Pod Insights Syntax - Tasty Web Development Treats 918: Extreme Native Perf on the Web with Superhuman About this episode Wes and Scott talk with Loïc Houssier about how Superhuman builds lightning-fast, delightfully-designed email software. They dig into engineering philosophy, offline-first architecture, local databases, AI-powered productivity, and what it takes to create tools that people love.
Show Notes
00:00 Welcome to Syntax!
03:05 Inside Superhuman and Loïc’s role
06:49 Is Superhuman native? What’s the tech stack?
08:16 How Superhuman approaches product design and speed
12:17 Local-first architecture – Sync, storage, and performance
Realm
13:46 Vector search, AI, and privacy considerations
18:12 How the team ships fast and stays focused
21:27 Rethinking email for the future
26:54 Brought to you by Sentry.io
27:19 How calendar integration and smart features work
29:54 Where new ideas come from
31:54 Will there ever be a true dark mode?
33:02 Are people actually using keyboard shortcuts?
36:42 How shortcuts work and the role of the command palette
41:28 Engineering for speed – Costs and trade-offs
43:32 How Superhuman’s sync engine works
46:09 What code runs locally and what runs on the server?
46:51 How Superhuman handled the Google and Cloudflare outage
Hit us up on Socials!
Syntax: X Instagram Tiktok LinkedIn Threads
Wes: X Instagram Tiktok LinkedIn Threads
Scott: X Instagram Tiktok LinkedIn Threads
Randy: X Instagram YouTube Threads
Transcript Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.
0:00 Wes Bos? Everyone, before we get in, real quick, this is an episode with the head of engineering at Superhuman, and we recorded this before the news of Superhuman was acquired by Grammarly, which is absolutely wild. We did not know that this was gonna happen, and I don't think that it changes any of the conversations, because we are talking about development methodology, CSS, their engineering philosophy, and how they make everything fast, tons of local first stuff. But I just wanted to tell you, because there's probably a couple keyboard warriors already in the comments being like, "This is a sponsored episode! We would never do that to you!" It's not sponsored at all. We've been trying to get somebody from Superhuman on for months now, and it just so happened to line up. So, enjoy the episode.
0:44 Scott Tolinski? Welcome to Syntax. Today, we have a really great one. We're gonna be talking about one of the most performant and excellent apps that I use in my day-to-day, every single day. We have an amazing guest on the show with us. We have
1:00 Wes Bos? Louis Ousier. Oh, I butchered it. We cut to me, because I can at least French-Canadian accent it, but why don't you use it for us?
1:09 Speaker 02 First name was perfect. First name was perfect. Ousier. You got it right, Scott. Thanks.
1:14 Scott Tolinski? Louis Ousier. And we're gonna be talking all about how the heck this thing is so good. You know, it's got an army of superfans, and I was kind of one of those people who saw a lot of people talking about Superhuman being so good, and was like, "How can it be better than an email app that I'm using today? Email apps are all the same." And then I bit the bullet, I signed up, and I regret doing that, because it's now ruined a lot of software for me. It is so wonderful in many ways. So, we're gonna be talking all about that. Louis, welcome to the show.
1:48 Speaker 02 Thank you. Thanks for having me, guys. I mean, I listened to some of you on previous shows, and it was like, super interesting to listen to it. And I feel humbled, like, now to be, like, joining you there, and maybe, like, sharing some of my I would say perspective or like experiences if I can provide some good content. So we'll see about that.
2:08 Scott Tolinski? Yeah, because I mean, you've been you've been, you know, around a long time. You were at DocuSign for a while before, right? And that's like a massively used app. So you're no stranger to high performing applications on the web, right?
2:22 Speaker 02 Yeah, I mean, I've been around like I have some gray hair and it shows. So, so yes, I've been around and I've been both in like a massive companies earlier in my career in France, like big corp type of companies. I've been in small companies through acquisition. I joined DocuSign, so where we had like the scale up IPO trajectory just before COVID, when it was a bit, a bit crazy. So there's some fun stories about it. But then I decided to move back to like smaller companies, small startups, usually in a remote setup, remote and hybrid and distributed, which had some complexity, but also like more fun. And just joined like Superhuman six months ago. So pretty cool.
3:05 Scott Tolinski? Yeah. So, yeah, do you want to tell us a little bit about, yeah, a little bit about maybe superhuman as what it is? And then also what you're doing on Superhuman?
3:17 Speaker 02 Yeah, I mean, I was a Q. Like I was not using the product before getting in touch with the team. And I was like, how come I would pay for something that is free? Like I'm using Gmail, I'm a Gmail power user. I like all my setup is done there, like all my filters and all. So I felt productive. I felt productive. And then I tried the product during like when I was chatting with like their teams and before joining. And I was like, yeah, it's nice. It's beautiful. It's fast. Now the question is like, okay, it's better, but like, would I pay 30, 40 bucks a month to get, I would say that more the better. So I was like, nah, not sure, not sure. And we keep talking and I love the people and I love their perspective.
4:02 Speaker 02 I love like the vision they had about like how the craft should be done and how Delight can basically create a business. And after three weeks, they have this wonderful PLG motion, like product led growth motion. And they kind of like trick you into using the keyboard shortcuts. So like the education in product was really good. And after three weeks, I was like, holy cow, this is best. There's no freaking way I can go back to Gmail. And I was like, okay, I need to understand how they do this. I need to understand how they basically make me happy to pay forty bucks a month. Like for something where the competition is basically free. And and it's super interesting.
4:45 Speaker 02 So I joined six months ago there. And with the sort of like a mandate, like when you come as a CTO or as a VP edge, there's always some sort of like a mandate. And and my goal was to, and the goal is to help the team embrace like this new paradigm, like with AI. And how you bring like AI into basically a craft oriented organization, where speed of execution was maybe not the first focus, because the perceived quality was way more important. And how do you balance that more? How do you use AI basically to get that productivity gain while still maintaining the bar super high?
5:27 Speaker 02 So that's that's what I joined and that's what I'm trying to, I would say, to help the team also get better with.
5:33 Wes Bos? Are you talking like adding AI features to the product? Or you're talking engineers using AI to build the product or both?
5:41 Speaker 02 That's a great question because both. That's that's both because as like every people that are like focused on the craft, you have your own set of tools, like you have your flow as an engineer. And you tweaked it to to be like and feel great in your flow. And AI is coming here and is like kind of like redefining what even the flow is. And so that's that's one aspect. The other aspect is we are a highly qualitative product. And all the quality just depended on us, like on engineers, on quality and like all the work that you're doing. And all of a sudden with L L Ms coming, some of your features, the quality, like garbage in, garbage out, depends on the user now. How do you?
6:24 Speaker 02 Control the perception of the user so that when the result is not up to their bar, they understand that they are doing something wrong and you help them like increase or improve their prompting capabilities when needed. So like both aspects are really key for our success, given our level of expectations, like our users, they don't accept, okay.
6:46 Scott Tolinski? Yeah, they're paying a lot for the product, right?
6:48 Speaker 02 Which is probably like the biggest change for me.
6:49 Wes Bos? Well, let's talk about that. Like, when you talk about like this like top tier software that is expensive but worth it, you know, people always think about like, like Panic, big I O R, big Mac O S and I O S applications, they produce really nice stuff. And almost always, when it's this like high quality software, it's always like native. But Superhuman is not native, is it?
7:14 Speaker 02 It is.
7:15 Wes Bos? It is?
7:16 Speaker 02 It is on mobile. And for desktop, we have basically an Electron app. But like everything, like the web app and the desktop app basically share like 90% of the code. Which interesting, using like a bunch of the browser capacity for everything that is storage and everything, because we are offline first. And we might talk about it, because that was like one of the design choices, because we have this 100 millisecond rule for every interaction in the product. And you have to be like intentional about it.
7:44 Scott Tolinski? How many milliseconds?
7:46 Speaker 02 100 milliseconds. We're targeting 50. But like, an interaction that is taking more than 100 milliseconds is considered bad or poor quality.
7:56 Scott Tolinski? Mm hmm. And it shows when you use the app. I mean, especially because so much of the general workflow, for people who might not know, it is via like keyboard shortcuts. So when you're using and you really get the keyboard shortcuts, if it's not feeling that fast, I mean, you're gonna notice it right away. So you're saying the I O S app is native, but the desktop app is Electron?
8:16 Speaker 02 Yep.
8:16 Scott Tolinski? And what's the UI done with?
8:19 Speaker 02 React.
8:20 Scott Tolinski? React?
8:21 Speaker 02 It's basically mostly React. Interestingly, we don't use just yet any design system. So everything is bespoke, mostly because we have an emphasis on local optimum when it comes to the user interaction. You want everything to feel great. And to do that, sometimes with design system. You're somewhat limited. So for quite a long time, like everything was bespoke in every single pieces of the of the product.
8:52 Scott Tolinski? Wow.
8:53 Speaker 02 Yeah, that's that's intense.
8:56 Scott Tolinski? And how do you mean consistency then if you have no design system?
9:00 Speaker 02 You need a bunch of senior people. You need to hire engineers that are product slash design engineers. So you need people that just have an eye for something that feels off and that can like make it right. So we are very intentional in the people that we hire. We don't hire like software engineers. I mean, it's kind of like a caricature to say that we hire like product engineers. Everyone is saying they hire like product engineers. But a good portion of our interview process is focused on, especially for the front end people, is like focused on your ability to to craft like a great user experience. So we just don't want like people that are great at like whatever the frameworks or all of that, but like very CSS oriented, very like detail oriented. So yeah, you you
9:45 Speaker 02 have to you have to be intentional when you hire. And we have a lot of like internal trainings, I would say, coming back and recurring about like all the engineering principles that we need to apply in the in the way we work.
9:57 Scott Tolinski? Man, that's wild to know. It's interesting, too, especially because I would have thought, you know, some of the other apps that we've used that have like an electron based desktop app that you would use also something similar for their their mobile app. And the mobile app does feel like it could be very much the same UI code base or anything like that. That's interesting to know that it's it's different. Also, like, I have found that Superhuman looks different than many apps that I use. It doesn't have that same kind of like generic looking style to it. So I would imagine that that designer kind of product developer fits with that as well. And those type of product developers, they're they're focused
10:41 Scott Tolinski? specifically on speed as well? Or is the speed aspect of it like an entirely separate team?
10:47 Speaker 02 No, no, no. Speed is not a team. Speed is like, I would say, part of the quality process. So everyone is looking at speed as something that is mandatory as part of the user experience. So it's not an afterthought. And Mike, our chief architect, would basically say that speed needs to be like the first thing you think about, rather than the last thing you think about. Like, if you think about like building a feature, you need to understand how this would be fast while you're designing the application. And that's why we moved to like offline first. It's not because we wanted to support offline. It's because the number one culprit when it comes to like slowness is the network. Especially when you're talking emails and you're on your mobile phone,
11:31 Speaker 02 you have a shitty network, like you don't want this to be felt on the end user's side of things. So we do a lot locally. We do a lot locally. So it might surprise you, but for example, there's a database on your iPhone mobile app. We are not getting the data live. We basically like sync the last 30 days and moving onwards. So we have a huge footprint on your phone so that everything like search can work offline and be like crazy fast. We don't rely on the providers, like Gmail or Outlook, for example, only for stuff that are like back in history. But like the last days, it will feel like instantaneous because it's local.
12:10 Scott Tolinski? Yeah, I I've been getting into a lot of local data stuff myself and sync engines and CRDTs and those types of things. Can you, can you maybe like talk a little bit about the tech behind what local database you are using? And how the sync process works? Yeah.
12:25 Speaker 02 So take everything with a grain of salt. I'm the VP edge. I'm not the engineer, neither am I the chief architect. But so right now, we are, on mobile, we are using Realm, but like we are moving away and we are like at the decision process to understand which tech we will use globally. It's kind of like a SQLite approach.
12:44 Scott Tolinski? I remember Realm, yeah.
12:46 Speaker 02 On the browser side, same. So we use an extension. So we have an extension that is, that is used both by the Electron app and the mobile, the web experience.
12:57 Speaker 02 And we store like we use the browser APIs and everything to basically store locally. And we use a submission of a SQL light as well to store stuff. We store basically emails, we store your events, agenda, the calendaring, like everything like this is locally stored. But it's stored like as we store the emails, but like we have, of course, some backend capacity where we store more stuff. More history, for example, but also like everything that is needed for RAG systems. Where we have like a dedicated vector database, which is not local. So we don't do. AI locally on device just yet. So when you're offline.
13:37 Speaker 02 You don't have AI and that's that's a design choice for now. Maybe it will change eventually, but right now we are sticking with it.
13:46 Wes Bos? And why, why is that? Is because like. I've I've been talking to a lot of people who are starting to do, especially as the. Processors get faster in these phones and and we have web GPU and whatnot. Where you're able to do embeddings like fairly fast on device. What's the choice for not doing that vector search and embeddings on device right now?
14:11 Speaker 02 Because usually we do the vector search to do something. It's like the vector search is only one piece of the feature that you want to to provide. So I will give you some example, but we do some drafts creation for you. So funny enough, Superhuman moved from being the fastest email app to the most productive email app. And part of the productivity is creating for you drafts that are ready to be sent. Like you don't even have to think about it. Someone asks you a question, we look at your, I would say, capacity and we provide some answers that. Hopefully are ninety percent of the time right and you just have to hit send, boom, you're done. Typical use case as well is like you ask a question to someone.
14:54 Speaker 02 And you have like some sort of like a reminder that is ready. So that you can nudge people pretty fast without thinking about it. And to do this, we need some tones, we need some information about like the past emails that you sent to those people. And you need to process the email and everything that is generative on device is still relatively slow. Right now, so the experience wouldn't be great. So the question is like, are we okay with somewhat? Like a slow, like generative feature that would make you wait, or we just don't allow it at all. And coming back to our principles, okay, is not good enough.
15:35 Speaker 02 So this is typically the case where devices and models like running on device are not good enough just yet for our quality bar. So, and that stuff, because we, we could, I would say, implement something, uh, but like our user base, like they need something that is clean and well done. So you have always to ask your question, is it good enough for them, given the past of quality and delight that we are providing for every feature?
16:03 Scott Tolinski? Yeah, and, and as far as AI features go, like, and you mentioned, you know, speed being a highest priority. And I know that like AI isn't always seen as being like the fastest thing, you know, you're waiting for a response streaming in, et cetera. But like, what kind of AI features does superhuman implement or can they implement that don't need to feel like instantaneously fast, like async kind of features?
16:31 Speaker 02 Yeah, so, so that's one point, like we do it, I think, so that it's ready when, I would say, you open the, uh, I would say, your superhuman app. So I was mentioning those drafts or like the classification of your email. Everything is happening on our backend. And when we open, you open the app, all your emails are filtered automatically and the draft are created. So we try to optimize for those processing that can happen basically before you open the, uh, the app. Most of the features are like this, we still have some features like augmented writing, like you want to kind of like rewrite, uh, like an email.
17:10 Speaker 02 Because you don't feel right about it, like you don't want to go to chat GPT and everything, so like come on Jay and improve writing, boom. And you have like this being processed and then it's kind of like using some UX tweaks to make it feel that it's instantaneous. So you cannot just wait with like a spinner saying like, hey, uh, reviewing your comments. So for example, like if you compare to Notion AI, if you're using Notion AI to rewrite some content. It will feel relatively slow. They will basically strike through stuff. So they add some complexity to the user experience, showing they do something. And eventually you accept the modification. We don't do that.
17:51 Speaker 02 Like, we basically like write and basically like the streaming APIs are good enough to to make things like feel fast. Basically, it's writing faster than you can read. So you don't need those extra stuff and those extra things that Notion is doing, for example. So we try to get rid of the clutter so that the experience feels like really fast.
18:13 Scott Tolinski? Wow. Yeah, I I I just popped open my little latest updates thing because it feels like there's always like new updates. So you all are shipping a lot of things really fast, which on the surface level, when you you look at Superhuman and UI is not like changing substantially all the time or doesn't feel like a different app. But here's just like four new features I didn't even know about. So like, how is your team able to ship so much so quickly?
18:39 Speaker 02 We are a startup and we work hard. So that's, I mean, honestly, that's that's a part of it. Like, we work hard and I mean, I have freaking good engineers, to be fair. Most of the team, like the average tenure is probably around like three years on average with people like two, like six to seven years. So we have a bunch of seniority and knowledge that is already in the team, which helps. We are pretty intentional and we try to prepare things so that when we work on things, on on features, they are baked already to some extent so that we can keep the flow going. We put in place an organization that is pretty domain centric.
19:22 Speaker 02 So we have three big domains in Superhuman. The calendar experience, which is a new one. The triage experience, so we have a team dedicated on the like the focus that you need to get when you join. Like, for example, you go to Gmail, you have a long list of emails and it's kind of a mess. You have the cognitive load, you need to go, then you need to go through it. So we have a bunch of features like splits or like automatic archiving of like the clutter so that you can just focus on the right thing. We have a team dedicated to it and we have a team dedicated with. Engaging with your emails where this forwarding, commenting, responding, basically. So those are the three teams. And when you have teams that are like laser focused on their persona,
20:05 Speaker 02 the way they use it, the complaint that they have, it's easier to get velocity because you know your stack, you know your code base, you know the complaints, you know like the how people are using it. It makes things, I would say, faster. And to be fair, AI is helping a lot. Anecdotal feedback, like it's based on like self, like we did a survey and people were like providing like their own appreciation of like the productivity gain they get from AI. The answers were roughly between 20 and 30% productivity gain. So it's not doubling our capacity just yet. But it's significant. It's significant.
20:46 Wes Bos? I want to talk a little bit about just like email overload and being able to find what is important. Because this is something I think a lot about right now is that tools like Superhuman allow you to just send a whole lot more email than you normally would because it it'll you said it'll generate drafts, right? And so there's there's like this new area of it's not necessarily spam, but it's just like this like generated cold outreach that I get. I probably got 15 already this morning and I'm I'm kind of at my wits end with my email right now. I'm just I'm sick of it because it's it's so hard to find what is important. So I'm curious, like, like, do you have any thoughts about
21:30 Wes Bos? how email in the future is going to look, both from sending lots more because it's good for business, but then on the inbound, having to be offensive and try filter for filter all this crap out?
21:44 Speaker 02 No, no, no. I mean, this is a this is a good point. Like the one thing that was interesting, like before talking about the future of emails, like if you look back, what was the past of email, like everyone is like Our customer base, like ninety percent of our users are using Gmail. And the enterprise customers are on Outlook roughly. So that's kind of like the ninety ten is like the the way we think about it. Like, think about Gmail. Gmail has like no interest in getting rid of the spam. Or like the marketing spam. Because of course Gmail is a marketing engine, like Google is a marketing engine. And that's why I would say they are kind of like bread and butter. So they cannot just help you only receive what makes sense for you. Like, they need that volume of data.
22:27 Speaker 02 We don't care, because that's not our pricing strategy. That's not the way we make money. And that's coming like today, like this is what you have right now. We, especially using AI, it's so much easier to detect whether an email is a pitch, like your company reaching out to you to sell you like the new C I C D tool that they've been working, they're out of Y C, whatever. And they want to pitch you. Or like those marketing emails because you use GitHub or GitLab on a daily basis. They send you like some marketing content associated with like the new features and all of that. Yeah, pardon my French. That close, that close. But we all get like all those emails.
23:10 Speaker 02 And honestly, to be efficient, and especially when you're a C label, there are some stuff that are really crazy important and that clutter that you would love to get rid of. And now we get rid of it. Like our users, like everything that is marked as a pitch or marked as a marketing email, you can select auto archive it. So you don't even see it. It's not even showing up in your in your email. And then you can do some pretty like a nice thing. Then you can prompt the classification. You can say like, oh, you know what? Everything that is like a podcast related or everything that is coming from my board.
23:47 Speaker 02 Like as wide as this and my board are like people like this and this and this. So you can write prompts now in Superhuman to classify your email.
23:55 Wes Bos? Ah, that's what I want.
23:57 Speaker 02 So that you can say, this is important, this is important. And and it's working great, it's working great. So that's that's kind of like one way, but the same like you can expect those marketing people. I would say to build some outreach like hidden text that we try to mess up with the L L M. That is doing the classification like it would be kind of like the the typical spam, anti spam. Uh, war and it's just like the the weapons are different right now, but it it would be.
24:24 Wes Bos? Yeah, like we get people now, they will ingest the transcript from the last six episodes. And then they will say something like, oh, I also live in Denver to Scott. And they'll they'll try to make it as a personal, you know, or they'll ingest like three episodes and ten of my last Instagram stories and like they'll it's it's awful. How personal they're trying to make it, you know, and now I'm trying to write. I use I use Missive and I'm trying to write these like A I rules that I'm on the offense of it. And it's just a waste of my time.
24:57 Speaker 02 Give it a try, give it a try, let me, let me know, let me know how it works.
25:01 Scott Tolinski? I have found the the labels, the auto labels to be super useful. I I I posted one in our show notes so Wes could see just look at labeling a pitch that somebody had uh sent me and it clearly labeled this pitch. I didn't know you could set up your own custom auto labels, though I haven't done it yet.
25:19 Wes Bos? And search with A I as well, you know, show me the emails that I don't know, six months ago, somebody emailed me requesting a print of a sticker, you know? I I hate having to like try to hit all the keywords properly.
25:33 Speaker 02 I use I use this A I search like almost on a daily basis. Uh, because like a lot of my knowledge and content is in my email. The the last example, over the weekend, I was working on my citizenship uh application. And for the people that are going through this, like in the U S, you need to go through the last five years of travels uh outside of the U S. And I was like, oh, damn, I remember like I went to Japan, I don't remember the dates. Hey, can you find the tickets for my trip in Japan that I believe happened in around that timeline? Boom, boom, boom. And I get those emails and I have the dates and done. Like I was done with it. So this is like silly examples, but AI search is definitely something that is, I would say, crazy good.
26:18 Speaker 02 The next piece that will be like very interesting and probably will ship that early Q3 is like someone asking like a friend of you already in your like, I would say indoor list of contact and all of that, asking for like a slot. And like, hey, can we meet somewhere next week, please? We have access to your calendar. We have access to your availability. So we can auto draft and answer and propose like a slot or a set of slots that are working for you. And then you just have to basically hit send and boom, you're done. So a lot of stuff like this now are possible because of LLMs that used to be just impossible to do before.
26:54 Scott Tolinski? And if you want to see all of the errors in your application, you'll want to check out Sentri. It's Yeah, I was wondering about, I even had on my question list, given that there is like, there is a calendar integration and it does seem to be getting more and more. I haven't found it to be able to actually replace my phone calendar just yet, but like the fact that you have so much information. One of the things Superhuman does already really well is contextually aware things. So Wes, you might not know about this, but somebody in an email will say like currently, this is the way they currently work. So they'll be like, what do you think about Wednesday? When you open that email, it will actually pop open the calendar and show you Wednesday already. So you can see exactly what your calendar is like on Wednesday. And I use that obviously all the time. How did those type of features work?
27:39 Scott Tolinski? Is it just simply like looking for certain strings and then choosing what to do with them?
27:45 Speaker 02 Yeah, totally. This is the type of features that might feel overkill for like a typical B2B SaaS company. They would be like, nah, let's not do like a feature like this. People can click on the calendar and go to, like they can do it. There's a way. And this is where like we have a CEO founder that is kind of like an art director to some extent and has like a product test and understand what are the features that will create delight in the end user. And this is typically one. You open this calendar and boom, we can, we see there's a date.
28:21 Speaker 02 And it's not that complex to find a date like in a text and boom, you open like the, I would say the right sidebar and it's probably working like ninety five percent of the time, but it's just delightful because you you feel you're in the context. And this is something that is really important. Like you were mentioning the fact that we have a lot of features, but we are not cluttering the experience like the experience still looks sleek. There's not a lot of buttons like we are not adding like a new button. Like every time there's a new feature, we want this to be discoverable, but hidden, I would say first. So you need to kind of like learn to use them so that the experience always feel sleek. And like you're always like in the focus mode. That's something that is part of our world, like design principles.
29:06 Scott Tolinski? That parsing of the date, is that done like programmatically? Or is that done with like a model?
29:14 Speaker 02 No, it's not a model. It's programmatically done. So this is typically the a place where like with the right reg apps and everything, like token detection and all of that, there's a bunch of libraries that can do this. Honestly, I would need to check like which one we're using. And I can like provide that after the show. No, clearly this is this is doable without without L L M's.
29:34 Scott Tolinski? Yeah, which is so nice because yeah, that allows it to be fast. It allows it to just happen. But like as you it's funny that you said delight, obviously, because that seems like really what it is. But the first time it happened to me, that was the discovery moment. When you open an email, all of a sudden my calendar is open and I'm like, whoa. Oh, this is a this is it's a delightful experience. I get I know you've only been there six months, but how how do those types of experiences even like? How do those types of ideas even come to people? Because
30:06 Speaker 02 No, it comes down it comes down to the product taste. So I think product taste is something that you need someone to teach you. To some extent and Rahul is probably the person that has the most impressive product taste. So like he's doing every single product reviews and design reviews in the company. Kind of like the Brian Shevsky of the world at MD and D. Like it's kind of like a version of that. I don't want to enter like the founder mode like discussion like it's a really great founder that we have and we love him. But he has that. He always asks those questions. He always tries to understand like how far are you from getting to that wow moment? Like how many steps to hit?
30:49 Speaker 02 And I've been here like only six months and now I'm asking those questions. Now I'm looking at the demo that my team are doing and like feels clunky, doesn't feel superhuman, feels too complex. We need to do something like this. So it's it's kind of like trained to some extent by being exposed to Rahul that has that as a, I don't know if he learned it. The gift. But it's definitely, it's a gift. I think it is. I think it is. But like the same way, I would say when you work with someone that is just great and you get mentored and you see how they operate, you start to build a muscle and you start to build, I would say that taste. It's kind of like drinking wine. Like you drink like shitty wine at first, you start to have like a better taste about
31:34 Speaker 02 it and you meet someone that is just like great at explaining why it feels better. And you're like, ah, you're right. I see the fruit now. I or like I can feel the hook. Now you tell it and then you build on it and you become like more expert. But you need that first person to sort of like open your mind about those like different aspects and that's a big part of it.
31:54 Scott Tolinski? Here's a product question. Will there ever be a true dark mode? The dark mode, Wes, is one of the disappointingly light gray dark modes in my mind.
32:04 Speaker 02 So we call it carbon. We call it carbon. But like I'm a dark mode user and I'm complaining every freaking day.
32:13 Scott Tolinski? I think that's the only thing I've ever submitted to feedback in the app was, can you get real dark mode here?
32:18 Speaker 02 No, no, no. So that's, that's very true. Like it was like mostly like, uh, I would say made for like white and clear like, uh, mode for like a long time. And then we basically implemented a dark mode because of the request, but we didn't do it everywhere. And as we are not heavy on design systems, making those changes, like that was a text time. And we are, we want to optimize locally. So like every single interaction needs to be thought through with dark mode now, uh, like in, in mind. But no, I'm with you. I'm with you. And you know what? Even worse. I'm an Android users, uh, user and Android is lagging behind iOS. So like for my team, I complain about like the two things that are lagging. So it's, it's really bad. It's really bad.
33:02 Scott Tolinski? I, I've noticed that, you know, the two main ways that you interact with superhuman are through the command palette and keyboard shortcuts, which are Wes, most of the keyboard shortcuts are just like one character. So like, you know, E or enter, you know, those types of things, right? Like I hit E and it immediately marks an email as done. And you kind of J and K, you kind of really get used to using the keyboard to navigate this thing. To me, these all feel like very, um, power user-y type of even like, um, like programmer type of ways of interacting with an app. Do you see that like normal people are able to utilize these features effectively?
33:44 Speaker 02 This is a great question. And this is a question we have internally because our initial ICP, uh, so like the ideal customer profile, uh, was the like barrier or by extension, uh, like tech industry founder. So usually young, usually tech savvy and highly productive and wanted to like, uh, I would say gain some time. So those people were the right, the right target to get like keyboard shortcuts and everything. They were like looking for it and even like expecting it. It's more clicky, uh, now as we, we were like more and more, um, present on the market. So basically all startups were, yeah. Are using like superhuman. Those startups got acquired by by bigger groups and.
34:27 Speaker 02 You see those like C level of startups being acquired complaining, I would say, uh. To the bigger groups saying like, no, no, no, no, we want to keep it. So now we have like big companies basically saying to us. Hey guys, we need to standardize and superhuman. But like not all the population is. Tech savvy like this, uh, what do you propose? And and that's something that um. That is tricky uh to some extent, but and this is. Where the I was mentioning in the beginning, like the PLG motion and the education. Motion is something that uh is very important for us and and our success because. Even the people that are not that tech savvy, if you go through. The basically those like in product uh education moments.
35:12 Speaker 02 You get used to it and you're like, oh damn, it's easy like enter. Anyone can do enter and then you start with enter. And then maybe you start with like the E, which is archiving things like, oh. Archiving works well and maybe you use that one. And then you learn about the second one, maybe like weeks later. So. The curve of adoption is might be might be like a bit slower for the non tech savvy people. But eventually like the the speed at which. You operate with those keyboard shortcuts is crazy and it's interesting because there's a. A lot of discussion about like the change of user interaction with devices and voice. Becoming more and more important. I still believe that keyboard shortcuts like. Would be like so much easier than voice like for like doing the archive for a while. Sure.
35:56 Speaker 02 Like and and things like this. So. So yes, we are talking about it. Not yet to the point where we want to. Add like a lot of like ribbon like. I can't imagine having a ribbon with like a different actions for people that are very outlooky. Uh, to to to some extent can feel that, oh yeah, that's the same or close enough. So, but that's the question we have.
36:17 Scott Tolinski? Yeah, I do. I would see that the like more like voice type of thing would be very useful on on mobile where you don't have access to that, right? Although you do have swipes and and those types of things that that tend to fill that space. The keyboard shortcuts I use the most are archive, the up and down to traverse them, and then the H to reschedule an email. So, yeah, that one to me, you know, Wes, you really liked sentries date selector, where you could just like kind of type whatever you want in. This thing is like next level. Yeah, this thing is next level like that. You could say, you know, three days, 8 a.m., next week, sometime, whatever. I mean, you can, you can type in whatever the heck you
37:00 Scott Tolinski? want in this thing, and it it manages to get your next Thursday, whatever, and it just works. How does something like that work?
37:07 Speaker 02 So like the H one, for example, that's the one that I use the most. So H to basically like make your email disappear and only come back when you believe it will make sense. Good example is like, oh, I don't want to treat that today. I want to focus on this tomorrow. So HT, enter, boom, your email is gone, and T for tomorrow. That's probably the one that I use the most, like next week, next month, or like I would say today or tomorrow at 2 p.m. So that's, I would say, the type of things that I use a lot because I want to reach inbox zero. Wes, like one thing like that is really key for our users is reaching inbox zero, because you just get rid of the cognitive load. And when you reach that point where you have nothing, like, oh, everything has been treated, and
37:52 Speaker 02 you don't have to mark stuff and read because you you want to get to it at some point and everything, which is like the typical flow for like Gmail or Outlook, it feels so great. And to do this, basically, you postpone stuff for to when it makes sense. So usually like, oh, yeah, I know that I want to tweet like those type of emails on Friday. HF, and Friday, boom, it's done. And one second, and it's ready. So you can go through like a hundred of emails a day, which is roughly said what I receive this way into like the like 15 minutes at the end of the day. It's pretty awesome.
38:28 Scott Tolinski? I don't want the audience to think this episode is a
38:30 Speaker 02 No, yeah, sorry.
38:32 Scott Tolinski? endorsement. No, no, you're you're fine. But I have found personally, because I, when I get excited about things, folks, I tend to like turn into a salesman for it. So I apologize for that.
38:45 Scott Tolinski? But I have found that I hit inbox zero just about every morning with this thing. And it's turned what I do in my daily ritual for email is that I tend to not look at email more than twice a day. I'll open my email in the morning. I'll either get rid of things I need to do in another day or whatever, mark them as done. Take care of everything I need to do, get to inbox zero and then close out the email for the day. And I have found another system that allows me to do this as quickly, primarily because just yeah. HT, HT for the stuff I don't want to do today, I can take care of it tomorrow. And as somebody with like attention issues, treating your inbox as your to do list ends up becoming
39:27 Scott Tolinski? just like a really pleasant experience overall, which I tend to hate email. So yeah.
39:32 Speaker 02 Everyone hates email. Everyone hates email. So that's a good way. And interestingly, and maybe like the last plug I would say for Superhuman. But what we see is in the, I would say our customer base is like they save like four hours a week. So it doesn't apply to, I would say if you're a typical engineer with like in a big team. So you don't manage third parties, you don't manage a board, you don't manage like a lot of people. You will be probably living your life in Slack way more and hopefully in Slack and not in Teams or like any other one. So that might not apply, but when you start to be like in leadership roles or your role, maybe you're like a freelancer and you spend a lot of time in emails night and day.
40:14 Speaker 02 So like four hours a week, this is typically what people save and they reply usually like 48 hours faster than before. So that's the type of like metric that we have and some people are crazy. I would say that there's like a wide range of people within that, but like that's the average.
40:33 Scott Tolinski? I think one of the big takeaways I have from Superhuman in general is that like. You all have made a technical choice based on, you know, company-wide philosophies of it being fast and quality, right? And you made your technical choices to align with that, to use local data. And through that, you've created a product that delights people, it is fast as hell, and overall, like, really feels good to use. And I've been talking a lot about local data on this show, and I think that's like an important thing, is that oftentimes we think about, "Well, I'll use local data." But like, you all started with the experience, what experience needs to be there, and then
41:17 Scott Tolinski? that probably determined to use the local data and then the keyboard shortcuts and all these things. And it really, really, I think, worked out in a great way for you all, so...
41:28 Speaker 02 But it has a cost. Like, one thing that is really important for people to understand, like, if you're starting your company and this has like a freaking cost. Like, you need to basically understand that delight and like that speed is something that is an investment. We are native, most like for iOS and Android for that purpose. Being native has a cost. We could have used Flutter or whatever, like, or the framework to basically build ones and use everywhere. We decided to be like a very intentional... We, I would say, are local first, so you were talking about like sync engines. Like, syncing is probably like the number one, not pale, but like channels that we have that is tricky, because you have network, no network, so you have some sort of like
42:11 Speaker 02 a pub-sub, like, mechanism on every single device in every single platform that we are using. But that's typically business logic that you would typically provide and only once in your backend, or your backend for frontend, like whatever the architecture that you're following. But typically, in a B2B SaaS company that doesn't care about speed, you would manage that there and manage like the pub-sub, like on the backend. But we want to be offline first, so you need to have that approach on the local device. It has a cost. Every time we want to do some modification about syncing, so you were talking about calendar, number one, I would say, feature of all our users, like there... Bring us calendar the same way you save email and we're working on it.
42:53 Speaker 02 But it means that on every single device, we need to kind of like adapt our like syncing engine for managing like the different events to say work with the API of Outlook and Gmail. So it has a cost. It has a cost. But we do believe that this cost is worth it because, I mean, we have like our users, they love the product and they don't go elsewhere because the experience is so great. So again, it's different business strategy. The business strategy here was a bit like Apple. We might not be the first. Well, we'll be a close second, but we will be the best. So it has a cost. It has a cost.
43:32 Scott Tolinski? Your sync engine, is that all custom stuff or is that being done through Realm?
43:36 Speaker 02 So the the the storage capacity is like through Realm, but like we have this concept of modifiers. So it's roughly said a pub sub mechanism. So when we want to archive an email, so we have a modifier that will basically log like every single stuff that you're doing on that email, archive, unarchive, archive, unarchive, if you want to play with it. On top of it, we have like the undo capacity, like we give you like 10 seconds because oh shoot, I made a typo, a typo and I didn't use AI to rewrite my email. No, no, no, no, no. So that's you don't send the email. So like you need to have like basically like a stack of actions that you can go through and we want to implement that offline. So it needs to be present on every single device.
44:20 Speaker 02 So we call that modifier and it's roughly said like a stack of actions that will be synced. I would say as soon as you have the network, it will be synced with the backend that implements also those modifiers and will then implement the changes in Gmail or in Outlook. Well, it's it needs to happen.
44:39 Scott Tolinski? Awesome. Yeah, this is so fascinating as a one, a fan of local data stuff and two, as a big time user of this app, I sincerely appreciate all the time you've taken to to break this down for us.
44:49 Speaker 02 Yeah, and maybe like one one interesting things like our front end engineers and like JavaScript engineers, we have two kinds of engineers, like people working on the UX and the people working on what we call the background, which is kind of like the backend of the local stacks. And we have like a platform team that is usually treating those like meaty gritty.
45:12 Speaker 02 Almost like backend problems on the device. Because it's, I mean, you face like the same race conditions, the same like
45:21 Wes Bos? Yeah.
45:21 Speaker 02 eventual consistency of shit that you, sorry, this time I said it.
45:25 Wes Bos? Yes.
45:25 Speaker 02 That you have typically on the backend, which is not typically common for like JavaScript users, unless you're using Node on the backend, but we're a go shop. Sick. Wes, do you have anything else you want to get to here?
45:37 Wes Bos? No, I think that's good. I was specifically going to ask about that of like, like the logic layer and the UI layer, but you pretty much just answered it. So that's really interesting when it's, when most of the code runs on the device, there's not really a backend, right? So that was really neat to hear.
45:57 Speaker 02 No, the backend, our backend is lightweight to some extent.
46:02 Wes Bos? Yeah.
46:02 Speaker 02 Compared to the number of like emails we are managing. But like the backend on the device is pretty extensive.
46:09 Wes Bos? What code do you run on your servers?
46:11 Speaker 02 So it's go, we're a GCP shop. Everything is running on GCP. We have a like small cluster running on a GKE, which is like the managed Kubernetes platform. We use Postgres. There's some, I would say, Redis here and there for caching. And the Pup Sub is like the, I would say the Pup Sub from Google there. So pretty, pretty lightweight. I think we have something like four or five, I wouldn't say like services, but like we have one monolith. So it's mostly like a moduli architecture to some extent on the backend.
46:51 Wes Bos? So here, here's a question. Last week, Google went down for a couple hours and it took CloudFlare down with it as well.
46:59 Speaker 02 Correct.
46:59 Wes Bos? Did you have downtime and what did that look like for your app?
47:03 Speaker 02 Yeah, so this is a great question. So we are offline first.
47:08 Wes Bos? Yeah.
47:08 Speaker 02 So which means that during that outage, the only problem that you had was that we couldn't receive new emails because Google was basically down. Gmail was downish, so receiving new emails was not possible. But we stack everything. So we had to replay transparent for our user. They send emails, like everything that is email answering, email sending, and all of that, totally transparent because it's managed locally. And then it's like going to our backend when they have the network. And the backend is kind of like replaying everything. So we had like a long stack of emails that were waiting for the outage to be done. And tack, tack, tack, tack, tack, we basically managed that. We had some issues because of the attachment.
47:51 Speaker 02 And there were some issues with attachment because like we had a one bug. So it was like an incident, not an incident, like it was an incident internally, but like not seen by our end users.
48:01 Wes Bos? Wow.
48:02 Speaker 02 So we're used to it. Like Outlook, like their Graph API tends to break every new version because they're, I don't know what they're doing, but like they break stuff relatively often. But it's mostly not seen on our side. So the only impact is like you will have like a small notification, hey, Gmail is facing some issues. But that's about it. That's about it.
48:25 Wes Bos? All right. I think that's all for today. Anything else you want to add before we wrap this sucker up?
48:31 Speaker 02 No, no. Thanks for having me. That was super cool, guys. Great discussions. And always great to talk about like tech stuff.
48:37 Wes Bos? Yeah, I appreciate it. Well, thank you so much for coming on and appreciate all your time.
48:42 Speaker 02 Thanks, guys.
48:42 Wes Bos? All right.
48:43 Speaker 02 Peace. Thank you.
Previous 917: AI Tools You Should Know Next 919: Better Auth with Better Auth Page updated November 1, 2026 · transcript and summary are machine-generated