497: Axiom with Seif Lotfy
Giant Robots Smashing Into Other Giant Robots - A podcast by thoughtbot - Giovedì
Victoria is joined by guest co-host Joe Ferris, CTO at thoughtbot, and Seif Lotfy, the CTO and Co-Founder of Axiom. Seif discusses the journey, challenges, and strategies behind his data analytics and observability platform. Seif, who has a background in robotics and was a 2008 Sony AIBO robotic soccer world champion, shares that Axiom pivoted from being a Datadog competitor to focusing on logs and event data. The company even built its own logs database to provide a cost-effective solution for large-scale analytics. Seif is driven by his passion for his team and the invaluable feedback from the community, emphasizing that sales validate the effectiveness of a product. The conversation also delves into Axiom's shift in focus towards developers to address their need for better and more affordable observability tools. On the business front, Seif reveals the company's challenges in scaling across multiple domains without compromising its core offerings. He discusses the importance of internal values like moving with urgency and high velocity to guide the company's future. Furthermore, he touches on the challenges and strategies of open-sourcing projects and advises avoiding platforms like Reddit and Hacker News to maintain focus. Axiom Follow Axiom on LinkedIn, X, GitHub, or Discord. Follow Seif Lotfy on LinkedIn or X. Visit his website at seif.codes. Follow thoughtbot on X or LinkedIn. Become a Sponsor of Giant Robots! Transcript: VICTORIA: This is the Giant Robots Smashing Into Other Giant Robots Podcast, where we explore the design, development, and business of great products. I'm your host, Victoria Guido, and with me today is Seif Lotfy, CTO and Co-Founder of Axiom, the best home for your event data. Seif, thank you for joining me. SEIF: Hey, everybody. Thanks for having me. This is awesome. I love the name of the podcast, given that I used to compete in robotics. VICTORIA: What? All right, we're going to have to talk about that. And I also want to introduce a guest co-host today. Since we're talking about cloud, and observability, and data, I invited Joe Ferris, thoughtbot CTO and Director of Development of our platform engineering team, Mission Control. Welcome, Joe. How are you? JOE: Good, thanks. Good to be back again. VICTORIA: Okay. I am excited to talk to you all about observability. But I need to go back to Seif's comment on competing with robots. Can you tell me a little bit more about what robots you've built in the past? SEIF: I didn't build robots; I used to program them. Remember the Sony AIBOs, where Sony made these dog robots? And we would make them compete. There was an international competition where we made them play soccer, and they had to be completely autonomous. They only communicate via Bluetooth or via wireless protocols. And you only have the camera as your sensor as well as...a chest sensor throws the ball near you, and then yeah, you make them play football against each other, four versus four with a goalkeeper and everything. Just look it up: RoboCup AIBO. Look it up on YouTube. And I...2008 world champion with the German team. VICTORIA: That sounds incredible. What kind of crowds are you drawing out for a robot soccer match? Is that a lot of people involved with that? SEIF: You would be surprised how big the RoboCup competition is. It's ridiculous. VICTORIA: I want to go. I'm ready. I want to, like, I'll look it up and find out when the next one is. SEIF: No more Sony robots but other robots. Now, there's two-legged robots. So, they make them play as two-legged robots, much slower than four-legged robots, but works. VICTORIA: Wait. So, the robots you were playing soccer with had four legs they were running around on? SEIF: Yeah, they were dogs [laughter]. VICTORIA: That's awesome. SEIF: We all get the same robot. It's just a competition on software, right? On a software level. And some other competitions within the RoboCup actually use...you build your own robot and stuff like that. But this one was...it's called the Standard League, where we all have a robot, and we have to program it. JOE: And the standard robot was a dog. SEIF: Yeah, I think back then...we're talking...it's been a long time. I think it started in 2001 or something. I think the competition started in 2001 or 2002. And I compete from 2006 to 2008. Robots back then were just, you know, simple. VICTORIA: Robots today are way too complicated [laughs]. SEIF: Even AI is more complicated. VICTORIA: That's right. Yeah, everything has gotten a lot more complicated [laughs]. I'm so curious how you went from being a world-champion robot dog soccer player [laughs] programmer [laughs] to where you are today with Axiom. Can you tell me a little bit more about your journey? SEIF: The journey is interesting because it came from open source. I used to do open source on the side a lot–part of the GNOME Project. That's where I met Neil and the rest of my team, Mikkel Kamstrup, the whole crowd, basically. We worked on GNOME. We worked on Ubuntu. Like, most of them were working professionally on it. I was working for another company, but we worked on the same project. We ended up at Xamarin, which was bought by Microsoft. And then we ended up doing Axiom. But we've been around each other professionally since 2009, most of us. It's like a little family. But how we ended up exactly in observability, I think it's just trying to fix pain points in my life. VICTORIA: Yeah, I was reading through the docs on Axiom. And there's an interesting point you make about organizations having to choose between how much data they have and how much they want to spend on it. So, maybe you can tell me a little bit more about that pain point and what you really found in the early stages that you wanted to solve. SEIF: So, the early stages of what we wanted to solve we were mainly dealing with...so, the early, early stage, we were actually trying to be a Datadog competitor, where we were going to be self-hosted. Eventually, we focused on logs because we found out that's what was a big problem for most people, just event data, not just metric but generally event data, so logs, traces, et cetera. We built out our own logs database completely from scratch. And one of the things we stumbled upon was; basically, you have three things when it comes to logging, which is low cost, low latency, and large scale. That's what everybody wants. But you can't get all three of them; you can only get two of them. And we opted...like, we chose large scale and low cost. And when it comes to latency, we say it should be just fast enough, right? And that's where we focused on, and this is how we started building it. And with that, this is how we managed to stand out by just having way lower cost than anybody else in the industry and dealing with large scale. VICTORIA: That's really interesting. And how did you approach making the ingestion pipeline for masses amount of data more efficient? SEIF: Just make it coordination-free as possible, right? And get rid of Kafka because Kafka just, you know, drains your...it's where you throw in money. Like maintaining Kafka...it's like back then Elasticsearch, right? Elasticsearch was the biggest part of your infrastructure that would cost money. Now, it's also Kafka. So, we found a way to have our own internal way of queueing things without having to rely on Kafka. As I said, we wrote everything from scratch to make it work. Like, every now and then, I think that we can spin this out of the company and make it a new product. But now, eyes on the prize, right? JOE: It's interesting to hear that somebody who spent so much time in the open-source community ended up rolling their own solution to so many problems. Do you feel like you had some lessons learned from open source that led you to reject solutions like Kafka, or how did that journey go? SEIF: I don't think I'm rejecting Kafka. The problem is how Kafka is built, right? Kafka is still...you have to set up all these servers. They have to communicate, et cetera, etcetera. They didn't build it in a way where it's stateless, and that's what we're trying to go to. We're trying to make things as stateless as possible. So, Kafka was never built for the cloud-native era. And you can't really rely on SQS or something like that because it won't deal with this high throughput. So, that's why I said, like, we will sacrifice some latency, but at least the cost is low. So, if messages show after half a second or a second, I'm good. It doesn't have to be real-time for me. So, I had to write a couple of these things. But also, it doesn't mean that we reject open source. Like, we actually do like open source. We open-source a couple of libraries. We contribute back to open source, right? We needed a solution back then for that problem, and we couldn't find any. And maybe one day, open source will have, right? JOE: Yeah. I was going to ask if you considered open-sourcing any of your high latency, high throughput solutions. SEIF: Not high latency. You make it sound bad. JOE: [laughs] SEIF: You make it sound bad. It's, like, fast enough, right? I'm not going to compete on milliseconds because, also, I'm competing with ClickHouse. I don't want to compete with ClickHouse. ClickHouse is low latency and large scale, right? But then the cost is, you know, off the charts a bit sometimes. I'm going the other route. Like, you know, it's fast enough. Like, how, you know, if it's under two, three seconds, everybody's happy, right? If the results come within two, three seconds, everybody is happy. If you're going to build a real-time trading system on top of it, I'll strongly advise against that. But if you're building, you know, you're looking at dashboards, you're more in the observability field, yeah, we're good. VICTORIA: Yeah, I'm curious what you found, like, which customer personas that market really resonated with. Like, is there a particular, like, industry type where you're noticing they really want to lower their cost, and they're okay with this just fast enough latency? SEIF: Honestly, with the current recession, everybody is okay with giving up some of the speed to reduce the money because I think it's not linear reduction. It's more exponential reduction at this point, right? You give up a second, and you're saving 30%. You give up two seconds, all of a sudden, you're saving 80%. So, I'd say in the beginning, everybody thought they need everything to be very, very fast. And now they're realizing, you know, with limitations you have around your budget and spending, you're like, okay, I'm okay with the speed. And, again, we're not slow. I'm just saying people realize they don't need everything under a second. They're okay with waiting for two seconds. VICTORIA: That totally resonates with me. And I'm curious if you can add maybe a non-technical or a real-life example of, like, how this impacts the operations of a company or organization, like, if you can give us, like, a business-y example of how this impacts how people work. SEIF: I don't know how, like, how do people work on that? Nothing changed, really. They're still doing the, like...really nothing because...and that aspect is you run a query, and, again, as I said, you're not getting the result in a second. You're just waiting two seconds or three seconds, and it's there. So, nothing really changed. I think people can wait three seconds. And we're still like–when I say this, we're still faster than most others. We're just not as fast as people who are trying to compete on a millisecond level. VICTORIA: Yeah, that's okay. Maybe I'll take it back even, like, a step further, right? Like, our audience is really sometimes just founders who almost have no formal technical training or background. So, when we talk about observability, sometimes people who work in DevOps and operations all understand it and kind of know why it's important [laughs] and what we're talking about. So, maybe you could, like, go back to -- SEIF: Oh, if you're asking about new types of people who've been using it -- VICTORIA: Yeah. Like, if you're going to explain to, like, a non-technical founder, like, why your product is important, or, like, how people in their organization might use it, what would you say? SEIF: Oh, okay, if you put it like that. It's more of if you have data, timestamp data, and you want to run analytics on top of it, so that could be transactions, that could be web vitals, rather than count every time somebody visits, you have a timestamp. So, you can count, like, how many visitors visited the website and what, you know, all these kinds of things. That's where you want to use something like Axiom. That's outside the DevOps space, of course. And in DevOps space, there's so many other things you use Axiom for, but that's outside the DevOps space. And we actually...we implemented as zero-config integration with Vercel that kind of went viral. And we were, for a while, the number one enterprise for self-integration because so many people were using it. So, Vercel users are usually not necessarily writing the most complex backends, but a lot of things are happening on the front-end side of things. And we would be giving them dashboards, automated dashboards about, you know, latencies, and how long a request took, and how long the response took, and the content type, and the status codes, et cetera, et cetera. And there's a huge user base around that. VICTORIA: I like that. And it's something, for me, you know, as a managing director of our platform engineering team, I want to talk more to founders about. It's great that you put this product and this app out into the world. But how do you know that people are actually using it? How do you know that people, like, maybe, are they all quitting after the first day and not coming back to your app? Or maybe, like, the page isn't loading or, like, it's not working as they expected it to. And, like, if you don't have anything observing what users are doing in your app, then it's going to be hard to show that you're getting any traction and know where you need to go in and make corrections and adjust. SEIF: We have two ways of doing this. Right now, internally, we use our own tools to see, like, who is sending us data. We have a deployment that's monitoring production deployment. And we're just, you know, seeing how people are using it, how much data they're sending every day, who stopped sending data, who spiked in sending data sets, et cetera. But we're using Mixpanel, and Dominic, our Head of Product, implemented a couple of key metrics to that for that specifically. So, we know, like, what's the average time until somebody starts going from building its own queries with the builder to writing APL, or how long it takes them from, you know, running two queries to five queries. And, you know, we just start measuring these things now. And it's been going...we've been growing healthy around that. So, we tend to measure user interaction, but also, we tend to measure how much data is being sent. Because let's keep in mind, usually, people go in and check for things if there's a problem. So, if there's no problem, the user won't interact with us much unless there's a notification that kicks off. We also just check, like, how much data is being sent to us the whole time. VICTORIA: That makes sense. Like, you can't just rely on, like, well, if it was broken, they would write a [chuckles], like, a question or something. So, how do you get those metrics and that data around their interactions? So, that's really interesting. So, I wonder if we can go back and talk about, you know, we already mentioned a little bit about, like, the early days of Axiom and how you got started. Was there anything that you found in the early discovery process that was surprising and made you pivot strategy? SEIF: A couple of things. Basically, people don't really care about the tech as much as they care [inaudible 12:51] and the packaging, so that's something that we had to learn. And number two, continuous feedback. Continuous feedback changed the way we worked completely, right? And, you know, after that, we had a Slack channel, then we opened a Discord channel. And, like, this continuous feedback coming in just helps with iterating, helps us with prioritizing, et cetera. And that changed the way we actually developed product. VICTORIA: You use Slack and Discord? SEIF: No. No Slack anymore. We had a community Slack. We had a community [inaudible 13:19] Slack. Now, there's no community Slack. We only have a community Discord. And the community Slack is...sorry, internally, we use Slack, but there's a community Discord for the community. JOE: But how do you keep that staffed? Is it, like, everybody is in the Discord during working hours? Is it somebody's job to watch out for community questions? SEIF: I think everybody gets involved now just...and you can see it. If you go on our Discord, you will just see it. Just everyone just gets involved. I think just people are passionate about what they're doing. At least most people are involved on Discord, right? Because there's, like, Discord the help sections, and people are just asking questions and other people answering. And now, we reached a point where people in the community start answering the questions for other people in the community. So, that's how we see it's starting to become a healthy community, et cetera. But that is one of my favorite things: when I see somebody from the community answering somebody else, that's a highlight for me. Actually, we hired somebody from that community because they were so active. JOE: Yeah, I think one of the biggest signs that a product is healthy is when there's a healthy ecosystem building up around it. SEIF: Yeah, and Discord reminds me of the old days of open sources like IRC, just with memes now. But because all of us come from the old IRC days, being on Discord and chatting around, et cetera, et cetera, just gives us this momentum back, gave us this momentum back, whereas Slack always felt a bit too businessy to me. JOE: Slack is like IRC with emoji. Discord is IRC with memes. SEIF: I would say Slack reminds me somehow of MSN Messenger, right? JOE: I feel like there's a huge slam on MSN Messenger here. SEIF: [laughs] What do you guys use internally, Slack or? I think you're using Slack, right? Or Teams. Don't tell me you're using Teams. JOE: No, we're using Slack. SEIF: Okay, good, because I shit talk. Like, there is this, I’ll sh*t talk here–when I start talking about Teams, so...I remember that one thing Google did once, and that failed miserably. JOE: Google still has, like, seven active chat products. SEIF: Like, I think every department or every, like, group of engineers just uses one of them internally. I'm not sure. Never got to that point. But hey, who am I to judge? VICTORIA: I just feel like I end up using all of them, and then I'm just rotating between different tabs all day long. You maybe talked me into using Discord. I feel like I've been resisting it, but you got me with the memes. SEIF: Yeah, it's definitely worth it. It's more entertaining. More noise, but more entertaining. You feel it's alive, whereas Slack is...also because there's no, like, history is forever. So, you always go back, and you're like, oh my God, what the hell is this? VICTORIA: Yeah, I have, like, all of them. I'll do anything. SEIF: They should be using Axiom in the background. Just send data to Axiom; we can keep your chat history. VICTORIA: Yeah, maybe. I'm so curious because, you know, you mentioned something about how you realized that it didn't matter really how cool the tech was if the product packaging wasn't also appealing to people. Because you seem really excited about what you've built. So, I'm curious, so just tell us a little bit more about how you went about trying to, like, promote this thing you built. Or was, like, the continuous feedback really early on, or how did that all kind of come together? SEIF: The continuous feedback helped us with performance, but actually getting people to sign up and pay money it started early on. But with Vercel, it kind of skyrocketed, right? And that's mostly because we went with the whole zero-config approach where it's just literally two clicks. And all of a sudden, Vercel is sending your data to Axiom, and that's it. We will create [inaudible 16:33]. And we worked very closely with Vercel to do this, to make this happen, which was awesome. Like, yeah, hats off to them. They were fantastic. And just two clicks, three clicks away, and all of a sudden, we created Axiom organization for you, the data set for you. And then we're sending it...and the data from Vercel is being forwarded to it. I think that packaging was so simple that it made people try it out quickly. And then, the experience of actually using Axiom was sticky, so they continued using it. And then the price was so low because we give 500 gigs for free, right? You send us 500 gigs a month of logs for free, and we don't care. And you can start off here with one terabyte for 25 bucks. So, people just start signing up. Now, before that, it was five terabytes a month for $99, and then we changed the plan. But yeah, it was cheap enough, so people just start sending us more and more and more data eventually. They weren't thinking...we changed the way people start thinking of “what am I going to send to Axiom” or “what am I going to send to my logs provider or log storage?” To how much more can I send? And I think that's what we wanted to reach. We wanted people to think, how much more can I send? JOE: You mentioned latency and cost. I'm curious about...the other big challenge we've seen with observability platforms, including logs, is cardinality of labels. Was there anything you had to sacrifice upfront in terms of cardinality to manage either cost or volume? SEIF: No, not really. Because the way we designed it was that we should be able to deal with high cardinality from scratch, right? I mean, there's open-source ways of doing, like, if you look at how, like, a column store, if you look at a column store and every dimension is its own column, it's just that becomes, like, you can limit on the amount of columns you're creating, but you should never limit on the amount of different values in a column could be. So, if you're having something like stat tags, right? Let's say hosting, like, hostname should be a column, but then the different hostnames you have, we never limit that. So, the cardinality on a value is something that is unlimited for us, and we don't really see it in cost. It doesn't really hit us on cost. It reflects a bit on compression if you get into technical details of that because, you know, high cardinality means a lot of different data. So, compression is harder, but it's not repetitive. But then if you look at, you know, oh, I want to send a lot of different types of fields, not values with fields, so you have hostname, and latency, and whatnot, et cetera, et cetera, yeah, that's where limitation starts because then they have...it's like you're going to a wide range of...and a wider dimension. But even that, we, yeah, we can deal with thousands at this point. And we realize, like, most people will not need more than three or four. It's like a Postgres table. You don't need more than 3,000 to 4000 columns; else, you know, you're doing a lot. JOE: I think it's actually pretty compelling in terms of cost, though. Like, that's one of the things we've had to be most careful about in terms of containing cost for metrics and logs is, a lot of providers will...they'll either charge you based on the number of unique metric combinations or the performance suffers greatly. Like, we've used a lot of Prometheus-based solutions. And so, when we're working with developers, even though they don't need more than, you know, a few dozen metric combinations most of the time, it's hard for people to think of what they need upfront. It's much easier after you deploy it to be able to query your data and slice it retroactively based on what you're seeing. SEIF: That's the detail. When you say we're using Prometheus, a lot of the metrics tools out there are using, just like Prometheus, are using the Gorilla data structure. And the real data structure was never designed to deal with high cardinality labels. So, basically, to put it in a simple way, every combination of tags you send for metrics is its own file on disk. That's, like, the very simple way of explaining this. And then, when you're trying to search through everything, right? And you have a lot of these combinations. I actually have to get all these files from this conversion back together, you know, and then they're chunked, et cetera. So, it's a problem. Generally, how metrics are doing it...most metrics products are using it, even VictoriaMetrics, et cetera. What they're doing is they're using either the Prometheus TSDB data structure, which is based on Gorilla. Influx was doing the same thing. They pivoted to using more and more like the ones we use, and Honeycomb uses, right? So, we might not be as fast on metrics side as these highly optimized. But then when it comes to high [inaudible 20:49], once we start dealing with high cardinality, we will be faster than those solutions. And that's on a very technical level. JOE: That's pretty cool. I realize we're getting pretty technical here. Maybe it's worth defining cardinality for the audience. SEIF: Defining cardinality to the...I mean, we just did that, right? JOE: What do you think, Victoria? Do you know what cardinality is now? [laughs] VICTORIA: All right. Now I'm like, do I know? I was like, I think I know what it means. Cardinality is, like, let's say you have a piece of data like an event or a transaction. SEIF: It's like the distinct count on a property that gives you the cardinality of a property. VICTORIA: Right. It's like how many pieces of information you have about that one event, basically, yeah. JOE: But with some traditional metrics stores, it's easy to make mistakes. For example, you could have unbounded cardinality by including response time as one of the labels -- SEIF: Tags. JOE: And then it's just going to -- SEIF: Oh, no, no. Let me give you a better one. I put in timestamp at some point in my life. JOE: Yeah, I feel like everybody has done that one. [laughter] SEIF: I've put a system timestamp at some point in my life. There was the actual timestamp, and there was a system timestamp that I would put because I wanted to know when the...because I couldn't control the timestamp, and the only timestamp I had was a system timestamp. I would always add the actual timestamp of when that event actually happened into a metric, and yeah, that did not scale. MID-ROLL AD: Are you an entrepreneur or start-up founder looking to gain confidence in the way forward for your idea? At thoughtbot, we know you’re tight on time and investment, which is why we’ve created targeted 1-hour remote workshops to help you develop a concrete plan for your product’s next steps. Over four interactive sessions, we work with you on research, product design sprint, critical path, and presentation prep so that you and your team are better equipped with the skills and knowledge for success. Find out how we can help you move the needle at tbot.io/entrepreneurs. VICTORIA: Yeah. I wonder if you could maybe share, like, a story about when it's gone wrong, and you've suddenly charged a lot of money [laughs] just to get information about what's happening in the system. Any, like, personal experiences with observability that kind of informed what you did with Axiom? SEIF: Oof, I have a very bad one, like, a very, very bad one. I used to work for a company. We had to deploy Elasticsearch on Windows Servers, and it was US-East-1. So, just a combination of Elasticsearch back in 2013, 2014 together with Azure and Windows Server was not a good idea. So, you see where this is going, right? JOE: I see where it's going. SEIF: Eventually, we had, like, we get all these problems because we used Elasticsearch and Kibana as our, you know, observability platform to measure everything around the product we were building. And funny enough, it cost us more than actually maintaining the infrastructure of the product. But not just that, it also kept me up longer because most of the downtimes I would get were not because of the product going down. It's because my Elasticsearch cluster started going down, and there's reasons for that. Because back then, Microsoft Azure thought that it's okay for any VM to lose connection with the rest of the VMs for 30 seconds per day. And then, all of a sudden, you have Elasticsearch with a split-brain problem. And there was a phase where I started getting alerted so much that back then, my partner threatened to leave me. So I bought a...what I think was a shock bracelet or a shock collar via Bluetooth, and I connected it to phone for any notification. And I bought that off Alibaba, by the way. And I would charge it at night, put it on my wrist, and go to sleep. And then, when alert happens, it will fully discharge the battery on me every time. JOE: Okay, I have to admit, I did not see where that was going. SEIF: Yeah, did that for a while; definitely did not save my relationship either. But eventually, that was the point where, you know, we started looking into other observability tools like Datadog, et cetera, et cetera, et cetera. And that's where the actual journey began, where we moved away from Elasticsearch and Kibana to look for something, okay, that we don't have to maintain ourselves and we can use, et cetera. So, it's not about the costs as much; it was just pain. VICTORIA: Yeah, pain is a real pain point, actual physical [chuckles] and emotional pain point [laughter]. What, like, motivates you to keep going with Axiom and to keep, like, the wind in your sails to keep working on it? SEIF: There's a couple of things. I love working with my team. So, honestly, I just wake up, and I compliment my team. I just love working with them. They're a lot of fun to work with. And they challenge me, and I challenge them back. And I upset them a lot. And they can't upset me, but I upset them. But I love working with them, and I love working with that team. And the other thing is getting, like, having this constant feedback from customers just makes you want to do more and, you know, close sales, et cetera. It's interesting, like, how I'm a very technical person, and I'm more interested in sales because sales means your product works, the product, the technical parts, et cetera. Because if technically it's not working, you can't build a product on top of it. And if you're not selling it, then what's the point? You only sell when the product is good, more or less, unless you're Oracle. VICTORIA: I had someone ask me about Oracle recently, actually. They're like, "Are you considering going back to it?" And I'm maybe a little allergic to it from having a federal consulting background [laughs]. But maybe they'll come back around. I don't know. We'll see. SEIF: Did you sell your soul back then? VICTORIA: You know, I feel like I just grew up in a place where that's what everyone did was all. SEIF: It was Oracle, IBM, or HP back in the day. VICTORIA: Yeah. Well, basically, when you're working on applications that were built in, like, the '80s, Oracle was, like, this hot, new database technology [laughs] that they just got five years ago. So, that's just, yeah, interesting. SEIF: Although, from a database perspective, they did a lot of the innovations. A lot of first innovations could have come from Oracle. From a technical perspective, they're ridiculous. I'm not sure from a product perspective how good they are. But I know their sales team is so big, so huge. They don't care about the product anymore. They can still sell. VICTORIA: I think, you know, everything in tech is cyclical. So, you know, if they have the right strategy and they're making some interesting changes over there, there's always a chance [laughs]. Certain use cases, I mean, I think that's the interesting point about working in technology is that you know, every company is a tech company. And so, there's just a lot of different types of people, personas, and use cases for different types of products. So, I wonder, you know, you kind of mentioned earlier that, like, everyone is interested in Axiom. But, you know, I don't know, are you narrowing the market? Or, like, how are you trying to kind of focus your messaging and your sales for Axiom? SEIF: I'm trying to focus on developers. So, we're really trying to focus on developers because the experience around observability is crap. It's stupid expensive. Sorry for being straightforward, right? And that's what we're trying to change. And we're targeting developers mainly. We want developers to like us. And we'll find all these different types of developers who are using it, and that's the interesting thing. And because of them, we start adding more and more features, like, you know, we added tracing, and now that enables, like, billions of events pushed through for, you know, again, for almost no money, again, $25 a month for a terabyte of data. And we're doing this with metrics next. And that's just to address the developers who have been giving us feedback and the market demand. I will sum it up, again, like, the experience is crap, and it's stupid expensive. I think that's the [inaudible 28:07] of observability is just that's how I would sum it up. VICTORIA: If you could go back in time and talk to yourself when you were still a developer, now that you're CTO, what advice would you give yourself? JOE: Besides avoiding shock collars. VICTORIA: [laughs] Yes. SEIF: Get people's feedback quickly so you know you're on the right track. I think that's very, very, very, very important. Don't just work in the dark, or don't go too long into stealth mode because, eventually, people catch up. Also, ship when you're 80% ready because 100% is too late. I think it's the same thing here. JOE: Ship often and early. SEIF: Yeah, even if it's not fully ready, it's still feedback. VICTORIA: Ship often and early and talk to people [laughs]. Just, do you feel like, as a developer, did you have the skills you needed to be able to get the most out of those feedback and out of those conversations you were having with people around your product? SEIF: I still don't think I'm good enough. You're just constantly learning, right? I just accepted I'm part of a team, and I have my contributions. But as an individual, I still don't think I know enough. I think there's more I need to learn at this point. VICTORIA: I wonder, what questions do you have for me or Joe? SEIF: How did you start your podcast, and why the name? VICTORIA: Oh, man, I hope I can answer. So, the podcast was started...I think it's, like, we're actually about to be at our 500th Episode. So, I've only been a host for the last year. Maybe Joe even knows more than I do. But what I recall is that one person at thoughtbot thought it would be a great idea to start a podcast, and then they did it. And it seems like the whole company is obsessed with robots. I'm not really sure where that came from. There used to be a tiny robot in the office, is what I remember. And people started using that as, like, the mascot. And then, yeah, that's it, that's the whole thing. SEIF: Was the robot doing anything useful or just being cute? JOE: It was just cute, and it's hard to make a robot cute. SEIF: Was it a real robot, or was it like a -- JOE: No, there was, at one point, a toy robot. The name...I actually forget the origin–origin of the name, but the name Giant Robots comes from our blog. So, we named the podcast the same as the blog: Giant Robots Smashing Into Other Giant Robots. SEIF: Yes, it's called transformers. VICTORIA: Yeah, I like it. It's, I mean, now I feel like -- SEIF: [laughs] VICTORIA: We got to get more, like, robot dogs involved [laughs] in the podcast. SEIF: Like, I wanted to add one thing when we talked about, you know, what gets me going. And I want to mention that I have a six-month-old son now. He definitely adds a lot of motivation for me to wake up in the morning and work. But he also makes me wake up regardless if I want to or not. VICTORIA: Yeah, you said you had invented an alarm clock that never turns off. Never snoozes [laughs]. SEIF: Yes, absolutely. VICTORIA: I have the same thing, but it's my dog. But he does snooze, actually. He'll just, like, get tired and go back to sleep [laughs]. SEIF: Oh, I have a question. Do dogs have a Tamagotchi phase? Because, like, my son, the first three months was like a Tamagotchi. It was easy to read him. VICTORIA: Oh yeah, uh-huh. SEIF: Noisy but easy. VICTORIA: Yes, yes. SEIF: Now, it's just like, yeah, I don't know, like, the last month he has opinions at six months. I think it's because I raised him in Europe. I should take him back to the Middle East [laughs]. No opinions. VICTORIA: No, dogs totally have, like, a communication style, you know, I pretty much know what he, I mean, I can read his mind, obviously [laughs]. SEIF: Sure, but that's when they grow a bit. But what when they were very...when the dog was very young? VICTORIA: Yeah, they, I mean, they also learn, like, your stuff, too. So, they, like, learn how to get you to do stuff or, like, I know she'll feed me if I'm sitting here [laughs]. SEIF: And how much is one dog year, seven years? VICTORIA: Seven years. SEIF: Seven years? VICTORIA: Yeah, seven years? SEIF: Yeah. So, basically, in one year, like, three months, he's already...in one month, he's, you know, seven months old. He's like, yeah. VICTORIA: Yeah. In a year, they're, like, teenagers. And then, in two years, they're, like, full adults. SEIF: Yeah. So, the first month is basically going through the first six months of a human being. So yeah, you pass...the first two days or three days are the Tamagotchi phase that I'm talking about. VICTORIA: [chuckles] I read this book, and it was, like, to understand dogs, it's like, they're just like humans that are trying to, like, maximize the number of positive experiences that they have. So, like, if you think about that framing around all your interactions about, like, maybe you're trying to get your son to do something, you can be like, okay, how do I, like, I don't know, train him that good things happen when he does the things I want him to do? [laughs] That's kind of maybe manipulative but effective. So, you're not learning baby sign language? You're just, like, going off facial expressions? SEIF: I started. I know how Mama looks like. I know how Dada looks like. I know how more looks like, slowly. And he already does this thing that I know that when he's uncomfortable, he starts opening and closing his hands. And when he's completely uncomfortable and basically that he needs to go sleep, he starts pulling his own hair. VICTORIA: [laughs] I do the same thing [laughs]. SEIF: You pull your own hair when you go to sleep? I don't have that. I don't have hair. VICTORIA: I think I do start, like, touching my head though, yeah [inaudible 33:04]. SEIF: Azure took the last bit of hair I had! Went away with Azure, Elasticsearch, and the shock collar. VICTORIA: [laughs] SEIF: I have none of them left. Absolutely nothing. I should sue Elasticsearch for this shit. VICTORIA: [laughs] Let me know how that goes. Maybe there's more people who could join your lawsuit, you know, with a class action. SEIF: [laughs] Yeah. Well, one thing I wanted to also just highlight is, right now, one of the things that also makes the company move forward is we realized that in a single domain, we proved ourselves very valuable to specific companies, right? So, that was a big, big thing, milestone for us. And now we're trying to move into a handful of domains and see which one of those work out the best for us. Does that make sense? VICTORIA: Yeah. And I'm curious: what are the biggest challenges or hurdles that you associate with that? SEIF: At this point, you don't want just feedback. You want constructive criticism. Like, you want to work with people who will criticize the applic...and you iterate with them based on this criticism, right? They're just not happy about you and trying to create design partners. So, for us, it was very important to have these small design partners who can work with us to actually prove ourselves as valuable in a single domain. Right now, we need to find a way to scale this across several domains. And how do you do that without sacrificing? Like, how do you open into other domains without sacrificing the original domain you came from? So, there's a lot of things [inaudible 34:28]. And we are in the middle of this. Honestly, I Forrest Gumped my way through half of this, right? Like, I didn't know what I was doing. I had ideas. I think it's more of luck at this point. And I had luck. No, we did work. We did work a lot. We did sleepless nights and everything. But I think, in the last three years, we became more mature and started thinking more about product. And as I said, like, our CEO, Neil, and Dominic, our head of product, are putting everything behind being a product-led organization, not just a tech-led organization. VICTORIA: That's super interesting. I love to hear that that's the way you're thinking about it. JOE: I was just curious what other domains you're looking at pushing into if you can say. SEIF: So, we are going to start moving into ETL a bit more. We're trying to see how we can fit in specific ML scenarios. I can't say more about the other, though. JOE: Do you think you'll take the same approaches in terms of value proposition, like, low cost, good enough latency? SEIF: Yes, that's definitely one thing. But there's also...so, this is the values we're bringing to the customer. But also, now, our internal values are different. Now it's more of move with urgency and high velocity, as we said before, right? Think big, work small. The values in terms of values we're going to take to the customers it's the same ones. And maybe we'll add some more, but it's still going to be low-cost and large-scale. And, internally, we're just becoming more, excuse my French, agile. I hate that word so much. Should be good with Scrum. VICTORIA: It's painful, but everyone knows what you're talking about [laughs], you know, like -- SEIF: See, I have opinions here about Scrum. I think Scrum should be only used in terms of iceScrum [inaudible 36:04], or something like that. VICTORIA: Oh no [laughter]. Well, it's a Rugby term, right? Like, that's where it should probably stay. SEIF: I did not know it's a rugby term. VICTORIA: Yeah, so it should stay there, but -- SEIF: Yes [laughs]. VICTORIA: Yeah, I think it's interesting. Yeah, I like the being flexible. I like the just, like, continuous feedback and how you all have set up to, like, talk with your customers. Because you mentioned earlier that, like, you might open source some of your projects. And I'm just curious, like, what goes into that decision for you when you're going to do that? Like, what makes you think this project would be good for open source or when you think, actually, we need to, like, keep it? SEIF: So, we open source libraries, right? We actually do that already. And some other big organizations use our libraries; even our competitors use our libraries, that we do. The whole product itself or at least a big part of the product, like database, I'm not sure we're going to open source that, at least not anytime soon. And if we open source, it's going to be at a point where the value-add it brings is nothing compared to how well our product is, right? So, if we can replace whatever's at the back with...the storage engine we have in the back with something else and the product doesn't get affected, that's when we open source it. VICTORIA: That's interesting. That makes sense to me. But yeah, thank you for clarifying that. I just wanted to make sure to circle back. Since you have this big history in open source, yeah, I'm curious if you see... SEIF: Burning me out? VICTORIA: Burning you out, yeah [laughter]. Oh, that's a good question. Yeah, like, because, you know, we're about to be in October here. Do you have any advice or strategies as a maintainer for not getting burned out during the next couple of weeks besides, like, hide in a cave and without internet access [laughs]? SEIF: Stay away from Reddit and Hacker News. That's my goal for October now because I'm always afraid of getting too attached to an idea, or too motivated, or excited by an idea that I drift away from what I am actually supposed to be doing. VICTORIA: Last question is, is there anything else you would like to promote? SEIF: Yeah, check out our website; I think it's at axiom.co. Check it out. Sign up. And comment on Discord and talk to me. I don't bite, sometimes grumpy, but that's just because of lack of sleep in the morning. But, you know, around midday, I'm good. And if you're ever in Berlin and you want to hang out, I'm more than willing to hang out. VICTORIA: Whoo, that's awesome. Yeah, Berlin is great. I was there a couple of years ago but no plans to go back anytime soon, but maybe I'll keep that in mind. You can subscribe to the show and find notes along with a complete transcript for this episode at giantrobots.fm. If you have questions or comments, email us at [email protected]. And you could find me on Twitter @victori_ousg. And this podcast is brought to you by thoughtbot and produced and edited by Mandy Moore. Thanks for listening. See you next time. Did you know thoughtbot has a referral program? If you introduce us to someone looking for a design or development partner, we will compensate you if they decide to work with us. More info on our website at tbot.io/referral. Or you can email us at [email protected] with any questions.Special Guests: Joe Ferris and Seif Lotfy.Sponsored By:thoughtbot: Are you an entrepreneur or start-up founder looking to gain confidence in the way forward for your idea? At thoughtbot, we know you’re tight on time and investment, which is why we’ve created targeted 1-hour remote workshops to help you develop a concrete plan for your product’s next steps. Over four interactive sessions, we work with you on research, product design sprint, critical path, and presentation prep so that you and your team are better equipped with the skills and knowledge for success. Find out how we can help you move the needle at: tbot.io/entrepreneursSupport Giant Robots Smashing Into Other Giant Robots