's like uh so yeah what you touched on i think angie jones my my new manager who leads uh blocks devrel just gave a talk on this kind of like mashups like it's very reminiscent of the mashups of the heyday of 2.0 and uh i think what was the example they had like it was like google maps and craigslist and like that actually was what spurred google to create open api platforms for the maps product and you know so like there's a lot of like very interesting innovations that like ended up coming from this but yeah i think when the way that i view it now is like we're kind of in like the peak of the hype cycle like you're seeing this like massive explosion and people just like producing all these various things and the recoil i'm sure that we're going to be seeing is like all of the anti-patterns there were probably anti-patterns before but we just decided to like do them again in this new way um so like i think one one thing that you talk about is like it's probably useful for somebody to be able to talk directly with their postgres database i'm sure that there's some people that find a lot of value in that but it's also very quickly gonna you're gonna realize like well should i enable other people to talk to my postgres database and that's when i say like no there's a reason that we had these abstraction layers created because we don't think that, you know, everybody should just be talking directly to each other's Postgres databases.

1:03:23That's why we created those interfaces. And so I think that, like, that's what we're going to see start happening. I think we saw, like, some early signs of this, like where, you know, people identifying security vulnerabilities with MCP servers, you know, it's probably something that's not well-defined within the protocol. But, I mean, like, that's, you know, I think that's been true of a lot of standards as they've been developed. I think even like in like the first version of OAuth, there were security vulnerabilities within the format. For sure. And, you know, it unlocked a lot, but, you know, there's always going to be like ways that people are going to find like why it's not going to quite do or should do everything that we want it to.

1:04:07Right. And the stakes are higher. Of course, with OAuth, the stakes are high. But now when it's like plug an LLM into your life, every facet of your life, whether it's your bank account or your Foursquare account or whatever your square account. There's a lot, you know, when there is a vulnerability, there's just so much opportunity for bad actors that we definitely will see some trouble. This is a brand new thing. I mean, I think Anthropic came out with it back in November. And of course they had the the task of getting other people besides anthropic to actually adopt the protocol, which it was open.

1:04:46And I think designed to be not an anthropic thing from the start, which is a great way to get it going. But here we are in April and it seems like, will it be the protocol to rule them all in the next age of agentic things? Not sure, but right now it seems like it's got a lot of momentum behind it and quite possibly might be. um and so it's gonna be vulnerable until it gets robust i mean that's what happens you have to face the real world for a while yeah find all the obvious things and work out the kinks and that's part of the reason why like i still you know when i was writing about it on on linkedin that like i think that the the platforms that are going to be able to leverage this the most but also like last and survive some of like the the downsides a little bit better are the ones that already invested in really good api platforms like if you already have that built in like with oauth authentication permissioning like if you have that basis and then you're wrapping that your npcp server around that you have a lot of the safety guards already put into place my concern is for the people who just like haven't done that like base level of work and they're just like, hey, we're just going to stand up an MCP server.

1:06:03It talks to Postgres. It talks to this other thing with inside our internal things. That's, I think, going to be a recipe for disaster in the long term. You might see some initial short wins, but I think over time, it becomes just a huge exploit. I think that we debated this inside with a couple of other engineers in my company. Like, hey, it's really cool that we built an MCP server that can talk to Squares APIs, but why stop there? Why not have it talk to some of our internal services? And I had to point out to folks, I'm like, well, if we create an LLM agent that can talk to Squares internal services, that effectively makes them public.

1:06:45Because now I can go and like, I'm sure somebody could end up doing some sort of prompt engineering to basically dissect what are all the different tools it's calling internally and be like, great. I'm now going to go use those to write my own program on top of those. I don't need the LLM to do it for me. How about this as an idea? What if AGI is essentially a self-building MPC server that the LLM talks to to discover what it needs? It doesn't have what it needs, so it makes new stuff and it calls the same MPC server to discover more of itself. So it's like it needs to do something, can't do it, builds it, can do it, builds it, needs to do something, can't do it.

1:07:30It's like this constant talk between the LM and the MPC server. Well, I think like it isn't like that new. I haven't I'm going to be perfectly honest here. So I'm like kind of just shooting a shot in the dark of like with Google's new agent to agent protocol that they're trying to propose. Like one of the theories could be that you can have your agent go to this hub of agents and be like, hey, they're asking me to do this thing. I don't really know how to do it, but I'm going to ask all of these other agents in here. Can any of you solve this problem? And then like one of them goes, oh, yeah, I can.

1:08:00Maybe that's the way that you have this like this agent to agent hub. It's like Upwork for agents, you know, like, hey, I'm looking for a designer. Yeah. Looking for this function. Can't do this function. There we go. It's like Fiverr for agents. Exactly. Like throw the bids out there. I'll do it for five tokens. Right. Yeah. Race to the bottom. I'm trying to hack this square MCP server. Can somebody help me? I got an agent that does that. What's the race, you know, given MCP and what's happening? What is the what is the next arms race when it comes to that layer of the implementation? Like who's next?

1:08:39What's the next big thing? Past MCP or with it? Well, I mean, if it's if it's burgeoning, what is the you know, what is it going to enable? What's the next big thing? I mean, you kind of touched on it with Web 2.0 and access to APIs that had freedom and innovation. Well, that's why I was concerned that the providers may change their heart about the thing. Obviously, Richard, you are one, so that's why I was throwing it to you. Because the spirit of innovation and openness and Square obviously would be more useful if I could plug my agent into it and get information ad hoc and have it do things for me without having to code it up myself, etc.

1:09:20However, it leads to the agent gets super, the client gets superpowers. And so whoever controls the client, and right now OpenAI very much controls the client because ChatGPT has hundreds of millions of users, and it's starting to become the Google of this era in terms of just the default choice. And then Square becomes this thing that plugs into ChatGPT, and now open AI holds your interface and you're just commoditized? Like that's a concern, isn't it? I mean, it is a concern, but I think like, you know, we're, I mean, I want to believe always that open source will find a way to win. Yeah. And I think like the open sourcing of models, the trying to incentivize folks to releasing like the weights and like their measurements and sort of like insisting on the transparency so that we can like have confidence.

1:10:15Cause I think like that's a precarious relationship to be managing. Like if you don't open up a little bit and build that transparency, because I think that you can end up like triggering like a huge defensive move from everyone where they go, Oh, like at first signal, like that, you know, we realize like they're exploiting our relationship or our data, you know, everyone that's going to be like, all right, we're cutting off, we're pivoting over and we're going to like either do it in house or we're going to go pick a different provider. We're going to do it, you know, open source. Um, and so I think it's a tough thing to manage over time.

1:10:49Cause I think, I mean, one thing that is benefiting us all right now is at least there's a market of LLMs for you to be able to select from, um, some work better in certain contexts, you know, chat GPT is like really great for like sort of general purpose, you know, anthropic is really building stuff. It's like super awesome for developers. Um, and we were just hoping that like these, these open models can actually do more and more um so i mean i've definitely been dabbling and trying to run an llm locally on an old gaming laptop it's fun but it's not practical right i'd have to have a way way more powerful machine and i don't really want to spend that much money on it right yeah i mean adam and i have been doing some local stuff as well and the what i call it adam the gravitational pull the tractor beam the tractor beam that i think chat gpt has is their product design.

1:11:42And it's the old open source versus proprietary, you know, capitalized company who's laser focused on just like, what they're doing so well is they're rolling all the latest advancements into a product. And what all these other things are is they're models that you can use with a chat app UI built by one dev to scratch their own itch. Obviously, you can also have capitalized companies that use the open stuff to create a competing product and that's happening as well and i just think that i agree with you 100 that it's great that we have a diversity of options and i think when it comes to models they're being commoditized i do not think open ai will hold a lead over the open source world that's meaningful for very long um all the llama ford just dropped and it's disappointing so but deep seek is really good you know like there's there's things happening right now metas seems to be not impressing with their latest efforts but anyways that i think open source wins but that's implementation interface is what matters the end of the day like how people are going ahead and using these things which is why apple's so seriously dropping the ball right now because they have the interface to everything and it's terrible it's called siri and they can't ship intelligence in it um it is like think about the opportunity they had with siri they were so far ahead.

1:13:06They have the device. They literally have the interface. They have a five-year lead. I mean, Alexa was the next one that got some penetration, but Siri was there and they got millions of people talking to Siri and then Siri was just sucking over and over again for years. And it could have been so much better. And then they'd have, I'm glad that that happened because they'd have lock on everything. Everyone would be talking to Siri. At least now it's a separate company. With ChatGPT being the new Google. I mean, I have people in my life who just, and these are not techies, who are like, did you ChatGPT it?

1:13:46It's the new Google. And so I think OpenAI right now, even though in their technology and their models and stuff, it's not really going to hold a moat, but in their product and their momentum. And so all these people creating MCP servers might just be making chat GPT, you know, the muscle, the superpower client. And so that's why I feel like I was likening it back to that previous decade where everyone's like, yeah, mash up our API with everybody else's. And then people started doing it and they're like, wait a second, you're not coming to our website anymore because you're just using our data to create better products or different things.

1:14:24And so now we're going to have to charge you for that. obviously you don't represent Square and the entire company or block you represent a small portion of it but I think when it comes to MCP servers there has to be some trepidation in allowing the full not the full but a a lot of functionality that you provide to be used by people who maybe don't even know what Square is you know they're just using it through their chat GPT yeah i mean like i know that like you know for a lot of what we've been doing you know we have like safeguards in place there's a lot of folks out there who like you know they use aws bedrock to like do hosted models be more confident that their data isn't being like shared elsewhere um and i'm trying to like have some of these protections but i think yeah over time like the one thing i think about like what gives open ai i'm trying to i'm not super confident in this that like that they actually a lot of their moat right now is definitely like their llm model but the interesting thing to me has been recently like with these smaller i mean essentially like smaller updates that is just enough to cut out features that you find in other like essentially like you know chat gpt wrappers exactly when they like say in their desktop client like hey we're going to actually like apply code directly into your your vs code you're like yeah that's clearly like a cutout for cursor and like you know it's like oh the more you just bridge the little gap right like they can just choose to like solve little parts of the problem it's like well we're preventing them from like eating too much of our lunch right but we're going to continue to like i mean it's in a place where they don't even necessarily need to be front running everything they can actually let open source and all these other things kind of front run things and then pick and choose the things that go oh people really like this piece of that uh puzzle like let's actually just go like bake that into right uh chat gpt yeah they can sherlock the industry as apple would always do with their third-party developers yeah i think they're very well positioned to do that and i think that they've shown competency in that i say that from my personal experience because i've been trying to leave chat gpt and the product keeps pulling me back in most recently the image generation stuff which is a different you know it's a different thing altogether like it's multi-model now and so there's dolly involved and i don't even have to care about that it's like it pulls me back in and while i'm here i'm starting to talk to it again and i'm like oh it's better than it was last time i was here and so it makes it harder and harder for me to use what i had been using which is llama and i've been trying deep seek and i'm using you know google's gemini for coding and like trying out all these different things but at the end of It's like chat GPT just keeps getting better and better and better.

1:17:15I feel like they're running like that, like that Netflix or HBO kind of playbook. Yeah. It's like as soon as, as soon as the season's about to be over and you're like, all right, I don't need to be subbed to this anymore. Oh wait, we rolled out the hot new thing. Yeah. They're dragging us back away. Totally. So I don't know. I think obviously we don't know what's going to happen with all these MCP servers. And I think it's like similar to,

1:17:41Google search back in the day where it's like you have to be there. You can't not be there because then you're irrelevant. And obviously OpenAI is not the only people that will be using these MCP servers. So it'll serve all the needs of everybody. And so I think it's smart, at least for now, to be like, what's Square's approach right now? You guys have the hackathon, you tried some stuff. Is there going to be an official thing shipped? Do you know? Oh, I mean, well, now you're asking the things that I can't quite. So the thing I can say, because I mean, I did work on this. I know that from the DevRel side, we're definitely going to be pushing for getting this out there.

1:18:20Somebody from our Goose team actually did like a first version of a Square MCP server. In fact, when we were going to do this hack week, we looked at that and we were like, yeah, this is great. But like, you know, it's only covering like a very small portion. and then somebody on our team had like a really innovative idea on like how to do this and like basically cover the entire service area in a single shot and so we definitely want to get that out there and get people using it the thing that I still keep in mind and like you know this isn't like reflective of what everyone else thinks within block I think it still requires a certain level of care and I don't want to be like sort of like a the doomsayer of like what ends up happening but i can't help but like think like when you enable mcp servers for all these different api platforms so like i think of this is probably i'm gonna ballpark a metric from a long time ago but like say like the global number of developers was like 13 15 million people and this is like number i'm pulling from years ago so it's probably wrong when you enable this mcp server to connect connect into like cloud desktop chat gpt desktop plugging into anything else the number of how you viewed developers before has just exploded totally like good point 100x like it's now and that's just like thinking about like how a regular person can now go enable this stuff but that's also not accounting for like the the very savvy developer who goes well why spin up one agent when i could It's been up 20 agents to go do work for me.

1:19:58And so like, it's just going to be this like exploding problem. And so like, the thing I keep thinking about is like, can all of us handle the volume of a bunch of bots going around calling all of these things? Like, are we just going to see like a huge influx of DDoS? And like, I mean, it's hard to understand like what the second order effects of this are when you have just tons and tons of agents being spun up everywhere yeah well said good thoughts exciting times interesting times not sure exactly how it's all going to shake out but that's what makes it exciting and interesting right the question i think about is where's the toll booth that's where you make the money at right as a you got a need and you want to create your own version of a toll booth and to get access you have to pay x i suppose if you're thinking about that from a commercial standpoint or an enterprising standpoint it's like what is the next toll booth when it comes to this scenario you know like you've got agents that are going to go and spawn let's just say like you had just described and that that work was normally an individual opening up a book or going to the library or talking to so and so from you know the county about how regulations work like whatever however these npc servers mcp mpc gosh anyways anyways these mcp servers act because you got agents are gonna go out there and scour the whatever to figure out information and it may not be it may just be a dead thought or it may be a new thought that spawns a new thing that's now this next layer of whatever and it's so hard to even describe it so if you're lost i'm also lost the point is is like you've got all this traffic happening on the internet all this api traffic all this database traffic that is kind of probably not from a paid user but it's like data that's important to the world really hard to see where the toll booth is going to be at for this one though i can't uh i can't personally see where it's going to be yet i'm thinking about though that's an agent i mean it's interesting i think you make a good point because like uh the one thing i was like thinking about i used to be like very allergic to this still kind of am like uh when you i mean i'm not opposed to like monetizing apis but i feel like when you monetize apis you have to kind of go about it in the right way because otherwise developers will never use it.

1:22:30But now with MCP servers, I kind of go, well, it's not like developer, it costs almost nothing for the developer to now go use your APIs. So because they can just have an agent go do it for them and call them for them and they're not, you know, the barrier to adoption is like much, much lower. And so it does sort of beg the question of like, okay, if we're going to have like say 10X more traffic, do we need to actually put monetization in here in some way? Or, you know, go the chat GPT approach, I guess, is like, you know, we pitch our remote MCP server and it's, you know, five bucks a month. And like, you know, you get near unlimited calls or something to connect to any of your other AI tools.

1:23:17Right. Yeah, I think that seems somewhat feasible and likely, but can't see the future. I got some agents working on it though. Yeah. Nice line. I heard somebody say recently, and tell me if this is how you all feel. They said, if you take chat GPT away from us right now, we are dead in the wall. I'm paraphrasing, but it's a version of like just doom. We don't know how to work anymore. We have learned how to work because of the speed of trial and error and the speed of access to information and the speed of the user experience of asking and getting in this sort of volley back and forth. I heard someone say recently that they said, if you took this away from us, we wouldn't know how to work anymore.

1:24:07Is this how you all feel in any way, shape or form? No. No. I think it would suck. I mean, it would definitely suck because I think that there's many ways that it has accelerated us. but I don't think I would argue that we're not quite at a point where like we would be like oh my gosh like we can't we can't function there might be some companies that might not be able to function because they let too many people go thinking that AI was going to automate everything but I think there's still quite a few of us who'd be like yeah I still know how to write code by hand so you're only doing the code context then that's the that's the challenge I'm not I'm doing the whole life context I'm answering whole life my answer still no because I just fall back to where I was a couple of years ago, which was just Googling stuff and looking at YouTube videos, you know, like for me so far, and I've used it extensively aside from the joy aspect of the image generation, which for me is pure joy and I love it.

1:25:03And I don't want to go back to a life before I could turn someone into a walrus if I want to, you know, aside from that, I can just go back to Google and stuff. Like it's mostly just speeding me up. Like it's answering my questions, but I'll find the same answers. maybe it'll take me half an hour versus 30 seconds so i don't want to go back but i don't feel like i'd be dead in the water or i'd be a fool or whatever they does i couldn't do my job dims yeah yeah i don't know it's not that let me ask you a different question okay same question but different because this is actually a good analogy i think potentially okay so you're going from your house let's just say into town wherever into town is for you sure sure going from your house into town the old way was these back roads right stop sign and so-and-so's house scenic route right and you arrived at in town sure then you learned about this new thing called the highway right and fast and and you went there so much faster yeah this is how i liken it and i don't i don't disagree with what you're saying i just wonder how truthful it is or how truthful you're Let yourself be because once you've gone from your house to in town, the highway way, sans traffic.

1:26:15Okay, just bear with me here. Okay, don't put all the worst stuff in there. The speed version of it, the direct access, the get off the wherever and you're not in town. When you go from here to there, the old way, the scenic route, the 30-minute route versus the 10-minute route via the highway. And you have to go back to that old way. It does hurt. i don't care if you can still do it yeah i said that in my answer i said i would suck i would i don't want to go back right i'd like the fast but more i would say like here's here's a different kind of analogy where i think it would i want to sort of like tweak tweak what you had there like in the transition from you know a horse and buggy to a car there was a transition period where like people were like just not getting horses anymore and like they're like everyone's like getting cars but imagine that like some sort of key failure like all of a sudden we have no more fuel for cars so we have all these like hunks of junk everywhere and like now we're in this state of there's not enough horses for people to be able to get from point a to point b and now everyone is like extremely disconnected and so you can't get from where you are like going from san francisco to la now takes like you gotta hike because you have no way of like really getting there and you can only you know take what you can carry i think like there will be a point in time where we might get to that point where like we rely on ai for so many different components that if you suddenly shut it down that like getting from point a to point b might actually be like essentially impossible because it would take too long it becomes impossible not because it's true impossibility but because it's like gosh no no i'm never going that way again you know like i went the 30 minute route to in town and that was no fun okay i'm just never going back there again it's like five years from now i want to go check my square dashboard and they're like we don't have dashboards anymore we're just the mcp server you know exactly yeah you are literally describing what what i think can happen because like that's what people are proposing is like ai has like created a like so you know like everything is just like voice chat so you're just going to be like using super whisper to say like hey right you know go ahead and update my business hours for the upcoming holiday and then like suddenly you turn off and it's like wait how do i actually go update my business hours there's nothing in the dashboard here for me to go do this right you know we could end up in that that format where like as soon as you know somebody doesn't have access to ai it's like i don't actually have an interface and we have to have we have to keep the ejection you know have to be able to eject yeah um think about automatic cars you know there's a way of doing things manually i don't know maybe there's not anymore uh who's i just talking to where there's a oh like uh if a car's battery is out for instance and it has an electric start there has to be a way to start the car without the battery or at least to shift the gears sorry to shift the gears like to get it into neutral so we had like a we had somebody stuck in reverse you know and their car died and it's like real this is for real yeah yeah and it's like well we got to be able to, and there's like, and then a car guy came along and you see this little patch right here.

1:29:28You take this panel off and you stick a screwdriver in and you turn it and you can shift the gears manually. There's got to be a way to do it without the battery, even though a thousand times out of a thousand, you're just going to use the battery to start the car. And so I think we have to do that with these things. We can't go all agentic, you know, even though they're going to be the way in 10 years or whatever it is, because the agents will fail, you know, like something's going to go wrong and there has to be an injection process that square has to still have a dashboard richard you can't just get rid of it let me ask you a question jared yeah or richard too you can you can both answer this question okay let me think of a person do either of you have aunts or uncles both of you right yes just say yes just say yes of course you've got somebody you got to answer for me then yeah we do genealogy right okay tell me your uncle's number your phone the uncle's phone number could 5050505 you do not know do you really know your uncle's number that's actually his phone number okay so a different just docks memorized because it's some sort of special number he isn't the best phone number in history can you imagine that he his phone number was and he's passed away now so it was 5050505 i mean he didn't even pick that it just came to him isn't that amazing that's a cool number anyways aside from that yeah i wouldn't know it the point is is like you forget to remember phone numbers because your phone has them in the context like you have it in the context so there's no point of you've you've now just obliterated the need to ever remember now my wife's phone number i know uh my brother's phone number i don't know by heart who else's phone number do i know by heart not that many i mean a small few like sub my parents my wife right that's about it sub five and the reason why is because the phone or the thing or the cloud holds the data and it's so accessible and like you said a thousand out of a thousand times you're reaching for the phone to get the number because it has the information but how would you get your brother's number if you didn't have the phone yeah exactly yeah how are you calling anybody without my brother i'm like I'm like, okay, where'd that phone go?

1:31:41We're driving to Pennsylvania, right? Like, where's the internet so I can get access to this contact card? Because I have not personally remembered the phone number in my own brain. I've relied on a different brain, my second brain or whatever brain or whatever access to data. I mean, we're going to be like that at some point, right? Like, what you think is going to be the case is that there's just certain things you just don't have to remember anymore. Because for the most part, it's going to be there. Like there's my phone. Cool. Call brother. No. I stopped that from happening. It could totally do it.

1:32:17I'm sorry. I can't do that. Here's what I found on the web for call brother. Would you like me to call Chad GPT? Yeah. Sure. When it comes to memory, I think you're right. When it comes to functionality, there are certain things that you're going to have to be able to do. You know, like you can file your taxes manually with paper still. Right. Even though no one's doing that. So I think there's for mission critical things in life. Of course, if you lost your phone and you need to get ahold of your brother, you'd probably find somebody between him and you who knows his phone number. And you'd say, Hey, can you get me his phone number?

1:32:47And then you'd have it. So there's like, there are ways, but yes, we are going to probably forget a bunch of stuff and just ask LLMs. And then when they, if they don't have the answer for us, we're going to be like the people in idiocracy, you know? It doesn't exist. Yeah. It doesn't exist anymore. It's no longer true. Like that's, yeah, like if the agent can't find it, like I'm thinking of this thing now, like, gosh, can you, do you think we're heading to a point where you have an agent? I have an agent and everyone has a small army. Why just one? Proper word to use. Yeah. Of agents that we just now command.

1:33:21A host of agents. Have your agent do it. Yeah. Your agent. Talk to my agent. We'll do lunch. Right. And it's not a human agent. It's just some, it's what you refer to as your agent. It might actually have some humans behind it, but like generally it's not a human. it's some sort of software. I mean, I look forward to having a personal assistant. What I don't look forward to is my personal assistant learning exactly how I do my job and then replacing me. Right. Yeah, like, oh, I'm just your agent for now. I'm shadowing you, Richard. I'm learning all your, oh, Richard's not needed anymore. That reminds me of my latest Galaxy Brain meme that I don't think I actually shared with anybody.

1:33:57You guys know the Galaxy Brain one where it's like a progression of like smarter, smarter, smarter, smarter. so the first one which is like the small brain was i program the computer and then the bigger brain is the computer helps me program it okay the next bigger brain is i help the computer program itself and then the galaxy brain is the computer programs me

1:34:24that's where we're heading that's where we're heading man and that's the truth too let me tell how that happens okay algorithms when you go to the for you page that's called programming oh that's true it already is programming that's the way the computer's programming it's it's uh larger agents that are human potentially potentially putting out things you know deciding what content makes sense to you and you're in the bubble you've put yourself in and it's usually the world man this is getting doom and gloom and the world seems to be opt-in but it's mostly by force only you know i mean like i mean it's for the most part and only the privilege get to opt out you know it you have to be it's true you gotta be in a position to be able to opt out all right well let's close on this then if if the algorithms are programming us i think it might be time to pull the plug this might be it by the way we're producing podcasts that serve this i'm just kidding we're part of the problem around here oh man do we just get like deplatformed off youtube from this maybe maybe at least demonetized i'm gonna throw out a couple urls you can at least go to youtube.com slash at square dev i don't know if that's case sensitive let me check it because we've had some instances where case sensitive was was needed for yeah just search for square dev on youtube you'll get there even searching just youtube.com slash square dev works as well so all the pods are there videos that richard mentioned are there on youtube even that infamous uh conversation i had with jack dorsey is a scroll or two back sitting at 18 000 views let's get that out come on now yeah that's not it's an interesting conversations it's still relevant today it still is relevant today you know there was, there was more to that story, but we'll let it go for now.

1:36:17Samuel Jackson, call Siri and get us some directions to our favorite restaurant, please. It ain't going to happen. It's been fun. It's been Richard. Thanks for giving us the chance to help produce the square of the podcast. I think it's a super cool show. I'm glad that we finally got to a point where we can do it. I'm glad to have the season, you know, rolling out as we speak, a lot of good stuff on there. Very proud of you and the team behind the scenes. Make it happen. Could not be happier. Could not be happier. Thank you so much. Matt, thank you. I mean, really appreciate all the help that you've given us through this.

1:36:52And I got to say, it's also it's fun being a guest. I mean, I want to try this out a little bit more. It's almost more fun than being a host. Almost. Almost. Almost. Almost. Well, that's it. All right, friends. Bye, friends. All right. Bye. that is changelog for this week thanks for frenzing with us do check out the square developer pod links are in the show notes there's no way i'm gonna try to say that url another time and let us know what you think about mcp agi chat gpt versus the world of open source options or anything else on your mind in our totally free totally rad changelog zulip community Join right now, if you haven't yet, at changelog.com slash community.

1:37:39One more mention of our awesome partners at Fly.io, to our sponsors of this episode, Heroku, Retool, and Augment Code, and to our beat-freaking residents, the one, the only, the Breakmaster, Cylinder. Next week on the Change Log. News on Monday, Anthony Eden, founder of DN Simple, on Wednesday, and Mr. AhoyHoy himself, it's Nick Neesey on Friends on Friday. Have a great weekend. Like, subscribe, five-star review us if you dig it. And let's talk again real soon.

1:38:29But now it's over, it's time to go. We know your problems should be coding. And your deadline is pretty foreboding. Your ticket backlog is an actual problem. So why don't you go inside? No more listening to changelog and friends. A badminton chairman in Silicon Valley. Like no one gave the gag, we'll come to an end. But honestly, that will probably be our finale.

1:39:08You'd best be slinging ones and zeros. And that makes you one of our heroes. Your list of to-dos is waiting for you. So why don't you go inside? No more listening to change, rock and friends. With Adam and Jerry and people you know Change-logging friends, time to get back into the flow Change-logging friends, change-logging friends It's your favorite ever show Favorite ever show