Watch on YouTube: https://www.youtube.com/watch?v=H3h1WICelqs
Follow Sunil: https://twitter.com/threepointone
Watch on YouTube: https://www.youtube.com/watch?v=H3h1WICelqs
Follow Sunil: https://twitter.com/threepointone
- [00:01:40] React and Temporal, Declarative vs Imperative
- My Temporal Explainer: https://twitter.com/swyx/status/1417165270641045505
- [00:12:57] State Charts and Lucylang
- [00:17:08] The Future of React
- [00:25:03] React Streaming Server Rendering vs SSR/JAMstack/DSG/DPR/ISR
- [00:33:13] Next.js and the Open Source Commons
- Third Age of JS
- Benedict Evans (not Sinofsky) on Word Processors: https://www.ben-evans.com/benedictevans/2020/12/21/google-bundling-and-kill-zones
- [00:45:16] ESbuild vs SWC vs Bun
- Bun (Jarred Sumner) https://twitter.com/jarredsumner/status/1390084458724741121
- [00:50:46] Let Non-X Do X: Figma vs Canva, Webflow vs Wix/Squarespace
- Canva vs Figma valuations https://twitter.com/swyx/status/1438102616156917767
- [01:06:33] React Server Components and Shopify Hydrogen/Oxygen
- [01:09:18] Categorical Imperatives of Web Platforms: Cloudflare vs AWS, MongoDB vs Auth0, Gatsby vs Netlify
- [01:18:34] Wrap-up
[00:01:40] React and Temporal, Declarative vs Imperative
[00:01:40] swyx: Okay. So the first topic we want to talk about is React and Temporal, right?
[00:01:54] swyx: That's the hope. I don't know if like my explanation of Temporal has reached everybody or has reached you. There are three core opinions, right? The first is that whenever you cross system boundaries, when you call it external API. So when you call internal microservices, there's a chance of failure and that multiplies, the more complex the system gets.
[00:02:11] So you need a central orchestrator that holds all the retry states and logic, as well as timers And it tracks all the events and is able to resume from it from failure.
[00:02:21] Second opinion that you should have is you should do event sourcing rather than try to just write your business logic and then instrument with observability logs after the fact you should have your logs as the source of truth. And if it's not in the log, it did not happen.
[00:02:34] And then the final piece is the workflows as code, which is the one that you're focusing on, which is the programming model, in the sense that like all the other competitive workflow engines, like, Amazon step functions, Apache airflow, Dagster, like there's a bunch in this category.
[00:02:48] They're all sort of JSON and YML DSLs, and the bind that you find yourself in is that basically you're reinventing a general purpose programming language inside of these JSON and YML DSLs because you find a need for loops, branching, variables functions, all the basic stuff. And, people find that like at the end of the day, all this tooling is available, you just have to make it run in inside of a general purpose programming language. So that's what Temporal offers.
[00:03:12] But it's very interesting because it kind of straddles the imperative versus declarative debate, right?
[00:03:17] React, people view as declarative. And I think it's mostly declarative, like there's imperative escape hatches, and because it's declarative, people can have a single sort of render model of their entire app for the entire tree. And I think it makes sense to them.
[00:03:32] And you're saying that that's better, right? That's better than the imperative predecessor of like jQuery and randomly hooking up stuff and not having things tied up together. You sounded like you want it to
[00:03:42] Sunil Pai: interrupt. So it's actually two things. One is the jQuery had an imperative API, and then they went way too hard into the declarative side with templating languages and then started reinventing stuff there.
[00:03:54] So really react was like, no, you need access to an imperative language to create, you need a fully featured programming language to generate description trees like Dom trees or in this case, a workflow graphs.
[00:04:10] swyx: Got it. So it's kind of like a halfway solution, maybe, maybe anyway. So the problem with us is that we're trying to say that imperative is better than declarative, for the purposes of expressing general purpose business logic, which is an interesting sell for me because in all other respects, I'm very used to arguing to declarative is better.
[00:04:33] Then there's also an idea that people should build declarative layers on top of us. And I, it's just a very interesting, like back and forth between declarative and imperative that I don't know where I really stands apart from like, wherever we are is never good enough. So we need to add another layer to solve the current problems
[00:04:51] Sunil Pai: there.
[00:04:51] So there's a phrase for it and I forget what it's called the mechanism. It says that, uh, the system that allows you to execute stuff should not be the same system that prevents you from doing bad things. So there's a core, which is basically a fully featured API. And then you put guard rails around like the experiences.
[00:05:29] I need an escape hatch to actually like, do something like really funky here, X, Y, and Z, that that's not even well expressed in either the type system or sometimes even the language itself. You need to like hack it. And like, you might even email a couple of things. Uh, and in react, this was, I think when react came.
[00:05:47] It wasn't just that it was a, oh, like there's JSX. It was very much, uh, okay. Uh, I have a lot of existing code, so I can add, React to one part of it and then hook onto the DOM, it renders and then have like this whole jQuery widget that I would like render onto the thing. Uh, so it gave you this whole incremental part to adopting the system, but then like after a point, like react consumes all of it.
[00:06:11] And the fuck up with react is if you go too hard into react, doing stuff like animations is like impossible, which is why like we are at least a year or two away from a good animation API in React, or while you use, Framer or whatever Framer has become right now. Like frame of
[00:06:27] swyx: motion. No. Um,
[00:06:31] Sunil Pai: Yeah, but he's working.
[00:06:53] swyx: beautiful.
[00:06:54] It's a perfect blend.
[00:06:56] Sunil Pai: Right. And you would suffer with this in. So there was the jQuery prototype phase, which was like directly imperative. And then they went hard in the other direction with type templating languages, like Jade and dust. And, uh, there were a number of popular ones at the time. And that's when like even Angular 1 became super popular because they're like, here's the whole kit and caboodle full whole framework.
[00:07:18] And then React came and said, oh, well just the view. But that's because they didn't want to release like really yet. And they were like, yeah, this is all you need and the whole ecosystem. But anyway, so in temporal temporal for me is particularly interesting for that because it is now clearly making that.
[00:07:35] I hate the phrase, but it's a good one. The paradigm shift of like how you start thinking about these systems and you just write some fucking code and then like you start adding on bits and guardrails for the things you want to do, which is on for the few hours I spent going through the docs and failing to get it running on my laptop.
[00:07:53] That's my understanding of it. Feel free to correct me.
[00:07:56] swyx: Okay. Yeah. And I think you're right, actually, I'll try this messaging on you because, it's something that we're consciously designing for. In fact, I have a, one of my API proposals was, reacts like API for tempo. And so essentially what we enable you to do is bundle up each individual service or job into a component that we happen to call workflow.
[00:08:21] And my struggle here is that I currently tie component to workflow because what is the component like? It's, it's something that's self-contained that is a deterministic. Like it has a strict rule of execution from top to bottom, right. It just does the same thing every single time, uh, where we differ and why I struggle with this is because we put all the side effects into things that we call activities.
[00:08:44] That's where all the non-deterministic stuff goes. And that one gets retried, basically at Temporal's will and essentially Temporal is serving as the central runtime or framework that has knowledge of all these workflows and activities. And can re-render them based on its internal rules, I retries timeouts, uh, heartbeats, all that good stuff. So I struggled with things like, which is the component and which is the hook or the effect.
[00:09:08] And then there's other concepts. So, uh, we have ways to send signals into individual workflows, right? That's a very important property of the system that you can send data in while it's running and you can get data out while it's running. I'm not sure that's reflected in React at all. So maybe I'm stretching the analogy too much,
[00:09:24] Sunil Pai: Solid, had an answer for that the word signal. So like solid JS. This is by Ryan Carniato the Marco folks, signals are a first-class concept in the framework. Again, I haven't dived into it in detail in a while, but it feels like an important thing. And I always wondered why React actually didn't have it because props are something that you just like pass.
[00:09:46] Right. And it's just a value, like if you like plot it on a graph, for example, it's, let's say if you had to have like a graph of binary values, it would be either zero or like one, and that would be the shape of the graph, but signals are something that can be like something that happens and yeah, just pops up and goes down, like pressing a key on the keyboard.
[00:10:06] And that's actually not so easy to define in a, in a react like system, like, uh, which is why it's kind of hard to build like audio processing graphs with like React or JSX. Um, I don't have like a good answer. I'd probably have to like hack on Temporal a little more, but the idea of like signals as a channel, through which you can like send information and having it as a first class part of the system is something that's not represented well in well, in React at least.
[00:10:33] Yeah. Well,
[00:10:34] swyx: isn't that in an action? For reducers
[00:10:38] Sunil Pai: and event effectively. Yes. Like it's basically one of those actions.
[00:10:42] swyx: The problem is that everything just ties right into the component tree instead of just having the component B and sort of isolated unit that can function independently.
[00:10:50] Sunil Pai: That's the other thing, which is a workflow engine isn't a directed acyclic graph. In fact, it could have cycles, it could have cycles and it could have a number of other things, which is the
[00:11:00] swyx: beautiful thing, by the way.
[00:11:02] For us coding, a subscription platform literally is charged Stripe sleeps 30 days, charged Stripe again, and then infinite loop until you cancel and then you break out of the loop.
[00:11:13] That's it.
[00:11:13] Sunil Pai: That's awesome by the way. So I was actually thinking that someone's going to implement not someone's going to implement, uh, someone's going to use Redux saga on top of Temporal, that's what I was thinking, because then you will have generators that define like long running processes that are just talking to each other.
[00:11:30] I think that would be good. CloudFlare also loved Temporal, by the way, like we were talking about it, like for awhile, they're like, oh, this is like fundamentally a new thing. And as you can imagine, some engineers were like, well, why isn't this running on workers? I'm like, I don't know why isn't it running on workers?
[00:11:43] Like maybe we should get it there.
[00:11:45] swyx: It is fairly heavy duty right now. We're trying to reduce that to a single binary, which could maybe run a workers. I'm not sure about the memory requirements that you guys have. It could, it's just not a priority for us based on our existing users.
[00:12:00] Sunil Pai: Um, I was just, I was saying what they're saying.
[00:12:04] They want everything to run on workers and I'm like, dude, it's just like one small, weird isolated like condo.
[00:12:10] swyx: Ironically we also using V8 isolates for our TypeScript runtime. And that's just to make sure that people don't do non-deterministic stuff. So we did mock out everything, which is also pretty cool because whenever you use a library with, like setTimeout inside of that library, that persists to us as well.
[00:12:39] So most of, because you have to inject them into the environment of the V8 Isolate, you can't just randomly import stuff that as freely as you would in a normal node environment. So dependency injection and becomes a topic for us.
[00:12:57] State Charts and Lucylang
[00:12:57] swyx: Um, yeah. We actually clashed a little bit with David Khourshid because David is on this warpath of like everything in a state machine, right. Everything in the time-tested 40 year old JSON format that describes state machines. And we actually thought we were going to be competitive with him for a while because for him, the thing about writing imperative code is that it's prone to bugs, right? Like you can not really see the full, possibly the full span of like all the possible states that you're exposing, but in a state machine everything's explicit so he was butting heads with our founder for awhile.
[00:13:31] But I think recently he decided that he is better at building on top of us than trying to compete with us on the reliability front. So that's, that's kind of an interesting evolution that has happened over the past year on this topic of declarative versus imperative.
[00:13:44] I'm still like coming to terms with it. Like I'm not fully okay with it yet, but, it clearly is more expressive and that's something I am Very in favor of, and I have genuinely looked at like the workflow solution from Google, the workflow solution from Amazon, and they are literally have you write the abstract syntax tree by hand in JSON and that's just absolutely no way that that's going to work. So I'm pretty down with the imperative approach for now.
[00:14:09] Sunil Pai: Well, that's, I figured at some point you will run XState on it and extent should work fairly well. I think contemporary, I don't see why it would. I think that that would actually,
[00:14:19] swyx: Honestly, I'm not really sure what he's going to charge for. He's pushing the idea of state machines and making it more of a commonly accepted thing.
[00:14:26] Sunil Pai: Well, his pitch isn't even state machines. It's very specifically state charts and I love state charts. I even bought the book by the way, the Ian Horrocks $700. So when I got it on Amazon, it was $180. I was like, cheap. Let's do it. I got really lucky at the time. It, it fluctuates like mad by the way that that value, well, you should expense it now is what it is.
[00:14:46] Um, but, uh, what struck me about the thing? Here's what I tried. I really liked it. And I took a course, a couple of steps back and I was trying to understand, well, why isn't it like a success? Why don't people get into it? And the truth is that this falls not just into the intersection of this is the intersection of like computers and humans in the sense that sure.
[00:15:07] There are things that can be correct, but there are things that can be expressible as well. Like I don't even know what code I want to write when I'm sitting down to write it. I love to like discover it while I'm writing it and really. All the syntax that we have created and abstractions, we have created around programming languages have been purely to express these things and have let's call it implicit state machines, even though that implies that it's bad.
[00:15:32] Um, so for example, if you look at state charts, there's no real good way to compose two state charts together. You have to like manually start wiring them together. And like, there's, you know, like you've got in react, you say, oh, combo, if you have two components to put it together, you put like a little, uh, function around it.
[00:15:49] And now it's two components in one component. So it's important not just to have a good unit of computation, but to have it like be composable with each other so that you can gather it and then make this whole nesting doll react, Dom tree of things. And I think. Until there's an actual language that supports that has state charts as a first class primitive, much like Lucy, I think that's what Matthew Phillips built.
[00:16:15] He wrote a, he wrote an actual language that compares to state charts called Lucy Lang. That was very cool by the way. Like, I really like it. Uh, well, and it's fairly young, so it's too early to say whether people like love it or not. And other than, but people like you and me who look at something like, wow, this is awesome.
[00:16:33] Let's all use it. No, like to take a while to grow. But I think that's the state charts has a bit of dissonance with the languages that it's written in right now, because it's not a first-class thing. I mean, it's adjacent object with keys and. Okay. Like we can do better maybe. Uh, but I would not bet against David and the people he's hiring.
[00:16:53] Like he's hurting some smart people, you know, they're all like pretty intelligent. So I'm curious to see how that plays out.
[00:17:00] swyx: I'm just glad that we're not competing. Uh, so that's, that's something that, that, that resolved itself without any intervention from me, which is very good.
[00:17:08] The Future of React
[00:17:08] swyx: Well, let's have this conversation since it's related, should React to be more of a DSL,
[00:17:14] you know, this conversation that happened over this week, so I'll pull it up.
[00:17:20] Sunil Pai: Uh, wait, so I've, I'm seeing, is this the whole Svelte versus React thing that's been happening over the last two, three days?
[00:17:25] swyx: Yes. So basically it's saying React is already so far down almost like its own language.
[00:17:30] They should just embrace it more. And instead of using linting to catch rule violations, just make a DSL, people are gonna use it. It's fine. And just like build things in so that it's impossible to make these errors that, that people commonly make.
[00:17:47] Sunil Pai: So this is Mike Sherov, uh, he was smoking about it.
[00:17:51] He mentioned how it shouldn't be a lint rule. And since we already have customs, insects and GSX, he should introduce a couple of other things. So as you can imagine, the react team has thought about this a lot. So the big problem with this all boils down to that fucking dependency area on use effect, by the way, that's the one that trips, everything else is fine.
[00:18:09] Like you stayed all that is like fine. You can get. This is
[00:18:13] swyx: what it was. Yeah. People want like state something memos and things like, you know, just build the reactor primitives into the language.
[00:18:22] Sunil Pai: So yeah, I think this, this actually, isn't a bad idea and I think that was the whole deal with hooks. Whereas what's the phrase that they use in the docks.
[00:18:30] A sufficiently advanced compiler might comply with these things at some point, and you're like, oh wow, great job. On pushing that responsibility onto the community, React team, well done.
[00:18:41] swyx: My joke is like it's the react teams equivalent of a assume, a frictionless spherical cow from physics.
[00:18:48] Sunil Pai: Exactly.
[00:18:48] That's a perfectly spherical code.
[00:18:54] swyx: It will exist.
[00:18:57] Sunil Pai: And it's just the five of them or six of six of them hacking on this. And they have to make sure they don't break like facebook.com whenever they're working on these things. Imagine it's taken this long for Concurrent to show up and Concurrent is nice by the way. And we can talk about the server rendering API.
[00:19:14] Okay. Uh, so react right now is, uh, yeah, that's the one like that. It shouldn't just be an intruder, but, uh, inside the inside Facebook only, well, not everybody can see it, but it's an in an internal, uh, uh, Facebook Wiki page, which is a list of potential F projects. You know, how the react team has fiber, whatever the hell.
[00:19:47] Right? So there's a list of these projects that, or when we do this, uh, project F F I forget what the one for, uh, uh, animation that's called, is it called flat? Flat was the dumb one. And so there are lists of them and there are about 15, 20. I'm pretty sure my India has done. So Hey, so, uh, there's a list of them.
[00:20:09] And if you look at them and you start assigning values in terms of work, oh, this is about six months of work. This is about, uh, another six months of work. It strikes you that there's a roadmap for about five to 10 years. At least if not more than that, I mean, look at how long it took to get like this. Of course this was very more foundational.
[00:20:26] Those could probably happen a little quicker when it comes, which means the react team is like solely aware of what's missing in react right now. And to an extent that they can talk about it because if they do it becomes like a whole thing and like don't really engage in that conversation. They don't, I, I, and I don't blame them for it.
[00:20:44] It's very hard to have this discourse without somebody coming in and saying, well, have you considered CSS transitions? I like that. Yes, we have. We have, we have considered CSS a lot. Uh, so, uh, so. There are all these projects like a sufficiently advanced compiler that compiles down to hooks. There's the animation API.
[00:21:04] There's a welcome current, et cetera. This whole data fetching thing has been going on for years. And now it's finally starting to come to light, thankfully with collaboration, with the relay team and effectively all of the core when they built out facebook.com and, and that is the length that those are the time periods that Sebastian looks at and says, yeah, this is how we can execute on this because it can be prioritized.
[00:21:33] It has to be prioritized by either Facebook wanting it or making Facebook wanted. So for example, the pitch was, Hey, let's rewrite facebook.com the desktop version because they haven't, it's a film mishmash of like hundreds of react routes on one page. It should be a single react route that does this thing.
[00:21:52] Now that we have gotten management to agree to a rewrite, let us now attach it to the concurrent mode thing. And that was also part of it, which is in the older version, there was a lot of CPU fighting that used to happen between routes, which is why the whole work for the share dealers started and took like two years to like fix effectively.
[00:22:18] swyx: was that ever offloaded to the browsers, by the way? Like I know there was an effort to split it out of react.
[00:22:24] Sunil Pai: So I think last, I checked they were talking to Chrome literally every week.
[00:22:29] Uh, but I think it's also been down to, uh, well, what Chrome wants to prioritize at the time. I think it is still going ahead again. It's the sort of work that takes years, so it's not going ahead. Nice and slowly, uh, which is why. Which is why it's architected inside react for the same reason as like it's attached to global and then read off the global.
[00:22:52] I think it's also why you can't have two versions of React on the same page. There's the whole hooks thing. But also if you have two versions of React, and they'll just start fighting with each other on the scheduler, because the scheduler would yield to one than to the other than to the other one.
[00:23:08] And there would be no like central thing that controls what is on the scheduling pipeline. That's from the last, again, this conversation is at least two years, or maybe they fixed that, but that's the goal of the dealer. There has to be one scheduler for the thread that everybody comes on to, and like tries to pull stuff, uh, with it.
[00:23:26] I think it will become a browser API. It's just a question of like, when, like, yeah, I mean, the shared dealer in react itself has undergone so much change over the last three years. Uh, so maybe we should be glad that it isn't in the browser yet, because like, it's changed so much. It's coming there. It's I mean, the fact that they're releasing in November is a big deal.
[00:23:45] swyx: You said there's so many projects that you want to ship, and the way to ship it in Facebook is to either convince them that this feature itself is worth it, or you tie it together with something else, like the Facebook, I think it's called FB5 rewrite.
[00:24:00] Sunil Pai: Oh yeah. I think it's good for them. Like it worked because the Facebook, facebook.com is now more performant. Like it actually works well and they don't have CPU fighting. The fact that Facebook itself is becoming slightly irrelevant in the world is a whole other conversation.
[00:24:17] swyx: Well, you know, I still use my billions, so, uh, it's it's, it improves the experience for them.
[00:24:23] Sunil Pai: I'm only being snarky.
[00:24:25] swyx: Uh, but I, you know, hopefully hopefully you're like, you know, there's other properties like Instagram and WhatsApp and what is, uh, which hopefully it will apply there. And then obviously like there there's the VR efforts as well. Absolutely. Yeah.
[00:24:39] Sunil Pai: And that is the future. In fact, uh, several components also happened because they suddenly realized what they could do for how the deal with server components and server-side streaming rendering was never about an SSR story, or even a CEO.
[00:24:54] Facebook doesn't give a fuck about SEO, right. It was about finally they figured out how to use concurrent mode to have a better UX altogether.
[00:25:03] React Streaming Server Rendering vs SSR/JAMstack/DSG/DPR/ISR
[00:25:03] Sunil Pai: So, okay. I should probably just keep Server components aside for right now.
[00:25:06] And I'll just talk about the new streaming rendering API. Okay.
[00:25:09] Okay. So I know there's like about three styles of rendering.
[00:25:14] I say legacy, but legacy is such a dirty word. I don't mean it in the form that it's old it's in fact,
[00:25:20] swyx: traditionally, like, sorry.
[00:25:44] Uh, uh, web components actually falls square into this, where it just comes to life only on the browser and then like make stuff interactive. Uh, then there's the whole client fully client side rendered one. So this is create react app or, well, a number of like smaller players then there's server side rendered.
[00:26:04] And so as I rendered is actually like, it's not just next year. It's also your Gatsby. I feel like pretty much every, uh, react framework now has some kind of service side rendering story. Okay. So the next slide goes into what types of server-side rendering things happen.
[00:26:20] swyx: there are a lot of subdivisions within here, right?
[00:26:22] Like, uh, Gatsby is up here trying to reinvent like D S R D P R or something like that, which is like deferred,
[00:26:29] Sunil Pai: static,
[00:26:32] swyx: DSG, deferred static generation. That's the one. My former employer, Netlify also DPR, and is all, these is all like variations of this stuff with,
[00:26:41] Sunil Pai: like, it's a question of where you put the cache is what it is.
[00:26:46] It's a TLA three letter acronym to decide where you put the caching in.
[00:26:49] Yeah, so there's the whole JAMstack and that's like the whole Netlify story, but also CloudFlare pages, or even GitHub pages.
[00:26:56] There's no real runtime server rendering. You just generate a bunch of static assets and you Chuck it and it just works. Then there's fully dynamic, which would be next JS without any caching. Right? Like every request gets server-side rendered then like a bundle loads on top of it. And, um, like suddenly makes it alive, like sort of like it hydrates it.
[00:27:16] And then after that it's effectively a fully clients rendered application then there's okay. So I just said ISR, but like you said, there are like three or four after this as well. There's this whole DSP. Yeah. Oh wait. So the new streaming API is actually fundamentally new because. I don't know if people even know this, but react already has a streaming rendering API.
[00:27:37] It's called a render to node stream. I think that's the API for it. And the reason that that exists is so that, uh, only for a performance thing on the server where otherwise synchronous renders would block like other requests. And it would make like if for a server that was very, uh, uh, there was heavily trafficked.
[00:27:57] It would become like really slow. So at least with the streaming API, yeah. That's the one learner to notice the stream, at least with this one, it wouldn't clash and you could interleave requests from there happening, but it didn't solve like anything else, like nothing, you couldn't actually do anything asynchronous on it, which is kind of that fucking sucks because like, it looks like it's an asynchronous API, but you can't do anything asynchronous through it.
[00:28:18] It's the only thing that, okay, so vendor to readable stream is cool because I can, even if you go to the very last slide last bit, once. You know what this is, where the very first link open it up. Like it says react, lazy.cool computer club. So this is the demo that they have that exists with this new API.
[00:28:36] This is what they link to. So if you refresh it a couple of times and I'll show you something that happens here, so you see the little spinner that shows up there and then the content loads. Yep. So, um, you know what, maybe I can share my screen because I want to show like a couple of things. Uh,
[00:28:53] swyx: yeah. I'll fill in some context, like I knew that the renderToNodeStream API was not good enough, basically because everyone who is doing SSR was doing like a double pass render just to get the data in. Um, and I noticed a very big sticking point for Airbnb so much that they were almost like forking react to something like that to,
[00:29:11] Sunil Pai: they invented a caching API.
[00:29:13] They did like a whole bunch of things. Okay. So if you have a look here, you'll see that there's a little bit of spinner and then the content comes in. But now what I'm going to do is I'm going to show you the actual HTML. So let's just go to prettier and just pretty far this, for that, we can see the content and I'll show you something that's very like fundament.
[00:29:32] That's the playground playground paste, big HTML. All right. So are you looking at this HTML it's rendering rendering by the way, this, these are special comments that mark suspense boundaries. It's very cool. If you come down here, you'll see a dev, which is the spinner. So this is the spinner that you see when you refresh the page.
[00:29:52] So this is. And then the rest of like then, like the, like the bits that are below that close and the HTML closes, but content still start stream is streaming in at that point. So like, this is the actual, like devs that are coming in with the content. And then a script tag gets injected that says, Hey, this thing that just came in, shove it into where the spinner was.
[00:30:13] This template
[00:30:14] swyx: tag is so small. I would, I would have imagined it was much bigger.
[00:30:18] Sunil Pai: It's not. So by the way, at this point, the react has not loaded. This is happening without react. This is just a little DOM, much like swelled ha uh, just a little operation that does it. So you, you, you get this content. And, uh, so, so that's the first feature which is that suspense.
[00:30:35] It not only works out of the box, but fallbacks and replacing or fallbacks with actual content also happened. Um, I want to pull this outside of this main window to show you something. Um, so you can see the content load in, but keep an eye on the loading spinner. Okay. Just to prove a point. So the content loads in, oh man.
[00:30:56] Oh, is it cash just that way? Uh, the content loads in, but the spinner is still going on. That's because there's an artificial delay for the react bundle to show, to show up. That's the point of this demo, which is to show that it can do async. Now you can imagine that it's not just one part of the page.
[00:31:42] So it loads before the react, the server. Finish the streaming. So at that point, the react says, fuck you, I don't care about the streaming bit anymore. I'm taking over, it's now a client set up like just automatically out of the box, because now that would be faster. So it basically raises the client and suicide.
[00:31:58] So suspends working out of the box itself is like a big deal first. So people will start using it like with react dot lazy, but then with data fetching and a bunch of slate styling solutions, which they're also working on. Um, but this is the new server entering API. The reason I was talking about this, I keep losing context about these things.
[00:32:19] I should stop sharing, I guess. Um, the absolute best feature of this of course is the reason why is something that comes out of Facebook, which is it works with existing applications and you can incrementally add it. So the first thing you will do is you'll take your render to string that one line somewhere in your code base, which says rendered to.
[00:32:39] And you'll replace it with vendor to notable readable string. I mean,
[00:32:43] swyx: either way 99% of users have never used render to string. Right. That's what next year is for.
[00:32:51] Sunil Pai: Well, that's the, that's all my God. That's part of a whole other conversation, right?
[00:32:54] swyx: This is rendered a string as a service.
[00:32:59] Sunil Pai: The moment you update next, year's your version of next year? So work on yes.
[00:33:04] swyx: Which is good, which is good. Right? Because, uh, people won't even know and they will just benefit, but it's, it's a little bit bad. Okay.
[00:33:13] Next.js and the Open Source Commons
[00:33:13] swyx: And this is a little bit of my criticism, which is that your blessing, a meta framework, at the expense of all the others, right? Like which admittedly have not been as successful, but, uh, basically reacts Chrome, picked a winner and it was next year.
[00:33:27] Sunil Pai: I've been thinking about this so much. Oh, look, it let's get into them at our conversation now. So let's standard disclaimers. I think Guillermo is a mench.
[00:33:35] I think the people who work there are incredible. There are some people I'm close to. I'm so happy for them. I know people on the Chrome team who work with these folks. I love them as well. Nicole for me is, uh, is a hero. Uh, and of course the React team at all my buddies, I love them. Okay. That being said, the React team is six people and they don't have the time to build the meta framework and Guillermo, uh the one thing he's incredible at is he's great at building relationships.
[00:34:03] He's just amazing at that. Like he, uh, in a very genuine way, like this, there's nothing like ulterior about it. Next JS is open-source and runs on any node runtime and it's designed to do so. There's nothing about it. That's become special on Vercel. Because of that the React team felt, feel like, okay, fine.
[00:34:20] We can have a primitive and meta frameworks will solve it. And let's just make sure it works with next two years, because so many other people who are just reach out to them and say, Hey, this new API is showing up. Uh, this is not just with next.js. It's a similar thing is with like react testing library.
[00:34:34] When the new activity I showed up, right. I made the PRS to react testing library. I was like, what you should do is have every function and react testing library be wrapped in back act. So nobody really has to like use the API by hand. I just, it's now it's the D and it's a very good testing framework, the Chrome team.
[00:34:53] And this is my, I'm not saying this, like, it's a bad thing. I think they did the right thing. The Chrome team realized that if they provide performance enhancements to next years directly, they can have so much impact on the internet because so much of the react tool is running on next year. So fixing how the images are loaded in next year certainly makes the internet faster.
[00:35:15] Yeah. And maybe that's what we should do also like for the accessibility, just ship acts in, uh, all the acts rules in development mode, either in like react Dom directly, or at least the next years. Oh yeah. The sweatshop, the axles. Yeah.
[00:35:33] swyx: Oh, they're enabled by default. And, uh, your, your app one compile, uh, actually I think it would warn you won't fail by a worn.
[00:35:40] Sunil Pai: Okay. So you should be making the swag folks should be making way more noise about that. That is such an incredible draw for accessibility.
[00:35:48] swyx: The thing is like, uh, if you encourage, if you think that your, your problems are solved by X, then you're taking a very sort of paint by numbers approach to accessibility.
[00:35:57] Right. Which is actually kind of against the spirits of, of, uh, what people really want, which is, um, real audits with like tap through everything. Like the stuff that machines could catch is so little,
[00:36:08] Sunil Pai: I agree. The whole point of actual SIS to make sure that all the low hanging fruit is done by default.
[00:36:15] It's like TypeScript, like I guess, which is a TypeScript. Doesn't solve all your bugs, but the stupid undefined is not a function once it does. Yeah, exactly. Make sure that your images have. Just by default, like we can have stronger conversations about tab order once you make sure all your images have all tags.
[00:36:35] swyx: Uh, okay. Anyway, so, so yeah. So first of all, yeah, I agree with you on the, on this Chrome. And, uh, I think this is opensource winning, right? Like, uh, there's a, there's a commons. Vercel built the most successful react framework, Nate. They went the investor really hard at it. They had the right abstraction level, you know, not too much, not too little, just the right one.
[00:36:55] Uh, and now everyone is finding them as like the Schelling point, which is a word I'm coming to use a lot, uh, because you know, that is the most impact that you reach. Uh, so no hate on any of them. It's just like it happens that a venture backed startup benefits from all of this.
[00:37:11] Sunil Pai: Can you imagine how hard it makes my job?
[00:37:13] We don't run, not on CloudFlare workers, which means Next.js doesn't run on it. It's annoying.
[00:37:19] swyx: Oh, is there any attempt to make it run?
[00:37:22] Sunil Pai: There are a couple of ways where we can get it to work, but it like, it's a lot of polyfill and, uh, we'll get that. Like, I expect it to be fixed within the next three to six months, but out of the box, it doesn't run on it.
[00:37:35] And for me in my head, it doesn't, it's not even about CloudFlare workers. I'm like, oh shit. That's what makes Bezos like even richer because everyone's got, has, if you want to use Nadia using AWS or Lambda. And that just means more folks are using AWS. I'm just like, okay, I guess. Sure. I know you work there as well, but it's just very annoying to me where I'm like, shit.
[00:37:56] What's even more interesting is that node is now moving to implementing web standard APIs inside of it. So they already have the streams implementations. They will have fetch fetch will be a node API. Like it will be implemented based on standards, which means the request response objects. And once that happens and people, if people build frameworks on that, then you can say that it will run on CloudFlare workers because the cloud fed worker's API is also like a standards based thing.
[00:38:21] So it's an interesting shift of like what's happening in the, in the runtime world. Also conveniently the person who implemented the web stream implementation at node just started at CloudFlare like last month, like James. Oh, James
[00:38:38] swyx: now. Okay. Yeah. I recognize
[00:38:39] Sunil Pai: a great guy by the way. Uh, very, I just love these people who have like clarity of thought when they talk James as well.
[00:38:57] Sunil Pai: I have an internet information junkie problem.
[00:39:00] I replaced the weed smoking habit with a Twitter habit. This is what it is.
[00:39:05] swyx: You're not unlike some magic mailing lists that like tells you all this stuff. Okay.
[00:39:09] Sunil Pai: Like reading the tea leaves is what it is. Like. I keep trying to find out what's going on. The problem
[00:39:14] swyx: is I, I, I feel like I'm ready. I'm relatively plugged in, but you're like, you're way more plugged in than me.
[00:39:22] and then this development with node adopting web standard APIs, um, is this a response to Deno?
[00:39:28] Sunil Pai: I don't know if it's a response to Deno because I know Mikeal Rogers wrote about this. Like your. That we made a mistake by trying to polyfill note APS and browser code with like modules and stuff.
[00:39:41] Right? Like that's what the whole browser, if I, during those days, when we started actually using the same module system and the word isomorphic came up, what ended up happening was naughty APIs were polyfill in web land, but what should have happened is we should have gone the other way. And it would have kept like bundle this bundle size problem would have been a web smaller pro problem right now, just because of that.
[00:40:07] So I know that the folks at not have been thinking about it for awhile, maybe Deno finally pushed them to do it, but I don't, I don't, I don't, I don't think it's like that reductive. I don't think it's just, it's just dental. It's very much a, this is the right time to do it and we actually can do it now.
[00:40:22] So let's like flesh it out and do it the right way. Uh, and it's hard to do it in, in no, right. It's not just that you can just implement this thing. Like, what does making an HTTP server mean now? Because the request response objects are slightly different in shape. So you have to make sure that you don't break existing code.
[00:40:39] So it's not as simple as saying, oh, we're just implementing the APS. That being said, having fetch inside node proper is going to be great. I think. Excellent.
[00:40:55] Which is a blog post that I wrote last year that, um, honestly I feel quite a bit of imposter syndrome around it because all I did was name a thing and like it was already happening. It was, you already saw, like, I think basically when, when COVID hit, a lot of people were. I have a lot of time on my hands, I'm going to make new projects or something.
[00:41:14] Um, uh, and then, yeah, so I just, I named it and I just called it a few trends. So the, the trends I I'm talking about are the rise of IES modules first, you know, in, in development and in production, uh, concurrently the death of 11, which I'm also tracking.
[00:41:30] Sunil Pai: Yes, those are, those are both come to fruition.
[00:41:34] swyx: Which, by the way, I think the us government will have to drop by 11, uh, sometime in the next six months or so because, uh, the, the use, the usage levels have plummeted.
[00:41:43] 3.6% of all visits to the U S government website in November, 2020 was I 11 and now that has dropped to 1.6, um, all
[00:41:51] Sunil Pai: accelerating the drop is actually accelerating.
[00:41:53] swyx: Uh, I don't know if it's accelerated it's everything, but it's under the 2% mark that the us government sets for itself.
[00:41:59] They have an opportunity to essentially say like once it's stable, you know, there's no chance that it'll ever go back up again. Uh, they could just deprecate 11 for all government websites and then that, that will be the signal for all enterprises. And that's it. Yeah. So, um, and then the second.
[00:42:15] Oh, I was going to move on the second bit. But what was your calling?
[00:42:19] Sunil Pai: Oh, just saying that this happened, like, while I was working in JP Morgan over the last year, they did the same shift, but they're like, yeah, we are now a Chrome company. Literally none of our clients are asking for this and you know, it, it was just in rules somewhere, or we need to target, I 11, some people looked at it and said, okay, fine.
[00:42:35] What happened is people are spending money on something that wasn't giving them the returns. And that's when a bank is like, yeah, we don't need
[00:42:41] swyx: to do this anymore. Like you, you can deprecate free support. Right. And, and just make, just charge for 11 support, stop spreading it out among all the other users who are bearing the cost of development and maintenance.
[00:42:54] The other one was collapsing layers, which is the death of Unix philosophy. Like , we used to have one tool does one thing, but now we want to combine everything. So, uh, Deno and Rome both have ambitions of linter format or test runner, all of that into a single binary, because the idea of what we want out of a default runtime has changed, uh, from a, for a very minimalist thing. And I always made the comparison to what word processors used to be like.
[00:43:18] So, are you aware of Benedict Evans? He has a blog post, which is amazing about what a job of a platform should be. And he talks about like in 1980s word processors used to only let you type words. And if you wanted a horizontal layout, if you wanted word counts, if you wanted footnotes, these are all plugins that you buy and install separately.
[00:43:38] Right. Okay. So, but as we evolve, as we just use all these things, we realize that these are just like the same tool that we want out of a word processor. So then they absorb all these features instead of plugins. They're just part of the platform now. They're there now in the new table stakes.
[00:43:53] So I make that analogy to the runtimes that already doing, right. Like, Node used to be this like much more minimal thing. And, uh, but now we are expecting more and more out of our default setup with all these tools . Um, it's also very wasteful because when each of these tools don't know each other, they're all parsing their own ASTs running, running their own code. And then yeah, that's the whole
[00:44:12] Sunil Pai: proposition, but yeah.
[00:44:14] swyx: Any, any tool that collapses layers will, we will meet this, like, ESBuild, um, collapsed. Like a standard web pack would do like five or six AST runs. ESBuild collapsed it to two to three. That's a source of its speed as well.
[00:44:27] Sunil Pai: One of my favorite facts about ESBuild is that it is faster to minify the code than to not modify the code when you run. Yes. And the reason for that is because when you try, when it tries to do the full AST, keep comment notes, everything else, it has to do a lot more bookkeeping, but the moment it just ditches all those things, because ESBuild doesn't do like full magnification, like something like a torso, but it does do like a smaller symbol substitution, white space, uh, uh, removes all white spaces.
[00:44:59] And it does like some dead code elimination. Uh, and it's a lot more work to keep the bookkeeping for everything and all the white space notes than to not do it. So he has built is actually faster when you have a modification turned on, love it.
[00:45:14] swyx: It's amazing. It's amazing.
[00:45:16] ESbuild vs SWC vs Zig
[00:45:16] swyx: Do you have opinions on ESBuild versus SWC?
[00:45:18] Sunil Pai: Okay. So I like ESBuild. Uh, because I was very strongly looking for something a lot more opinionated. I've noticed that the reason that code basis Surat usually boiled down to the acute decisions that you make. Like in the very beginning of the project, you can do anything. I mean, whichever dumbass came up with the idea of baby plugin, macros has like ruined a lot of lives.
[00:45:41] It was me. I came up today, but that is like, then you're like tight. So the thing that ESBuild does is very like its creator, Evan Wallace, which is that it's, it's one of a kind like, he's not really interested so much in having community, uh, uh, PRS or like having suggestions on how it should be built. He has a very strong vision of what it should be like, which is why there are no AST level plugins and all that jazz.
[00:46:08] And because of that, because of, like I said, because he's collapsed the lyrics and collapse, the size of the development team to just himself, he has like such a clear vision on what it should be. So it w is good. It would be great for, I want to say 95% of projects that fall under the things he has designed at four.
[00:46:28] Okay. Uh, and that's a lot of applications. That's a shit ton of applications. That's like everything, but your host, if you need anything, uh, unique, I'll give you one. That's like a very good use case that is bill will never use. Do you know what, um, uh, really has this idea of persistent queries. Okay. So like for whoever's hearing who doesn't know it, right?
[00:46:52] Like you can write a query inside Java. And when it compiles it out, it takes out the query and replaces it just with an identifier, like, like a little eight character identifier. And it hosts that query instead of like on the service side. And it says, oh, that eight, eight character query, you can just hit it as a restaurant point now.
[00:47:29] I guess like you can still use Babel would, uh, SWC is meant to be a platform and which is why next years will use it because next gen is the meta framework, not just for react, but also for like some programming opinions, extracting get server props, get started, props, which one you want to be that this thing after server components comes into play, but a number of things like there will be people who always want to do.
[00:47:54] The emotion macro now is like fairly, uh, popular that they will want to use it. So I assume they will implement it in, uh, interest. I know. Do you know what bun is by the way? Do you mean, do you know, how are you following Jared Sumner? Some
[00:48:10] swyx: summers, no, wait, so
[00:48:13] Sunil Pai: key is reimplementing ESB, but in a language called Zig it's another systems programming language.
[00:48:20] And he's his claim is that it's about three times faster than you spell it right now, which is already some 200 times faster than Babel loader. It is just our web pack, but it's a language you said it? No. So the language is called Zig lines at AIG, but the thing he's building is called a, B U N. He hasn't shared it in public yet.
[00:48:41] I think he's actually planning on sharing it like next week. Like I think it's that imminent. He's been sharing numbers right now. Yeah. That's the guy, Jared. Uh, I love, I should've followed him like a while ago, create great feed, uh, excellent content. And like, he's, he he's thinking that he's going to like implement.
[00:48:57] He might actually implement an AST level, uh, uh, plugin, micro API, possibly just implement the emotion one. I think he was just, yeah. See, oh, that's like literally the tweet would write under the main one right there where he's like, Hey, what if we actually just did this in? Uh, oh,
[00:49:14] swyx: he's right. He's he's right with you.
[00:49:17] Yeah. Like he's
[00:49:17] Sunil Pai: just talking about it, like right there. So, uh, so SWC versus ESBuild, I don't think is the conversation. I think ESBuild will have a rise. A bunch of people will use it. The nice thing, the best feature about ESBuild is because there are aren't any like cute decisions. You will be able to move away from it to whatever succeeds.
[00:49:39] Th there's nothing customer
[00:49:40] swyx: that I believe that was Evan's original idea. That IES build was a proof of existence that day there's a better way. And that he stuck to it for way longer than I thought he would.
[00:49:51] Sunil Pai: People are using it in production and everything know everything about the designers that it's replaceable.
[00:49:56] That it's just a,
[00:49:59] swyx: that's wonderful. Isn't that amazing when people design their stuff? W. You know, it
[00:50:04] Sunil Pai: isn't kind of pressure that he would have had the best. Thank goodness. It was the successful CTO of Figma with money in the bank who is implementing this and didn't have anyone to impress. You know what I mean?
[00:50:16] It was like, yeah, let's put a macro API and what else do you want? Like, whatever. No, he doesn't
[00:50:21] swyx: go. Yeah. But he just needs to police himself and no one else. Right. If you don't like it,
[00:50:26] Sunil Pai: this is during his downtime from Figma that he's working on this.
[00:50:30] swyx: Um, my, my secret theory is that he's doing this as an, as a Figma ad.
[00:50:33] Like, you know, if he, if the CTO of Figma does this for fun, imagine what it's like to work inside of Figma, you know, like of, I've heard it's pretty great,
[00:50:42] Sunil Pai: pretty great working inside of Figma too. Well, the code is like, it's really cool.
[00:50:46] Let Non-X Do X: Figma vs Canva, Webflow vs Wix/Squarespace
[00:50:46] Sunil Pai: Did you actually point out. Uh, Ken was like six times bigger than Figma.
[00:50:52] swyx: you wanna talk about that?
[00:50:53] Sunil Pai: Oh God. That's. I didn't realize until you pointed it out.
[00:50:58] swyx: Incredible. Imagine all the geniuses working in Figma and go looking at Canada and like, yo, like I, I have like a thousand times your features and your six times in my size as a business.
[00:51:10] Sunil Pai: Uh, but I hope every one of those engineers understands the value of sales and like reaching out to your actual customers because
[00:51:17] swyx: I don't think it's just sales.
[00:51:18] It's more like, uh, they're always going to be more non, like, this is a category of software called let Nanex do X, right? Let non-designers do design. Whereas Figma is clearly for designers doing design. Um, and there's always going to be like a tool, three orders of magnitude more non-experts uh, who just want to do basic shit.
[00:51:37] Sunil Pai: Oh man. I hope that flow has a multi-billion dollar buyout and at some point,
[00:51:42] swyx: uh, I mean, I, yeah, I mean there's clearly something that w the problem with flow is that. They're too close to code. Right? You have to learn CSS the box model.
[00:51:56] Sunil Pai: Yeah. I mean, they do say there's no code, but really they're a visual,
[00:52:00] swyx: if you don't know CSS when using Webflow you're screwed.
[00:52:04] Sunil Pai: that's right. It's uh, they have, they have the best grid editor on the market too. I have to say that. I
[00:52:10] swyx: mean, the UI is just amazing, right? It's just like, um, yeah, I mean, you know, there's a reason why like the Wix is, and the Squarespaces are actually worth more than the workflow and it's not just cause they were around earlier.
[00:52:22] Like, um, they're, they're just easier to use for non-technical people.
[00:52:26] Sunil Pai: That's a good, you you're talking about why did we even start talking about this? What did you want to talk about? Uh, we were talking
[00:52:39] Was there anything else that you want to talk about?
[00:53:02] And I used to like participate very heavily. And back to the thing that you, uh, that we were just discussing about the conversations that happened too, about like SBA versus MPA and about like the whole notion blow up about how they made them thing into like 800 KB. Yep. Uh, the easiest kind of discourse to have is to have like one absolutist opinion, uh, that I saw a number of people in like those threads and the surrounding threads have, which is a, well, this is bad or this is good.
[00:53:35] And, uh, that's, that's all I got to say about it. Now give me like 40 likes on this reply industry. Uh, whereas like there's real opportunity here to understand how and yeah, that's the one, that's the one with treat by the way. Clearly it got like attention. No,
[00:53:51] swyx: by the way I phrased it very neutrally. I actually was pretty careful.
[00:53:54] Cause I knew that it's going to attract some buzz. I had no idea what's going to be this much, but
[00:54:00] Sunil Pai: no, no, no. But like I'm so interested in talking about, uh, so this is what I was talking to you about, which is like, it's not just about a website at one point of time. It's about the system that generates these kinds of like artifacts, uh, of, so for example, with what, what did they say?
[00:54:24] They're there 8 47 KB right now. They're not 8 47 KV today. They were 8 47 KB. When you, uh, Uh, tweeted this, uh, on the 11th, they are not in 47 KB. Now they might be 852, or they might be 841. Are you about to check?
[00:54:43] swyx: No, no, no, no, I'm not. I'm not, it doesn't matter. The exact number. Doesn't matter. I'm going to give you another example, which also came up, which is Netflix.
[00:54:49] Remember they ripped out react and he said they have react back
[00:54:54] Sunil Pai: on Netflix. I use, are you serious on that? Wait, did they have like both Netflix, they have both react and jQuery, jQuery and react on that page right now. It's just, but like, for me, it's interesting that, which is like, I think the most insightful tweet in this was very pointed out that nobody noticed this until they told it to us.
[00:55:16] Nobody saw it. It bothered. Yeah. That's the one, like nobody bothered about it. It was still making the money. They were happy about it. And they wanted to share that. And we need more of them. We need more people to be like sharing the process because if we react very badly to these things, then fewer people will want to actually share the numbers.
[00:55:34] And you won't learn from the industry, but I don't know whether it's a good thing or a bad thing. It does mean that you can make a multi-billion dollar company with a marketing site. That's nine MB of Charles' script. And I think, I think people who have very strong opinions about how much jealous should be on a page to take a step back and wonder how do you make it?
[00:56:17] swyx: shipping velocity, right?
[00:56:19] Sunil Pai: You are somewhere, you are spending effort on it somewhere. Just so we're clear because somebody will look at it and say, fuck you, are you suggesting that we all put in that's not what I'm saying. I'm just saying that the resources, that word, but resources at these companies are limited and they are, they they're prioritized and sequenced and you should ask yourself in what order you want to do it and who you're trying to please, are you trying to please your customers and your users or the peanut gallery on Twitter?
[00:56:48] And I think that's something that like, I, it's why I don't engage so much anymore because it's so hard to communicate in once and somebody will come in with a, well, fuck you, you work for Facebook or used to work for Facebook. What would you know? I'm like, you got me that kind of ends the conversation that, right.
[00:57:04] Like I'm studying contributed to babies being burned alive or whatever it is like, this is what it is.
[00:57:12] swyx: Um, it's a nuanced debate, like, uh, because they also did some like notion clearly did some stupid stuff here. Right? Like it, it, they could have spent a day. Uh, so do you know why it was 9.9 megabytes?
[00:57:25] Sunil Pai: If I understand it was the whole notion that that was being used, the
[00:57:27] swyx: whole app.
[00:57:28] Yeah. They were shipping the whole, there was actually someone from notion, uh, answering me. Uh, it's here. Yeah. This guy's, this guy works at notion before the marketing site was another route in our, at the time 9.1 NBME and app, we load the whole app just to show the sign up button.
[00:57:44] So what,
[00:57:45] Sunil Pai: what it's worth Facebook sign up page does start prefetching actual Facebook code so that once you log in it loads instantaneously. So there's a reason to do it. It's just that it shouldn't be nine and B of course. That's
[00:58:00] swyx: yeah, they could have like took a day every, every six months or something like perfect day, you know, and do that.
[00:58:06] So that's why I'm hesitant, uh, giving them a pass for like, okay, so what your multi-billion dollar company? This is embarrassing. This is just an unprofessional. Um, so yes, but at the same time, the people who are upset about other people's sites on Twitter, uh, and then they move on to the next source of outrage the next day.
[00:58:26] Like they also very uneffective. So.
[00:58:31] Sunil Pai: I it's hard to have a conversation with them, and God bless you. I noticed that you try so hard to stay, uh, not, not emotionally detached, I guess, but that you're, you're just staying above the water and that you're not actually picking a fight or like fueling it on. But there are people who are getting in there to like, get their shots and to say, ha I've said the tweet against a multi-billion dollar company I'm better than them.
[00:58:54] swyx: no, that gets a little sad. You know, like the dev Twitters is a small microcosm of the overall universe. And in fact it's often the least powerful people and the most powerful people have no don't care. They don't care what you think. They, they just, they're looking at the OTRs and if they, if they meet it and they get promoted, they have more money.
[00:59:14] And you're, you're still complaining about your performance issues.
[00:59:19] Sunil Pai: It sucks that it's this way, because I don't know if I'm romanticizing the. Actually I probably am. Yeah, that's just, it, it feels like, oh, when, when I was younger, things were like way better than the music was better. You know what I mean?
[00:59:50] Thankfully it's a lot more diverse right now. We have people from way more walks of life. And, uh, uh, this obsession with computer science degrees appears to be dying down, which is just great. Absolutely great news because they, whenever they talk about computer science degrees, nobody actually is like, okay, what did you learn in your computer science degree?
[01:00:09] That's like useful here. There are some things that are, if you're building the shed EULAR and shady lyric and fiber for react. Yes. But really the people that I've seen do really well as web developers are the folks who read a lot of comic books and learn to design because they know how to lay out stuff and how to tell a story.
[01:00:28] And they like know how to like translate that onto the, on the side. So I've seen like, and designers who spend like a little time to learn how to do like just enough stuff to be dangerous. They're like incredible. Uh, Adam Morris is now doing a generative, whatever components.ai is, it's this?
[01:00:45] swyx: What is that
[01:00:48] Sunil Pai: strong pitch?
[01:00:49] And it's, he's taking his time to execute on it. And I love it, which is you should be able to search through a problem space to find what your UI should look like. It shouldn't arise from your brain. You should be able to turn knobs and adjust things on your system. And turning knobs is not just about a, Hey, let me look at this one button and change the color of it.
[01:01:12] You shouldn't be able to do. As a whole. So right now they're in this phase where they're exploring a generative technologies, like how do you generate the entire, uh, instead of just doing math dot random, how do you generate 40 pallets and then like just present them to the user one by one based on fact.
[01:01:32] So, uh, he's doing a bunch of that right now. And he's working with some very smart people, I think. Uh, uh
[01:01:41] swyx: huh. I think big this guy called murmurs. I think that's out of most. Yes.
[01:01:44] Sunil Pai: Norma says, oh, but also not as working, uh, not Alison is tested. Alice, Alice, I think is working with them. Uh, uh, Dean black is working with like a number of very interesting people and I'm very, they're taking their time with it, which is very strange.
[01:02:02] It's not as startup-y as I would have, like. But, oh yeah, this is the syntax a generation. I think we can literally click a button and it'll do a
[01:02:14] swyx: great, great grid. Okay. Yeah.
[01:02:20] Sunil Pai: These are all generated in your browser. I think, I don't think this is like from a database somewhere. Yeah. Why would it need to be?
[01:02:27] Um, so they're working through this. They're trying to get the fundamentals for a science that they're effectively like inventing right now. I mean, it exists for art and stuff and they're just, well, not even just art, uh, but they're trying to figure out how to productionize it for UI. And I think it's very exciting and it's very distinct and unique and I'm very, I'm keeping my eye on this all the time.
[01:02:49] I love it. I am.
[01:02:51] swyx: Yeah. Well, that's fascinating, man. Um, okay.
[01:02:59] Sunil Pai: So like I did a bunch of talking in your directions.
[01:03:02] swyx: No, no. Uh, you know, we talked about Twitter, like, yeah. I don't know. I feel, I feel hot and cold with my Twitter stuff. Uh, I started out tweeting, you know, a lot of react stuff. That's how I, I guess I initially got traction and then I did the career stuff and now I guess I'm.
[01:03:18] I don't have a particular theme. I am feeling a bit directionless as well, but I also think there's like a natural sweet spot for Twitter, where basically this is why I tell people you want, so you want enough followers so that you're not a nobody, you know, you have instant social proof, but you don't want too many that you get all the idiots coming in your, in your mentions.
[01:03:38] Right? So,
[01:03:40] Sunil Pai: uh, well, I don't think I have too many ideas. Everybody who follows me is very beautiful and smart and intelligent in the future. Uh, but I definitely, that's why I have an alt now and I do my whining and Randy's stuff there because it's very hard. I've not, I can only count on my friends to assume the nuance that I'm trying to communicate like good intent.
[01:04:19] swyx: Yes. It's, it's so helpful. Um, yeah, I mean, that's, that's, that's about what I, what I do for Twitter. I mean, I guess I might move into like founders slash VC Twitter eventually.
[01:04:31] Um, I am not, I'm not currently on the founder path, but I, I, I always keep my eye on it. I don't know what your thoughts are on like the whole, like, okay. Maybe, maybe, maybe this is a worthwhile topic. Like there's a lot of jumpsuit ecosystem. People who are, you know, our friends and now are startup founders.
[01:04:46] Right. Like, which is
[01:05:09] When art critics get together, they discuss, uh, uh, form and function and philosophy of art. But when artists get together, they discuss where to buy cheap. Turpentine always be a turpentine person. I think like, oh, like never lose sight of like reality. There will always be things to discuss. There are like metal things, which is what we're talking about right now.
[01:05:28] I think, but the true, uh, it feels, it has always felt this way to me, which is that true progress in our industry and other industries are the people who are putting actual skin in the game and actually working on it. This is, this is how I think about crypto Twitter. Also, by the way, I know a bunch of engineers who are working on crypto stuff, but refuse to talk about it on Twitter, because it's just so toxic there.
[01:05:52] And they are working on actually interesting use cases with a technology and immutable database that's spread across the world. And you're telling me, you can't think of like videos cases. There are people who have thought about it. That's that's not just an NFT. Let's get drugs delivered to our house, which by the way, you get good use cases, I guess.
[01:06:10] I don't know. So I think the same thing of like founder Twitter, there are good founders who actually like the, for example, one founder that I've been following a lot lately because I just admire the story so much is Toby Luca from Shopify. Uh, he's a hacker he's he did the video that expo that demonstrated the hydrogen oxygen reacts.
[01:06:33] React Server Components and Shopify Hydrogen/Oxygen
[01:06:33] Sunil Pai: What do you
[01:06:33] swyx: think? Or what do you think about that? I think
[01:06:36] Sunil Pai: what, what do I think about server components? Game-changer everybody should steal it.
[01:06:40] swyx: Hydrogen oxygen is a server components, uh, implementation. It is, uh, it's
[01:06:44] Sunil Pai: there, like hydrogen is basically react server components, uh, and oxygen is the run time slash network.
[01:06:52] swyx: it's there for cell phone, right?
[01:06:55] Sunil Pai: Uh, it's actually closer to workers, which suck because are already our customers at CloudFare. I'm like, why don't they use so their dealers that they want it running right next to their data. And we don't turn people's databases on the edge yet. Maybe we will. I dunno.
[01:07:10] swyx: I could change. Well, isn't that a durable objects is kind of databases on the edge.
[01:07:16] Sunil Pai: Well, it's a data store on the edge, but it's not running sql.js on it yet. Let's just never say these things. I suck at keeping secrets. Uh, the truth is that like, clearly people are talking about things internally or what happens when we can actually literally have SQL database that uses durable objects as a backing.
[01:07:34] swyx: Like there was one slide out. I already had some version of this. Right. So
[01:07:38] Sunil Pai: right. Shopify right now is just exploding this because it's not just about like running a database, it's also, you don't want to expose your database to the internet and how serverless workers like hitting it, even though that's planet skills data's plan, I think, which is they're developing my sequel, but for the serverless world.
[01:07:57] So you should be able to have like thousands of connections open under children, bring it down. Uh, another great company to keep an eye on them and super bass, super base. Just how, like good, good people in general, I spoke to a bunch of them. They donated during the India fundraiser that I was doing, and I spoke to a couple of them.
[01:08:14] Great, great team and nicely way. So it's basically react server components that have access to Shopify, API APIs. So, uh, the end, it's very strange for a company like them to go full in on, uh, basically all for technology. It's not even better yet. Like the implementation is not finished. It's not coming out with React 18.
[01:08:39] It'll be, it'll be an 18 dock techs release, but it's not coming out in November. Uh, and to see Shopify suddenly like it. It's the right choice. It feels like the technology is right. It's not something novel and people will be using it more and more. I think solid already has some kind of version of a server components because Ryan just like steals ideas wholesale.
[01:09:01] And I love that, uh, attitude. Uh, but, uh, what do I think about it? Uh, great call. What a, what a, what a way to like, make sure that you're a first adopter of something that I think is a sure thing remains to be seen.
[01:09:18] Categorical Imperatives of Web Platforms: Cloudflare vs AWS, MongoDB vs Auth0, Gatsby vs Netlify
[01:09:18] swyx: So there's a thesis that I'm drafting, which I'm not really sure about yet, but, uh, this might be useful for you because I, I spoke about this with, uh, Rita and with James Allworth right up there.
[01:09:34] Um, And so the original thesis was called fare specific, which is how is CloudFlare taking on AWS? It's CloudFare is, was the originally just the CDM they're free, CDN, whatever. Um, and, uh, obviously the, the best buy by any measure. Um, but it's eating AWS from the outside in. You first do the CDN, then you do the workers, then you do the durable objects.
[01:10:00] You, you, you kind of like reinvent every primitive from the first principles concept of like, everything is distributed. It's just a pretty amazing to watch, right? Like that's, that's kind of what, uh, what coffers, like, how do you take on a giant behemoth? Like you do the one thing that they cannot do, which is you start from like all this, all this, like distributed Stouffer's, um, And then I start thinking about things like Shopify doing, uh, hydrogen and oxygen, uh, start thinking about, uh, here's your other, your other interesting movements that I'm tracking, right?
[01:10:31] Um, Shopify, sorry, got to be making serverless functions and guests be making his own CDN. Um, so this is, this is how I blog by the way I have, I have all
[01:10:40] Sunil Pai: these insight and do the,
[01:10:43] swyx: okay. He's the only blog post ideas. That's, that's pretty good. Um, actually I want to kind of make my book, my podcast into something like, okay, basically mouth blog, like you have a, you have an idea that you're shaping and you want to blog it, but you don't have time to blog.
[01:10:56] Let's just talk it up, you know, that's kind of it that's of that idea. Yeah. So, so Netlify and guests be doing deferred generation. So these are all examples of like, uh, like a trilemma of like there's rendering. Uh, which is one group of constituencies, there's the cloud and for people, there's another group.
[01:11:13] So this is like Netlify herself up there, uh, rendered fairway. It's like, uh, next. Yes, but also gets be and react. And those guys, and in the CMS, which is like all the, like a different crew that like has, is completely above the free of all this fighting and fighting, but they always start competing with each other.
[01:11:28] Right. Um, like you're going to my turf, my turf going into your turf, that kind of thing. And, and here's, here's another one, uh, which I'm also, this is called fair inside it. Um, so I recently found out that Mongo DB that's off. Do they, do they do off
[01:11:48] Sunil Pai: by the way, that's the feature that they sell. Like, it's part of the.
[01:11:54] swyx: Oh, yeah, yeah, yeah. They, they, uh, they wrapped a Netlify identity and they made much more progress with it. Then now the fibers, uh, basically it's because they took it more seriously. Um, and obviously role does serverless functions. What the hell?
[01:12:09] Sunil Pai: Okay. All right.
[01:12:10] swyx: So here's a zero actions. These are all these are introduced. Uh, basically whenever you, you sign in, you want to execute some custom logic, they build an editor into their platform. So you can, you can write service code for office zero.
[01:12:29] Sunil Pai: Okay. That's sort of that's the land director. Okay. Sure.
[01:12:32] swyx: So, so you know, the off guys are doing compute, the storage guys are doing off, like, uh, I think there's a, there's like essentially what I'm trying to basically form a, uh, not this one. Let me see, trying to form a triangle ignore. Um, yeah, so like kind of like a triangle.
[01:12:54] Hear hear all the players and here here's where they all start out. And then, um, you know, I'll have my, I have my sort of beachhead here. right here. But in order to grow, I have to grow in one of these directions, you know? Um, and that's kinda what I'm, I'm trying to form a thesis around this. Like, so I have, this is for individual APIs and then this is for contents.
[01:13:19] Maybe this is maybe what I'm saying, that this is for apps, you know, storage competes with, uh, compute computers off. And then this is for sites where it's a content rendering and infra, something like that. You know, that that's kind of the thesis that I.
[01:13:37] Sunil Pai: That's fascinating. This is a good one. Oh God. I'm going to have to sit and think about this and get back to you.
[01:13:42] Aren't I, oh, this is really good. So there's a
[01:13:44] swyx: question of like, should they bother, right? Should Gatsby bother making serverless functions because they're basically a shitty version of Netlify functions. Right. Um, but at the same time, next year has, has functions and it's fine. Like, or API routes, right?
[01:13:59] Whatever you call it. What determines a good expansion and one determines a waste of time?
[01:14:07] Sunil Pai: Well, I assume that a bunch of these things they're doing is because their customers are asking for it, like Gatsby, for example, they're like, Hey, we just want to be able to call it an API and not deploy two separate things.
[01:14:19] So can you give us functions? And I assume it gets me, is looking for customers and getting more people to pay for that. I assume they are doing it because next year's and still are doing it and people can easily get. To versatile because we're still as functional. So they want to make sure that, so I assume that's one, one part of the story, but I keep with this whole DST thing, it feels like they're clearly moving away from their whole static asset as content story.
[01:14:46] Like my
[01:14:47] swyx: God. Yeah. It's uh, I don't know if you saw the, the announcements. I mean the, where it gets before. Okay. So I almost tweeted out the screenshot. Um, okay. Here, you know, you know how, when, uh, after the build of next year site, it'll give you a report of like, what was, uh, statically generated? What was, uh, dynamic what's the serverless functions.
[01:15:18] This looks familiar. Oh,
[01:15:23] Sunil Pai: I see. I see. Interesting. Okay. I think it's a little obvious than what's going on here.
[01:15:33] swyx: The obvious thing is like, it kind of Gatsby just capitulating and just saying like, okay, we're a Next.js clone with a GraphQL layer. Right. With a much bigger plugin ecosystem and hopefully that will, that would make a lot of people happy. Um, my it's interesting, like Gatsby, we can quite conceivably just be a Next.js layer right now. Like they don't have to build all this. It already exists with Next.js. What they should focus on is actually the graphical data layer.
[01:15:58] Right. And all their plugins.
[01:16:21] I haven't seen it yet. It's definitely going to become a thing. Oh
[01:16:25] swyx: yeah. Anyway, so, no, I don't, I don't mean to talk about Gatsby so much more just about, um, categorical imperative. Um, if you are a business in a, if you're a startup or a framework or a organization, you, uh, you have a goal, you, you achieve that goal.
[01:16:41] Then you start to look into what's what's next? Everyone does this, right? Like I'm Mongo DB. I like dominated the no SQL databases. Okay. What's next off
[01:17:03] swyx: but so far CloudFlare is the, the most concrete of those thesis thesis, which has started with the CDN and built in.
[01:17:10] Um, and so far, I don't see anyone else doing that. So, I mean, that's kind of like the framing of what I think your team is doing. Uh, it's kind of building
[01:17:22] Sunil Pai: is doing differently this year. The big deal is they want to fix their developer experience. Like they have no brand recognition. They're not even a contender when it comes.
[01:17:49] But then I actually tried using, it was like, have you tried using this? You end up with like 40 different tabs open. What is the local development experience supposed to be? And the thing about developers is not that they will give you feedback on how to make it better, right? They'll be like, fuck this.
[01:18:03] I'm going to go onto our salary, AWS and get to production tonight. I don't need, I don't need to kill. So, uh, the story is to fix the developer experience and to have a coherent story. One of the first part of that story is what we're calling CloudFlare pages, which is right now effectively, just JAMstack.
[01:18:19] But as you can imagine, it's not, it is going to become a bigger thing because it's like, oh no, you can run workers with it and have API end points. Oh, now that's durable objects. Now there's this. So I would keep an eye on that. And of course, Glen Martin is on it. Who's also my close friend, but an incredibly smart guy.
[01:18:34] Sunil Pai: Uh,
[01:19:10] swyx: Utah, so yeah. Oh, um, you know, the joke, right? That's a, the book of Mormon is actually the ECMAScript spec.
[01:19:16] Sunil Pai: That's what it is.
[01:19:17] That must be, I bet. I bet somewhere in the temple there, like classes, horses, functions and stuff like that. I don't get it. Australians and folks from UDA
[01:19:30] swyx: is what it is. Yeah. Cool. Um, okay. Well, uh, I think that's a natural ending point for conversation there. Uh, is there any, are, uh, Topics or questions?
[01:19:40] Sunil Pai: Not really.
[01:19:41] It was just so nice to see you and talk to you again. I want to hang with you in real life and grab a beer and we will do that when either I'm there or you come over and I look forward to it, man. It was so nice seeing you today. Absolutely.
[01:19:53] swyx: It's really great to catch up. Yeah, finally. Right.
[01:19:56] Sunil Pai: This was just a great experience for me.
[01:19:58] swyx: Awesome. All right. Well, have a, have a good weekend.
[01:20:02] Sunil Pai: I'll see you when I see you. Bye.