Today's episode is a different one. It's an AMA where I answer questions that you submitted. Ask them questions as Geeks, that is Volimer Gigniak, a C2 at Wordsmith. Wordsmith is a legal AI startup where I'm an investor and not a team well, and Geeks was just to tell it to help out with this AMA. We've moved the questions as, obviously, questions across the industry, opinions on AI, opinions on hiring, questions about myself, advice on specific situations, and the private engineer as a business.
Thanks to Antisysit for being our presenting sponsor. With Antisysit, you can verify your systems correctness without human review or traditional interrogation tests and avoid bugs or outages. With this, let's jump in. Hey, Gage. Welcome to the Srivastavast.com AMA.
It's really nice to be a guest on my own podcast. This is really cool. Thanks, Geeks, for coming here. For some background, we know each other from Wordsmith, which is one of the very few startups I still invest in, because I stopped investing, but about two years ago I invested with a friend, Ross, who I work together. It's really nice to have you here. I appreciate you putting trust in us and investing,
and let's get started. First question, what made you switch from a full I.C. role, like a Uber to focus on sharing reporting their content? Yeah, so at Uber, I started as an I.C. and I was an I.C.
for about 10 years before Uber. I started as a senior engineer. I became an injury manager pretty quickly. It wasn't an I.C. role, but I guess a manager role, but it doesn't change the story too much. I was hitting about four years at Uber, and two things happened at the same time.
One is Uber in 2020 had layoffs because COVID hit Uber's business really bad. I had access to our internal dashboards where we saw the revenue for rides, and it was just going down very close to zero. I was actually sharing it to my team because I figured transparency is a good thing. I'm not sure that was the smartest thing, but I probably still do it again. I was wrong.
People are like, that's not looking good. We were all collectively freaking out a little bit. And layoffs came very predictably. It was a 20% layoffs. About a quarter of my team was unfortunately gone. The remainder of my team, our mission no longer made sense in this new world where
we were building something for drivers when we thought there would not be as many drivers who would have to compete with them, but because the COVID drivers were actually flocking to a platform. And so I got a new team to work with, but it felt to me for the first time in four years that they were going really well. And I just felt demotivated. I knew that the business would be doing poorly. And I also asked myself what I wanted to do after Uber. And before I joined Uber, I got this offer,
which was an amazing compensation package, which was a bunch of stock. And I told myself, well, stock, I mean, who knows if we were a world publicer or not. But I said if Uber does go public, and this money turns into stock, I got about $500,000 worth of stock as a grand option. I'm like, if I have 500k in my bank account, well, I can take a risk and for the next thing, I can actually do a startup. So I remember this at Newbury had gone public and that 500k stock turned into 400k because of the stock prices a bit lower. And then you have to pay taxes on it. So it was less. But I still had
a lump sum sitting in my savings account that I was like, I don't have to work actually for like, I could not work for like two, three years easily. I was like, well, maybe I should take a risk. And my plan was leave Uber, finish writing the software in your guidebook, which is something I started writing at Uber just finished it in six months. And afterwards, do what you've done, which is start a startup, join a startup. Because I was a little bit tired of being a middle manager. They
tell you your manager, you want to go graduate, you become manager, they should have said you became a middle manager, because now your job is to keep your team happy to keep management happy. Especially I was in a different region. I was in Europe. So this was easy. But when layoffs came, it was a lot of politics, a lot of explaining regulations that I wasn't what I wanted to.
It also keep your peers happy in terms of your manager peers, don't be retiring. And I was like, I want to be in charge next time, because I have a lot of ideas, but I felt I was like fighting the machine, if you will, in some sense. But that was my plan. It involved nothing with writing, except just finish this book. I have a legacy. I can give this book to people. I can be proud of it.
But then what happened is so much software engineering. When you started a project in software and you've never done before, you're a junior engineer, you're doing your first migration. You think it'll take two days, and then two months later, you're still stuck there. And it was the same thing with writing this book. I've never written a book. I know it's a big project, but I was like,
yeah, six months should be enough. Six months later, I'm still, I'm like, treading water. I wrote three other short books. So but my main book was not progressing. I asked myself like, okay, like, I gave myself about six to eight months to like, I did this book out and then just go have a real job. My mind, a real job was either just to start a startup be a founder or go back to
being an engineer manager or staff engineer or CTO somewhere in this smaller place. And I was like, okay, well, I should be honest with myself. Like what am I doing right now? And what will I be doing? And it was like, either I start and I raise funds to start a startup. And my idea of startup was just
Uber inside had a lot of platform engineering teams, copy one of the things that they were doing. My idea was actually we had an internal RFC system request for comments where we actually had a system that put these Google Docs together and we graded and all that and it was a critical system. I thought maybe I could productionize that. A lot of Uber startups actually came from people looking at internal platform stuff and taking it and either making it open source, some Porel is exuber, promispere, exuber, or observability system and many others. So actually, it's not all that radical. But then I was like, well, if I did that, I just have to fully focus on that. And on
the side, I was doing writing, I was writing a few books actually, I was blogging, I was doing YouTube videos out of fun. And I was like, well, I need to stop that. If I do that, because if I raise money, I will have to my investors, I will hire people. And for about five to 10 years, I'm going to have it to be just focus 100% on that. I talk with my brother, if he was on a second startup, and he said, like, look, if you started startup, do it because you are ready to spend 10 years of your life on it. Like, you need to believe that right now, because if you don't, it's not going to work,
it's really hard. It's not a popular thing to say. I wasn't sure I was right to spend 10 years on an RFC system. I wasn't that excited about it. And then I asked myself, like, okay, what is this drive?
Like, why do I really want to do this startup or a startup? And I was trying to be on site to answers. One was the money. In the sense that this was 2021, it seemed everywhere I looked exuber, startups, they were valued a billion. They were unicorns in a matter of like a year or two. It
seemed too easy. And I was reasonable. I was like, that would probably not happen to me. But what might happen is I might be able to build a unicorn in like, I say, 10 years time. And by that time, I will still, you know, I'm a sole founder. I might have five or 10% stake because I'll come to a lot
high dilution, which is $50 million. And let's say we have an exit and I leave, and then I pay taxes and I still have 25 million, which is like exactly 24 more than I would need outside of buying a house. And then I have this, you know, FU money. What would I do? The answer was like, well, I probably like, share what I know. I probably like, you know, write a book. I probably like, you know, do some YouTube
videos. That's kind of interesting. Like, I could do that right now. And the other reason I wanted to do this startup was the small teams. I always loved working both at Uber and at my previous companies, at Sky Scanner where I met Ross, co-founder of WordSmith. We were a small team, us against the
world. And I love that feeling like being either an engineer on that team or the manager of that team. I didn't enjoy being a manager of managers, but I had no longer connection. And that was the other reason. And actually, that was, I guess, the more legit reason. But in the end, I didn't have
this exciting idea. And I was like, if this started up with success, well, I would just be writing probably. So I was like, let me try that. I saw Salka was taking off. Lenny Rochi's key shared that he had 2,000 paid subscribers for a product management newsletter. I thought if Lenny has 2,000
paid for product management, there's 10 times as many software engineers as product managers and single team. And they're not as likely to divide. But there was no paid newsletters for software. So I was like, let me try it out. I gave myself six months and I figured it might not work. And then
it just worked. It took off. Makes sense. Next question. Have you seen engineering teams at BEC, the top that AI native is DLC and how does I collaborate across engineering product and design?
Yeah. So AI native SDLC, a software development lifecycle. Even the whole, you know, like SDLC is an interesting one before we go into AI. Because like, what is SDLC? It used to be, you plan, you code, you deploy, you monitor. And some people used to call this waterfall. And then there was Agile
where you just like iterate a lot faster. And interesting thing, like outside of Bigtech or outside of these large companies, if you go to a large company that is not like a big tech, not one of the Googles or Metas, they often have like pretty rigid processes around scrum. Specifically, they say we're very agile. We have scrum or they have the safe system, the scaled agile framework, which has a bunch of meetings and like a really rigid way to be agile. And of course, there's a bunch of like money and consulting and all that. But they think they're very agile.
And then a very surprised to see how most of teams inside of the likes of Uber or Meta or even Google work, which is like, oh, we kind of have this like, you know, problem, we actually plan, we sit together, we kind of do like, I don't know, a few days of planning, and then we code it, and then we deploy it, and then we get some feedback and we might iterate and they're like, wow, that's waterfall. We're so much more agile. And actually, like the whole thing about waterfall and agile is it doesn't matter anymore. Waterfall used to be a thing I talked with Ken back when it literally used to be like a year or two, a planning and like having like this much documentation. And we don't do that anymore. So the software development lifecycle is an interesting one. And almost every
modern company up to AI used to have RFCs or RFDs or designed off where people would write down because they realized that you should if you plan things ahead and then you build, you'll have better results like planning, sort of thinking through. Now the whole AI native as the LC, the closest I've seen to a company who is big and successful and making a lot of money and employing, you know, like hundreds or thousands of engineers is and traffic. They don't employ thousands of engineers. They employ probably hundreds of engineers right now, but they're very interesting place. They're not a product company decisively. They're a research lab and they just do everything super
fluidly like uncle, you can see it in cloth code and I talked with Boris turning about this. They don't do design docs. They just do prototypes all the time. They kind of show it to themselves. But I wonder if it's really replicable and I also wonder when it will break down. In the sense that
clock codes are great product. It's now the leading coding harness. So like they did an amazing job and just with prototypes and iteration and using AI and getting feedback and fixing it and responding on social media, they respond to bugs. They fix it immediately. But there's a question to me, like sometimes like how much do they plan to have a strategy? Like with pricing, they keep
changing the tiers back and forth and traffic is the closest I can think of. But I did not see any company that managed to really retrofit anything. When I'm seeing almost every company do, they are building AI infr systems. So for example, they will build a coding agent that's offered all their internal services. That's plugged into Google is doing this. Ramp is doing this. Uber is doing it.
So I think what's happening is they're building a lot better tooling to make this easier. And I think that's where we'll see. And I still have one last question, which is if you have a business that is working, it's making money. It has a rhythm. You have customers who are used to certain things.
How much do you want to change inside everything versus just changing it slowly to make sure, for example, case of Uber, people expect that when you press the button, the car arrives, the drivers are there. There's there's processes behind this, which are non software like you need to do outreach campaigns for the drivers. You need to let them know weeks in advance when there will be a big event so that they can prepare for it. The pace of the business has not changed because of AI, even though AI is a piece of development. And finally, when you just go too fast, you might forget the basics, which I'm seeing a lot. Spotify is a good example where I've talked with their CTO on their team and
they say they're AI very responsibly, which is great to hear. But then again, as a customer and a user, I'm so frustrated because they seem to be down so much. I can publish an episode two or three weeks ago because they were down and they don't have a status page. I don't know if it's AR not, right? It might not be. But then the whole site just went down. And I'm like, if you're using AI,
you're sure not using it for to make better reliability. Have you seen how AI is impacting what employers look for in candidates? Yeah, it's impacting it because it feels to me that they don't really know what to look for. I mean, I'm going to ask you for this one. I'm going to turn it because you guys are hiring.
How did it change how you're hiring for software engineering and I'll answer. Yeah, so in our case, we definitely structured interview quite differently. So the main thing that we're looking for now is the ability to reason through what the AI is doing and correct it and do the appropriate research. So actually, it's interesting, like our entry process, we give away a home board, which is not pretty classic, but we expect that this home work will be down with AI. But then we basically have a very long discussion around this home board and we are checking, okay, you pick this algorithm, was it an I-P-Gantt for you? Or did you actually do research and you
figure it out what is appropriate? Or here's the design decisions that you made. How did you make this decision? Again, is it automatic? Decision by AI or you understand it deeply and you can course correct? And then we are looking, so we are peaking into different parts of the code and we
are seeing how candidate can react on the spot, whether they can spot an issue, whether they can come up quickly with the solution to the issue. So basically, the ability to reason through and and research and not just apply all the solutions that AI generates automatically. So this makes a lot of sense and this is, I've seen a lot of similar things with startups doing it. And when we think of how hiring is changing with AI, before AI, there were two worlds in hiring. There was the Google interview process, which is the lead code interview process. And this is
because Google decided early on that they want to hire for raw intelligence, that they had puzzles initially, like how many golf balls fit in New York or something like that. But they realized that doesn't really scale that well. And they found coding interviews, like algorithm coding interviews to work really well because it's selected for a few things. It's selected for people who have computer science basics, which Google needed specifically going to universities where they teach computational complexity on some of those things. It also selected to apply under pressure, explain your thinking, and it's very scalable, meaning you can train like a thousand interviewers and give them like a pool of 200 questions. And it doesn't matter if a few questions leak,
the bar will look the same. And it works great for Google. It really does. Oh, and the bonus is that people once they know that this is expected of them, you need to prepare. And if you're unwilling to prepare for this, you're not going to be a good fit at a place like Google where sometimes you need to do stupid stuff. There's performance reviews, you need to do this thing. There's a new project
coming up which makes no sense, but we need to do it, but we need to do it. And corporate needs people who put up with BS processes every now and then without too much complaint. So it kind of selects for that. So it's kind of wonderful. And this is why most of the big tech has just adopted that. And Google
knows that you're not going to do that work. You're not going to use those algorithms. But again, it works good enough for them because they hire people who are adaptable, you learn stuff and you pick up a new things anyway. And in startups, you just hire for practicality. So this is where trial weeks have been popular, where a lot of startups used to hire by just giving you real work. For
example, take home fix a real bug in a few hours or a few days. And they could actually see like, oh, you're actually doing the work and startups were doing open source often, which is higher the contributors to the repository. What AI has changed is, first of all, the algorithm interview, it just visits through it. So remotely doing it no longer makes sense. And with the take home where you used to give someone a difficult take home, you can do it in an AI will complete it pretty well.
So you don't really get that signal. So my bet is that what will happen is these worlds will stay except the in person part is what decision will be made. You'll have filtering, like how to take home tasks that you can do with AI and you can cheat if you will. But when you will talk with them, on Google, they will still have you come into the office and you'll have to do this whiteboard interview. If you didn't prepare, like, no, AI is not going to save you because you don't know how to do it. And startups will probably want you to what you did is explain what you did. And a small
person of startups who can do, they will just have a trial weeks. What one that linear does come work with us for a week, like you need to collaborate, you can use AI, of course, you can, but it's not the main thing of it. So I think hiring will be honestly just more, there'll be more as a candidate, it'll be more friction, you'll need to invest more time. It'll feel more unfair because there will be no clear rules that we have been gotten used to and it'll be messy. It'll be also more subjective.
Just reality. Yeah, work together. By the way, it's amazing way to hire. We did that at the earlier stage just with the party to scale, but it's interesting that we never managed to scale it. Well, am I scaling you mean that? Yes, I started to do it so most candidates will say, yes, because
you need to take time off. The only reason linear can do it is they have hair very well known in the industry. And even a lot of people say that I'm sorry, I cannot do it. I love to work there, but I just don't have the time. And so they lose a bunch of those folks.
What kind of engineers are thriving and excelling right now? We hear about layoffs and slowdowns, but surely some are doing better than I were. Yeah, so we do get very close, but I talk with engineers who are very much in demand, just as so, or maybe more so than before. And what these people have is they either work at startups or well-known tech companies. They are interested in the business, they're so called product minded, they don't sell at borders. And by this time, when AI came around,
they just got into it. They somehow was, we's their way either at their company saying, okay, I'm going to work on this AI project building something on top of AI. Often AI and I will help build this part. And now they are actually considered experts in this. And most companies that are hiring and trying to hire positions, the ones that are to fill, is I'd like an engineer who has a few years of experience, they've actually built something with AI. They're not an absolute
noob to this. They will help me able to decide what architecture should we use, or actually we use fine tuning, should we use an off-the-shelf model, should we use our own model, should we do an on-prem, should we do off-prem? What about the inference costs? What about should we use Grog, should we use server? So whatever, five years ago, this was you hired an engineer who knew about cloud and could help you figure out at a startup. Now you're hiring someone who knows about inference
and some of these things. And so engineers who have been doing this are in very high demand. The only problem they have is sometimes if they work at the likes of Google, or or wealth on the startup, these other companies are surprised at how high of a composition asks if they have. But these people are very in high demand. The people who are having trouble is either at their current work. They just have no exposure to use any AI. So they don't have this experience
with the AI in front. They still build software and they use Cloud Code and Codex, but everyone does that. They feel a bit stuck on how to go about this and they don't have good pedigree, meaning they don't work at a company that is assumed to be a modern company. And those people are finding it hard to make the jumps. And now they're thinking should I just do some side projects? And my answer
will be like, well, very least, if you want to make that jump between the tiers of companies, and in my mind, there's I have the try model model, of course, but also I have this model of the company where you have consulting companies, where you're just like an Accenture or Kabbjamai, or one of these where you're given the client projects, they're really struggling right now. You have the product companies where you work and you build products. And within the product companies, you have the venture of funded product companies where you actually have a bunch of money to build quickly, scale, the composition won't be higher. You're now competing and hiring from the likes of Big Tech. And then at the very top, you have right now, it's AI lab, the Entropics, the OpenAI, whatever Google used to be in 2004 and meta in 2010, that is right, an Uber and for a short time in 2015 or so. Now that's that's entropic and open AI. And it's hard
to jump between these tiers. So for example, a lot of people are like, I have to work at Entropic. Well, I mean, dream big, but the reality is that I know so many people working at Google and meta and Facebook, they want to get into those places, but these places are extremely selective. So, Entrop Level, Web Product Engineers, et cetera, but what's the hiring landscape for juniors in low level system hardware software integration and buy that or the fan stack looks like? SingsR Plus or Genonature of Social System Level Thinking?
I'm less familiar with with lower level systems programming. I would just assume that it's not as saturated. When I talk with the Primateics Summit in February, I talk with the engineer who is working on low level systems, mostly C++ assembly. And we talked about who's using AI, clock, code, X, cursor, et cetera. And he was the only one in the group. There was about eight of us
talking, everyone's like, yeah, using it almost all the time. My code is generated by back then it was Opus 4.5 or 4.6 or I think it was code is 5.4. And he was the only saying like, we're using it, but maybe like 30% of my code because it's just very low level. These areas have always been to me a different world than the general big tech like big tech hires these people. They feel a little closer to electrical engineering, hardware engineering. Now that area in general, I observe there's a
big demand. There's a lot more startups. There's a lot more money in hardware tech. So hopefully it will be good. And I also believe that knowing the basics, like knowing if you can code in C++ assembly, like I think that's really useful knowledge and you can build on top of that.
Because most people who know high level language, TypeScript, whatever, like most of them will not know how to go down to C++ if you know C++ and you can build high performance low level systems. You can learn easily the rest of the stack. And if you're in this situation, I would just look for those specific offerings in junior positions, either you have pedigree, which makes it easier, which means you're in a good school or you had an internship at a good place, or if you're in school, try to get that pedigree, try to get into an internship program or build some impressive projects either on the side or contribute to open source, which is a still a pretty good way to stand out, especially with AI contributions being rejected. You will have to work hard if you want to get a prestigious place and accept a stepping stone as well. Like right now, I think getting a junior age-ob is better than getting no job. And once you have a job, try to excel even if it's a shitty job,
try to be the best there. You'll build up a good network. And at some point, hopefully you'll have a stepping stone, a new opportunity to come in to go to the next level. A few questions about BigTek. So when a company like Metta lays off 10% of the record here and then reassigns another 10% without consent, how does leadership fail to anticipate the obvious heat to culture and morale when everyone inside and outside can see it? Yeah, this is the question, right? The interesting thing,
I talk with Metta inside of some directors and even above, and they see it. So this is not a question of does the leadership not see it? This is a question of does the founder specifically Mark Zuckerberg not see it and why does he not see it? Or if he sees it, why does he not care? And we're now going to territory of like assuming what a specific person is. In the case of Metta, like Metta is the
only one who's done this. No other company that has a career CEO, I'm looking at Uber, I'm looking at Microsoft, I'm looking at Google, they have not done this. Because they probably know what would happen. And they don't want they don't want a part of their business to go down for no reason in terms of outages, losing some of their best people. Because that what's happening right now, Metta is some of the best engineers who up to a few months ago thought, you know, I like Metta always treated me well, we're investing in AI, we might or might not be winning, but it's doing good. Stock is doing good.
I have a good work I've been doing for 10 years. Now, some of them have been reassigned to do this work that they don't want to do like this data labeling. You can make it interesting. I talk with people who are in this organization, this AI, a a deal organization advanced AI that's AI and a radio is a data organization, but they joined and they're making the most of them their ex-engineerless experience. But these people realize like, well, I mean, leadership specifically CEO no longer cares about engineering as a whole. So we can only speculate clearly, it feels like
Metta has had in the past some existential times. One of them was when Google plus launched somewhere in the 2010s and as well documented, there's a book about a chaos, I'm not sure if chaos on case covers it, but it has been really well documented where Metta went full on wartime mode. It was like, Google is coming after us, they want to kill us and everyone worked really hard because everyone understood that the fate of the company was on the line. And my sense is that Mark Zuckerberg probably thinks that this is the case right now for some reason that is not really articulated and under his own necessarily understand. And he probably has his reasons. I don't know why he's not telling
people because this is, Metta is operating a wartime mode except everyone is like, where's the enemy? Like revenue is record high. They're doing amazingly well in the ads business. Their products are growing. And for some reason, it seems existential to Mark Zuckerberg to own AI. But this is where
when you look at the patterns, like the metaverse also looked existential to some extent. And now AI is looking existential. And I think people are starting to ask a question like, okay, can you just pick a lane? And in all fairness, it might be hard for a metal Mark Zuckerberg because Metta still does not own any platform anywhere. They are an application layer still. And I think
he really wants to break out of that. And I think it's just being a bit reactive, essentially, this is all speculation. So I think the easiest thing would be just ask him if you can answer. And bank, a big company specifically Google Amazon, Matt, Microsoft and Apple, how do I feel they're doing on AI adoption and engineering, who is accelerating, who isn't and who is managing transition well? I think Google is trying the most. They have the one where they give a free rein to everyone's
ability to build AI tools. It's a bit chaotic, but people are building a lot of things internally. And they are the only big lab who actually have an AI model with Gemini. And they have a Gemini organization. And there's always talks about how they're doing compared to Open An and Traffic. But they're the only ones who have any sort of competition. In fact, Gemini is the
only product which is actually eating into a Chagibis market share. My editor was selling it. I don't use Chagibiti for my queries. I use Gemini because I like Gemini. And I think he also said that it's free.
Okay, I guess there you go. So in this way, they're actually, I think, way ahead of the others. Metas seems to be bogged down by building and training their own AI. And morale is just going down because people don't really see the point. Microsoft is in this weird place where it's still very political as far as I understand there's organizations. There's the co-pilot, there's the core AI organization.
GitHub is under core AI now. So it's AI there mandate or it's source control. They seem to forget about that. And they're reliably just not so as you're fighting with everyone for capacity. They don't have enough Microsoft is more focused on politics than AI in my assessment. Apple,
I talk with people at Apple, but Apple is very secret event. Amazon is secret because their engineering culture is pretty good. So I'm surprised they're so secretive, but Apple is secret because their engineering culture is absolute trash from all I got is duct tapes everywhere. I'm not sure much is happening in Apple, but because Apple is not doing too much, I personally hope that they will actually see local AI locally running on your hardware because they have a very strong hardware. So one thing I think Apple is doing good is they haven't forgotten about their core business, which is making devices and a software that's decent. It's not great, but it's decent enough that
people don't leave. And maybe that will actually be a winning strategy. Amazon, they also, they're an interesting one. So Amazon is the example to me on how difficult it is to retrofit innovation. Compared to Google, they're trying so hard to have AI everywhere internally. They built a key row, their
internal tool and they have their own models, but they're all subpar. It's all people dragging their feet. They rather use cloth code and Amazon is full of smart people. So to me, Amazon is a good example of just Amazon, how difficult it is to bring it to a large organization. Companies that I think are doing a lot better than all of these companies are, I guess, a little tech, not the big tech, but the public is really small, Uber, ramp, even intercom, block, say, say, say for the layoff, but they're the ones they're building AI, in fact, because they don't have any identity crisis.
All of these Amazon, Microsoft, meta, Google, they're like, look, we need to own the whole stack. We need to build AI model, we need to build the application layer, and then we need to become a platform and Uber and ramp is like, no, we know our place. We want to use these the very best possible way. We will take cloth code, we don't care. We don't know about one of those. We will integrate it as
much as we can inside of us. We will not have a foundation model. We will buy or use the best one. And so they're just focusing on optimizing for their business. So I think they're the ones who are kind of the most ahead in terms of large companies right now. And Tropic and specifically,
cloth code are shipping at extraordinary rate, using agents for implementation, test reviews, incident response and many other things. Is this how AI native development will look like, or is it very extreme environment? And others will be wrong to copy that directly. I think it's just very hard to copy on Tropic. So we cannot deny that on Tropic as the best example for AI native development at scale. Together with potentially the codecs team. And when I say
in Tropic, I actually mostly mean cloth code and also their model, but it's all interwinding it, because in AI lab, their product is, they'll forget, and Tropic is cloth code. It's not cloth code. Cloth code is a revenue generator until cloth is so good. Their product is the model that they get a new version every few months, and they do a bunch of work with training, pre training, post training, and then the tooling around it. Everything is like a beehive all around this one thing. So the only way you could copy it is you become an AI lab. And the product is just the by
product, which right now is doing great, even though on Tropic, for example, don't even have an enterprise sales team that a lot of other vendors would have a media habit. It's almost pretty small right now. But I always feel that they're a bit of anomaly where I'm interested in I'm not seeing all that much yet is startups on how startups are completely changing how they work. And I suspect the reason I'm not seeing it is when I talk with AI native startups who are like, okay, you know, we're founders, we'll use AI for everything. And you start a company, you realize the first hurdle is like, how do you get traction? And at worst,
I mean, like you guys luckily have gotten traction, you kind of passed that point. But a lot of founders, it doesn't matter how AI native you are. If you don't have customers, if you don't have a market segment, if you don't have any of this. And I suspect that I wonder if that's going to be more important, that like, get traction doesn't matter how. And once you have traction, it's a little bit like even pre AI, you could assemble an amazing engineering team and build a first version of a product, or you could just like have like a really bad engineer, but have a really good idea and launch that product. And it takes off. Uberber was a good example where when it took off, Travis
Collins just hired some contractors made an ugly app, but it did something that people wanted. Oh, and he was right place in San Francisco. So I wonder if like, AI native is overrated and like once you have a business model, of course, you can optimize it, but will AI native make all the difference? I'm not sure. And another good example is Coinbase, you know, they're really trying to be a native to all those things, but they're in the end, they're crypto company. The
crypto market goes up, they will do great. And now they've got crypto market just went down. So like, you can be as AI native as you want to maybe you'll be able to do the same with your people. But I'm not as sold on this. Yeah, to me, it feels like artificial, artificially trying to become AI native is a bad strategy, right? Like just saying and dropping is doing that. So we'll
copy it and try to implement what I think works really well is when you're seeing the problem and you understand that, oh, actually, this problem can be solved really well with AI, for example, you know, incident response, right? So why don't we try AI to do a first pass understanding what's happening, right? Like it seems like an obvious idea. And like, if we have problems with incidents and debugging time is taking a lot, we can try and if it's takes them good. But some other process might not work in the company. So it depends if there is a problem and it feels like it can
dissolve his AI, then it's like a good idea to adopt the practice. I wonder if instead of AI native, which is just think about like companies where like AI is a natural tool that you reach for, like you, for anything you try it out and it might or might not work, but you're not precious about it. You use it if it makes sense. And you throw it away. If it doesn't or you'll revisit it later.
Yeah. And just the other tool that can help you. Next one. Can you share something about today's presenting sponsor was like, is this really the question that people are asking? No, this was actually not submitted by anyone. But I still want to talk about it.
Now, I admit this was the one question I sneaked in because I really wanted to share something visually interesting about our presenting sponsor and this is how different their UI is. Let me show you with three examples. We already know that it is as verifies your systems correctness by running your whole system and hostile simulation and finding bugs. Here's a UI for casualty analysis. You can open a report for a bug and see the probability of a bug occurring throughout the timeline of the simulation. In this case, we can see that at virtual time 25, something happened that makes
this bug close to 100% occur. So we can jump into this point in the virtual timeline simulation to read the logs. This kind of bug probably visualization is one that I just not seen before. There's also this neat log explorer. You can filter on error messages and then visualize how common or uncommon the error is over time. For example, here we're looking for failing
linearization failures, the purple line. And you can understand how rare or common a specific failure was. Again, I've yet to see this kind of error visualization and I really like the innovation on the UI here. And finally, the multi-first debugger. You can go back in time and replay a debug timeline.
And you can inject bash commands at any time without affecting the playback of the bug. How cool is that? For example, here we're listing files in the current directory, but as you can imagine, you can debug the whole environment much easier. I really like how the team and instances are pushing what's possible with both debugging and verifying software. Head to instances.com slash pragmatic to learn more.
Is it going to encode quality for speed with AI, worse at long-term? Some engineers do review the plan architecture and code. Others rely on SDD plus harness in these regards to code. Last or short-term, but is it good enough to make up for worse code? This is a big question, isn't it? Like, I wonder if there's like any answer. I feel as
engineers, I think we know what answer we want. We want the answer to be, yes, quality is important, yes, care and craftsmanship is important. This hasn't changed. Even before AI, we wanted this to be true. But when I got inside of Uber, I learned about some horrible hacks that Uber did that looked really painful. For example, the old Uber app before 2016
before we had the rewrite, you would open the Uber app and you would see the ETA of the cars, you saw the products and you could pull the slider and it would show how many minutes the next category would be. Uber Black is two minutes, it's like six minutes and you pull it and you saw some information on the screen. What happened is that app was pulling the server every five seconds to give me all the information. It was a package. Every five seconds you would get an increasingly large data package, but by that time it was a few hundred kilobytes, I believe, that was coming back.
The reason that they did this is terrible. It's inaccurate, it's slow, it's really wasteful on resources. It's also just stupid, honestly. By that time, we should have just pushed this information. But the reason this happened is the back-net team was small and the front and the mobile and the web teams were larger and they were getting frustrated that when everything wanted to change the back and take some information back, it would take days, weeks, months. They asked the back and
team, hey, can we do something about it? They're like, well, there's a really hacky solution where we just send this big blob together and you can go on the back and add whatever you want and it's blob and they're perfect. It actually unblocked Uber for a long time to grow independently, but it was a terrible architecture. This is an example where this is technically tek-dup, but tek-dup can speed you up. I wonder if with AI, this is also true, that should we not look at tek-dup in the stages of a product or company? Early stage, you're looking for an idea, just like, go with tek-dup,
we don't know if it'll work, you'll probably toss it out. There's companies at this stage where we just try to prototype and it doesn't matter if it's beautiful or not. Once you've found product market fit, there's this Ken Beck has the three Xs, the, I think, Explore, Expand, Extend, and there's other ways to say this. But in the Expand phase, you've found product market that you want to scale up, you want to quickly reach a bunch more users and you're kind of okay with hacks at this point to grow faster and the last phase is when you're mature, you want to make things good. And what I've seen at the likes of Uber, again, pre-AI, is when you find product market that you have a bunch of customers, you have a bunch of demand, you will now have enough revenue and money that you can hire people who can help you fix these hacks. So I wonder if it's the same with AI, maybe we're overthinking that
if you're in early stages, you're just doing a prototype, it's cool, and don't worry about the cool quality, which, I might hurt you, if you're at a stage where you're now scaling up, I mean, pay more attention and if you're a stage where it's a mature product, it's actually making money, we don't want to mess it up. I'm looking at Instagram's product, for example, which is a mature one, but I've got to mess it up. That is probably where you want to be very careful and pay attention, understand it. Oh, and final thing is AI doesn't only let us build faster, it all allows a refactor faster. So we have no excuse not to do that every now and then. Yeah, I completely agree. I
think it's basically false dichotomy that it can be only speed or quality, like it's more about segmenting time or in cold based right, so infrastructure, maybe more attention to quality, products, maybe more attention to speed. There hadn't been repeated shapes in AI tooling and best practices. It makes it easier to find exploits and create them in AI jungle. What will it take for the industry to seriously create standards rather than hoping they emerge? Yeah, first AI zone you it keeps changing. I think any standards would make no sense. And I think standards just
naturally emerge. I haven't seen any patterns. So MCP and Tropic, when they're still a small app, they're not a leading lab, they're very small. They created this thing called MCP and I really thought it's kind of makes sense and it comes from a non-threatening place. It's a small lab, which we don't really know. They're kind of cool, but Google was bigger,
open, and I was bigger and then all these large companies adopted it because there was a lot of politics and it's accidental. And Tropic today, they tried to do MCP people were like, no, we don't want to be locked in. So I think they'll just, I'm sorry, I don't have anything like planned happening here. Companies like Entropic have engineering managers coding a lot. And at Metang, I presume a dober as well. The philosophy was actually the other way around.
That the M should mostly focus on people. What's the right approach for engineering managers in AI era? I mean, this is a philosophical question. And people have strong opinions about that. For example, we know from a Twitter, Elon Musk took over Twitter and then Rename is X to fire a bunch of people and he mandated that engineering manager to code while having 20 plus reports, which sounded like pretty insane to do both. I'm not sure there's a right or wrong
model. I've seen all sorts of models work out there. There's pros to both. There's like when an engineering manager does not code, they will care for more about people. They will pay more attention to what is frustrating people at the personal level, at the organization level, and they will try to fix those systems and they'll try to take really good care of people. Injury managers who code,
they will be more in the details. They will be able to give more technical guidance. They will have better technical discussions and they will care a lot less about this first category of things. And they also probably will not have bandwidth to make systems that will change or go to meetings, so for example, work with HR to actually change some policy that makes no sense and upsets a few people or work with a bunch of other other teams to have this new system instead of everyone just duplicating the work. Right now, the industry is definitely going very strong in the direction that managers should be technical. Let's forget about this. People mentioned stuff. I think people need to
unfortunately expect less guidance and support from managers. Managers who love doing this part and are very good at the people part will feel probably underappreciated for a while. I think it'll swing back. I think we've been at the side where we have been very focused on people and has been very rewarded as a manager and it was great to be an engineer at the companies like this. It's now going back where we left. So I wonder if it'll come back again.
At some large companies that you reported on not using AI aggressively is a career risk. How should leaders prevent adoption from becoming a theater? Talking leader boards, mandatory usage, code volume targets, rather than real outcomes. I wonder if this is almost over because there was a part where I talk with CCOs and engineering leaders at all sorts of companies and they're really frustrated saying, oh, man, you're not using AI. But this was before Opus 4.5. This was before,
almost before Opus 4.5 and GPT 5.4 before cloud code was used by many people. This was at the age of auto-complete with like, you know, GPT 4.0 or even worse models. And like, Grindjors are using it or one cursor was just the tab, you know, they had the golden tab key. I think this is almost like an on issue. Like everyone, mostly says I know what users are. And also that's when token leader boards made a lot of sense. Shopify the token leader boards in
that era. No one knows about this about them, but they did it back then and now they kind of deprecated it. So I think it's kind of moot points, especially with these strong models. I assume everyone will use it. And I think it's almost like meaningless to look at it a bit like lines of code made no real sense to look at it for most engineers. What have you done with Persuade to use that
organization, achieve an actual AI productivity gain, rather than jobs more cool and more PRs or more human story. That's a good one, right? Before I answer just like taking a step back, like when I worked at Uber, it was the first company where I joined where kind of like people told me like, don't worry about the revenue. We just care about growth. Like, as long as we grow, we're good. Like, we're just raising more money and then we hire more people and then we grow faster
and we raise our money and we hire more people. And even I remember my manager was telling me like headcount when I became manager. I was like, how does headcount allocation work? It's like, you know, doing a business plan or something. It's like, oh, no, no, no, like, it's kind of a black box here. Like, it's this weird thing where you get a headcount allocation. And if you fill it quickly,
you get some more. And I was like, how does that work? And turns out that because like, an Amsterdam at the time, we could hire quickly, the headcounts were reset at the end of the year. And if you didn't use it, they'd be real ahead within the way. It was a really, really weird time.
It felt off to me. I'm like, surely, like if I hire a person for it and it costs, they cost X, like they should generate at least as much value, right? But they're not right now, like, we don't look at an age like that. Like, this is like different. I always felt that wrong. And so there were
opportunities where I could have worked in a team or led a team, which was a purely platform team with no direct business value. And it was kind of, I'm not sure like, it was a cool technology. It was a team who was building something similar to React Native just internally because we had enough that our needs. And I was like, I'm not sure I see the business use case. So I always stayed on teams where I was very comfortable that we are actually making money. Like, I knew how I
was making money. And I always had this in my mind that if someone asks, like, what would you do if you hire two more people, I would have an answer here so much from a revenue would generate. And if someone asked, what would happen if I took away two of your people or half your team or your whole team, I'd be like, no problem. Here is the business impact. Here is how much revenue would make. And so when it comes to AI productivity, can we really distinguish from business
productivity? I mean, there's only two ways that a business revenue wise can make a difference. And this is just a very capitalist way of thinking about things, of course. But one is either you make increments or you mean the money that you would have not made before. If you wouldn't make that money before, it doesn't matter. Like if you're a crypto exchange and we're making more money
because there's more crypto volume, that's not AI is it, it's the market. But if we launch a new product and it's now making money that we didn't do and AI is helping with that, that's, I guess, the value for AI or cost savings. And I wonder if AI is because use cases just cost savings, which is kind of depressing to me. But the AI companies that are making money, I do see the ones which are selling an AI product, you know, the AI labs are obvious ones are startups. Let's AI is an interview who are making money because of that product. So I think that's a use case.
But otherwise, it's pretty iffy, pretty finicky. And I still have this private thought of like, will AI be a bit more like cloud in the sense that cloud is everywhere now, and including in banks that we will never go on cloud in AWS? But like as a customer, no one cares if you have cloud or not, it used to be as a cost saver, more a more flexible way to control cost. And I think AI, maybe it's a more flexible way to control your own cost or like what people do work. It's a weird thing. But to me, it feels closer to cloud than like to
know, like mobile, which created a whole new market of everything. What is a popular current belief about AI and engineering that you think isn't correct? I think it's incorrect. It's anything that it just makes things easier. If you're using AI and your life is getting a lot easier, like, are you trying hard enough or are you like delegate and stuff? And to me, like I use some of it for my business. And it actually like
makes me think just as hard if not harder work is harder. So I think like believing that AI makes work easier or jobs easier, it's just it's just wrong. How important are degree and university prestige in hiring today? Is computer science becoming a prestige field like law or architecture leading to fewer self-not professionals? Unfortunately, I believe it is. And this is less to do with the degree and what they're teaching,
but more about the market. There was a time around like 2015 to 2020 where you could get hired at a company for a well-paying job by doing a bootcamp, which is like three months to six months, sometimes 12 months, versus a four year or five year degree in computer science. And the reason was there was just a huge shortage. Like all of the people graduating from university were swapped up. That has ended. Majority of companies do not hire from boot camps. Very, very few in pockets,
maybe in the UK or else we do apprenticeships, but they're very small. And the top universities are still getting those graduates are getting hunted down at the likes of MIT, Caltech, Harvard, many others, Waterloo and Canada, Imperial College and the UK and so on. But they're not getting as many competing offers as before. And even the middle of schools, it's just harder. So when it was hard to hire someone with a computer science degree, people went for like lower self-taught and those things. But now they do it less.
I even had someone tell me who is self-taught, worked in university for five years, lost their job about two, I think a year and a half ago that for a year she couldn't find a position, even though she was doing like a sorry work and infrastructure work. And I think in the end, she said that she's either considering changing fields or just doing her own thing. I think it's easier than ever to do your own thing, but companies I think will be more picky. And the value of the degree, it's a bit underrated. If you're living in your current country and you don't plan to leave, like it might matter a bit less. But first of all, large employers often have this requirement just
for filtering saying we need a degree, it just filters out a bunch of non-qualified people saying we need a computer science degree, this filters out the art majors and they don't have to look through as many resumes as they already have too much, even if they have this one thing. But the degree is very important for visas. If you're, for example, in a country and you'd like to move to another country, typically more towards the West, when they like without a degree, it will be very difficult with the immigration system. So like that's something that's worth keeping in mind. That thing can pay dividends even decades later when you're not thinking too much about it.
So a few questions about yourself now. Do you still spend time programming yourself or testing large language models? And if so, what percentage of the time? I spend most of my time researching and writing, but increasing an hour for my business, the primary engineer, I have a backend that manages group subscriptions, some customer support functionality that I'm building. I'm building it myself. And now I might have some folks helping me on my team as well.
But when I could get a staff now, I don't want to get a staff. I just want to build it myself. So it's simpler stuff. Honestly, it's a crud database that it runs on infrastructure like Render. I use the tools. I use codecs. I really code X and GPT 5.5. I also use a clock code as well.
I play with cursor. I sometimes write factory. I try to rotate these tools and it just makes it so much easier for me to get back into it. But I don't spend most of my time on it. And your own workload as a creator, writing, podcasting, and researching, have you seen productivity gains from AI? This is an interesting thing where I think I should have. So I don't use any AI for my
own writing. I did a few of these experiments more for Curiosity saying, hey, here's some notes, generate an article in the voice tone of the private engineer. First of all, it doesn't address his job on it. I don't think it sounds like me. Second of all, it just feels artificial.
Most importantly, I really, really enjoy. I love writing. It's not the thing of writing. It's the thinking. I keep thinking. A lot of times on social media when I will post something
and I guess a bunch of likes or views, it's often I'm just writing and I have this idea when I'm revisiting this topic for the third time and that's an interesting idea. I just post that idea out there and I just go back to writing and then later I see people responsible. I guess what people see is just an original idea that comes. Most of my social media is my byproduct of writing and researching. Most people don't know this. There are so many people who are optimizing social
media for likes or things or all of this thing. But for myself and a bunch of people that I know and respect, it's kind of like your side thing. One good example I read on Hacker News wrote about is that their favorite YouTube creators in photography, this person was a hobby photographer. Their favorite photography creators are not professional YouTube creators about photography. They're photographers who have business and they actually do shots and then they have a YouTube channel where they share every now and that's infrequent. It's not there. And I also think of
myself as my main thing is I research what's happening in the tech industry. I talk with engineers. I try to keep an ear on the ground as much as I can because I talk with and I do this by just being in touch with a bunch of software engineering folks. I know some friends and when I see interesting things, I dig into it. That's for example, how I know that something was really off at meta. I've
only ever sensed things being slightly off at meta for a long time, but now I have 10 or 15 people who I know there for years. And now most of them were sounding the alarm bell. I haven't heard that before. And it turns out I was right about just how bad things I've got in there. But in my workflow, I use it for research when I'm like, here's a topic. I'm going to research RAM,
engineering, culture, deep research on all the platforms. I would have thought that this would have freed up time. And I guess it frees up some of that time, but I would have never spent that much time researching. So I don't feel that I'm working less interesting enough. And what capability do you worry I might weaken in your personal way? For example, coding fluency
or technical recall or writing from blank page? I don't think the writing will suffer because I just don't use it there. I don't even have spell checks on. I just don't like it. Or I know I turned grammarly off as well because I hate when it once to reorganize it. I think it's whenever you over rely on something, it could make it less
efficient. Like for example, one thing I now over rely on is deep research. I want to find all the things on the web. So my ability to find things on the web might be worse, but I'm not too worried about that. Because first of all, it was just crudgy task. Second of all, I don't
retrust the internet that much. In deep research, I still check where it's referenced. I'm not sure this is going to be 100% checked out. But with coding, I now just prompt and write the code in that my ability to write code by hand will probably be degrading. But I don't personally mind that part all that much. So I think it goes back to like, look, whenever using it for a bunch of stuff,
just know that that skill will go down and are you okay with that? I'm kind of okay with it. Has AI ever attempted you to go back to building software? It's not so much easier to build software. It probably would have tempted me, but right now I just love what I do. I actually love the human connection with actually talking to people and getting a window
into what other people are doing. But it is making me build more software and be more ambitious. So there's this project that I've been putting off for a while, which is a self-service signup flow for companies, for the pragmatic engineer, so like the whole company domain. And I'm actually just building it because it's so much easier to get started with. It's less intimidating. Vladimir is an engineer
in banking early sorties and he's worried about staying relevant. So he's trying to create for full CS education, but it's quite scary to give up good paycheck. I feel stretched how should he think about future-proofing his options. What I see in terms of future-proofing is the single best way to improve it is work at a company, which is doing stuff that is very relevant. You know, this is building products, building modern products, building products that incorporate some level of AI, where it's okay to experiment banking when it's a rigid place. It might be the opposite. But my first
advice will be inside the company and you start a project where you are just doing some experience with AI. This is why Google is such a great place right now. I know it might not be too popular to say, but they encourage doing this. Like, oh, you're on your team. You're building a product pool.
And you have a suggestion to like build this new experiment with AI. Yeah, go ahead and do it. And I have a feeling that a lot of companies will be receptive to this because right now there's a bit of like every leader thinks like we should use AI more. And if someone comes up and says like, I have an idea and I'll do it on part time, it's a win-win. Worst case is, you know, you learn about RAG or you learn how to implement this thing. It can be an internal tool. And that's why there's
an explosion, even at larger companies like Uber with internal AI tools. Just start doing that. I think that's the best way to stay relevant because if you take a computer science degree or do it full time, it's able to be behind the industry right now. Also, like you can do it at a great part time, but because it's such a big, it's all just like the best way is to be hands-on. So my advice, we try to do that as part of your job. That's easy. Everything else is harder, leaving for a new place,
interviewing for a new place all harder. Of course, you can try to do side projects, but I find that unless it's something that truly motivates you, like unless you have this thing that you really want to build like this health app that you really want and it doesn't exist and do it. But other than that, it could be easier to do it at work. Why two cents? How can you surround yourself with highly motivated top-notch programmers when your classmates aren't at that level and feels like too much to catch up to? I mean, if your class days are not that motivated and you are trying to find a different
group of friends, well, it depends on where you're at with high school, that you're stuck with them, which is fine that even while I was at high school, there was only two of us who were coding. Luckily, there was another person. Maybe you can find someone from a different class. Maybe on an online community. I've heard some others right real on my podcast. When she was in high school, she joined
online communities and started to build. She actually started to contribute to some software there. So that's about one way to find. If this would be at work, probably to either change teams, if you can internally to move there or outside of your project, like take projects where you can work with other people, like seek out and try to follow those people or get towards them, because a lot of people will be motivated. And also, this is the thing where when you're in that situation, you can change companies that make a difference. When I worked in banking, one of my first jobs, my colleagues were super nice.
They were such nice people, but they were not in love with technology. None of them were. And then when I moved to Skype, everyone was. And it was just such a big difference. So Akash is saying that his son is heading for an IT focused high school dreaming of becoming a gaming developer. What does a pass and the job market looks like in five years from now?
And what should he do to prepare himself? Ever-ending this question, right? If only we knew. I mean, I personally believe that. I tried to draw parallels from other industries because we don't know what's going to happen exactly with AI, this tool that we know that coding is so much easier. It will probably make some of the other
parts of the jobs easier. But I like to think of a parallel, for example, construction, where if you want to build a house today or renovate your house significantly, you could walk into the DIY store or you can go online and you can order a bunch of equipment, including professional equipment. You can get the same equipment as professionals. On YouTube, you have professionals making videos of how to build a wall, renovate a wall, tear down a wall, do that. You could do all of that.
You have the information and you have the tools and you have the materials. You can buy the top-notch materials. It just takes a bit of work. So why do people on construction have a job? Well, I guess most people don't want to do all that and they'd rather hire a professional. So I
think what will happen in the second industry is exactly this where, and of course, more people are calling out a lot of electricians like Change of Lightbulb or even some of the more advanced work. A lot of people are using YouTube and DIY jobs are probably getting way more business. But I think there will be professionals. So if you want to be a professional in a field, there will be a path to that. And to get into that, it will go through universities and education.
I'm fairly certain that the game that will be released in 10 years, which Akwes' son will hopefully be working on, it will be built by a studio that's either a startup or a AAA studio. And if it's a AAA studio, they will hire graduates from some of the top universities, from people who have been building games on the side. And for Akwes' son specifically, I have an episode with John's Tyrolar who builds games and one of his games got a million sales with two of them building it. I would suggest that to watch that episode. But also Johnas, he shared a video of all the games he built over like 10 years or 15 or 20 years, and he has been building games on the side. So if this one wants to become discouraged, I'm just going to start building
games on the side right now. In this hard market, what do you recommend for engineers in the EU? Keep aiming for tier one companies or stay with tier two jobs? Yeah, so this is the in the tri-model structure. I have a tier one, I put it as the local companies, like the local supermarkets, the ones that are really competing for local town tier two as regional and tier three as global, that's the big tech. And in the job market, well, first of all,
when the job market is really volatile and uncertain, saying put can be a good strategy, at the same point, I would not stop looking for opportunities because on one end, the job market feels a bit different than in 2023. 2023 was a brutal market. It was layoffs everywhere, and no one was hiring. Right now, there are some layoffs, but so many companies are hiring. So now could be a great opportunity to jump a tier up, to start up to building products, to having more autonomy, to using more of these AI tools. And if you stick at a company that is
just really moving slowly, you might not have that opportunity. I talked about the engineers who are really in demand. They have a few years of hands-on experience with these tools. They will be in demand in a few years time as well. And if you will still have zero years of that, well, you're sitting in one place. So I would be opportunistic in looking out, maybe looking at job
openings, talking with your network, not ignoring fully recruiters, seeing what's out there. If you get a job offer, you can always say no. If you have no job offers, I mean, you're going to stay at your current place probably anyway. How can engineers and students use AI to learn and explore new technologies and concept better? I think you can do deep research a lot better.
You can ask to explain stuff. But the way I see it, AI only ever help me learn about something. I wanted to learn about something to start with, which you want to learn. It's a tool. It'll help you. But I wouldn't also fully throw away things like books, other resources, maybe videos,
tutorials, and also just building your own thing. That's what I mean. Biggest makes it easier. It's not going to make it easier to learn, especially when you're not motivated. So like, decide what you want to learn. And yeah, it can help you, but just learn it in that case.
Just have one for your excuse when you want to do it. And if you don't want to do it, just don't do it. So not IRS is asking a question. So I guess it's very safe to share all the information. How much do you earn from this? And why start this instead of that job?
The last time I shared specific numbers was, I think, I think in the first year of the publication, where I shared that I had like 2,700 paying customers. And it's going to be on that. It's now more than 10,000 paying customers of the newsletter. I also now have some sponsors in the podcast. And the reason I don't like to talk about the specific money, you know, there's people like here's exactly how much I make is every time I do that, I get so many questions coming in from people like, Oh, I also want to make this much. Can you advise me? Can you have a call with me? Can you
coach me? Can you mentor me? I want to quit my job. I want to do this thing. And first of all, I'm very grateful that it's amazing business. But it's just what I'm good at. Like, I don't want to
give financial advice to people. And I didn't even think this was possible. But actually not like be that like vague. When I left Uber, my conversation was going down a little bit because of the four year vesting. But in my best year at Uber in in the Netherlands, I made, I think it was like something like 288,000 euros back then it was like $300, $300, $300, $300 or something like that. And
120 of that was base salary. I think it was like a 26 or 27 k bonus or maybe 30 k bonus. It was a big cash bonus. And the rest was in equity. And like when I started this, I, I didn't think it would go too far.
I thought I'd give it a shot. But most of why I didn't think it would go to be realistic, like Lenny shared his numbers of 2000 pay subscribers and he was a massive $300,000 roughly give or take. And he was going up. And I thought, well, I mean, maybe I could I get there? Maybe yes, maybe no, but we'll see when we get there. But I in the first week of starting publication, I had a hundred
paying customers, which is like that was $10,000. So that's paid a front, which is okay. That's very nice. In six weeks, I got 2000 pay subscribers. It was still $100 before I raised. And I started to
raise the prices back then. But it was like around $100,000 and then I kept going up. And I started to be on a higher annual run rate in about like, I think four or five months, then my old Uber best total compensation. And it was still going up. And I was like, okay, what's going on? So I just kind
of stopped looking at or thinking too much about the money or these things. I started to focus on just writing that one really good article. I did this for a year and a half, two years, actually. And then I looked up and I was like, well, I actually really love doing this. And actually, I didn't know that you could make more than working at a big tech by doing this thing, your own business. And this is also something that you can realize if you're like with your own business,
you have the potential to make more. And also, you know, one of the reasons you probably left meta as well, where you were probably very highly paid is you have the opportunity with a startup with your own business. I'm very lucky that this has happened. But also one thing, like I love my days. I find it very, very exciting every day what I'm doing. And that is what keeps me doing this.
And I honestly, I just love being in charge. Like right now I'm sitting here because I like to sit here and I having a great time with you. But if I didn't want to, I didn't have to do this, and I I'm too, I do well when I create my own structure, but it really helped me. I don't think I could have done any of this without going through that like 15 years as being a developer, like just doing the way I always try to do the best work that I could. I had a lot of structure. I have a lot of
I mean, a lot of connections to actually help so much with this business. Like a lot of times, my guests are people that I know or I reach out for them to advice. So luckily, I feel almost like, wow, like, was this possible? And I didn't think this was possible. But now I'm just kind of rolling with it. I'm like, yeah, it's great. I love it. I enjoy it. I'm also not too attached to it in the
sense that like, look, if business wouldn't do that well, or people for some reason, you know, they saw being interested is like, well, I can live with it as long as I help some people, I give value to some people. And also, this is an interesting thing, like, I could make more revenue by like juicing it more like I could put more things behind paywall. I bought in feedback from people saying, why did you put so much of this outside of the paywall? And whenever I think something is important and more people should get access to it, I try to not put it behind the paywall, even if it hurts the business. Because again, it's kind of nice to be able to do that. What's next,
and any expansion plans for the probatic engineer? Yes, so the interesting thing is if it was a VC funded company, I would have to expand. But I don't, the only plan I have is I would like to make the pragmatic sum much more regular. There was one in February in San Francisco. There will be one in in the beginning of the year also in San Francisco. And I'd like to get to a point where I can have one
as well. And I'd like to be able to do this on a more regular basis. So ideally, my dream, but like this is more down to logistics and energy and some of those things is to have one in the US research industry. Sometimes some boring industries, like at some point I'd love to go into a utilities company and like go through like how they build software. It sounds pretty boring, but it's very darn important. Have you ever gotten in trouble over an article? Has everyone tried to
sue you? Yes, once. Two articles actually. One I never published because I decided not to publish. I, this will be beginning of the publication. For some reason, I really got upset at Neil
Bangbunk in the Netherlands because I read about their hiring practices. They do intelligence tests, rush to our tests before doing a technical interview and I thought that's going to mess up. And I tweeted about this and a bunch of people who were unhappy at the company wrote to me like, Oh, here's some juicy stories about how terrible this company is. And here's all the things that they do. And here's I have and they had evidence and all that. And it was like,
like, oh, wow, this is like crazy. And so I started writing article about that. This was in the first year of the project. This is December. So I started in August and this was in December. And I had an
article ready that was pretty, pretty damning. It probably read like a hit piece. Like I didn't have any agenda, but it was just like negative, negative, negative. And this and can you imagine this and that I was about to publish it. I even sent it over to the company to Bangbunk and Cinq do because I'm my editor was like, you should probably send this over to them. But then I slept on it and I was thinking,
what am I going to achieve with this? Like at the company inside of Bangbunk, I'm not helping anyone because they'll be defensive and it's actually business. It employs people and it's growing and it's playing more and more people. And then I also got a message from someone who said that they had a bad experience there, but it was also very helpful because this person that came from, I think, Egypt and no company would hire him in the visa on the Netherlands, but Bangbunk did. And they were pushing him really hard and something felt unfair. But it was a stepping stone and that person that works
at Facebook and said it could never happen without Bangbunk. And they took a chance on me. I was thinking like, well, I'm not going to help the company. The article has zero positives. It just says, don't do this, don't do that. And also despite this, they actually have this and I was like,
I'm probably missing something here. And I decided to not publish it because I decided that I want to publish things where I actually like share things at work. Like, and I wasn't sharing any of the things that made Bangbunk work and actually they're now even more successful companies. So, well, and I think this is the thing like every company has this ups and downs. So that was the thing that I did not publish and I didn't get in trouble for that. A bunch of journalists reach
out to me later. So I get all the juicy details because they want to read, but I just deleted the whole thing. The thing that I almost got in trouble for, I was really stressed about is the deep dive on Poland, Poland, the events company who really pissed me off because I was just covering layoffs across the industry. I mentioned I Poland was one of many who did layoffs and I knew people there who left Twitter and the luveru and some good companies to work at Poland because it was a good company, good salary, flexible perks. And I just briefly mentioned them in my article saying like, they did layoffs, it was poorly handled. On all hands, someone brought up saying the
pragmatic engine, I was the only one who mentioned it. The pragmatic engine mentioned that we did layoffs and it was poorly handled. What do you think of it as a C ono C o said, like, ah, this is not like, it's like a BBC or panorama. It's like some small publication they've got to get into the agenda against us and worry about it. It's incorrect anyway. And I was like, and they shared this back
with me and I was like, what? And so the company did not pay employees and they lied about them, they canceled health insurance, it was like, it was loss of lies and unpaid salaries. And I just decided like this thing was like, the guy said, I'm not upon a ramos. I did a proper investigative article where I collected a lot of stuff on how it went on, including a double charging of a payment. That was a double charge. This guy's an outage. They're now reporting out about it from the BBC.
I might or might have not helped with some of that reporting for the BBC, not for my, I couldn't put it in my article because when I sent it over to Poland, they said that this is libelous, this is libelous, this is libelous, meaning they could sue me. And I had to think about like, do I really want to do that? So I actually self censored and I put so much effort into the article, so much stress and I realized that investigative journalism is just not for me. And it's a good read that the BBC later made it made a documentary. I also helped them with that. But I realized this world is not for me.
Other than the book and newsletter, what's something surprising you have found through your writing? Usually just find ideas as they go because they're faster. I also have a long list of things that I collect. Like, I'm not sure if I have any specific things, it trends sometimes pop out a bit more as I'm seeing multiple people talk about them at the same time. For example, there was this, sometimes it just reinforces the things that I'm kind of thinking could happen in January when I started using over the Christmas break clock a lot more and I was really impressed with it. I was
like, wow, this is really good. But it's just me. And I started to read around it and I saw a lot of people saying the same thing. And that actually encouraged me to write the article saying like, I think coding by hand is over. And this was very early on. And I actually got some flack from it
from some people like, how can you say this for an AI show? But I was like, actually, like, I felt this is where it's going based on my experience. And then I got a bunch of evidence and I talked with a few more people. So it needed to reinforce the some opinions I have or it also gives you new ideas. Do you plan any additional guidebook updated for AI era? And what will you change to better reflect the
LMI era? Right now, this book is surprisingly durable for AI because it didn't contain too much about coding to start with. But the non-technical parts, things like understand the business, think about software architecture, those are more relevant, but at the lower levels, at some point, I'll probably update it. But I think I want to like wait until we figure out like how, like what are practices actually work? Like when we'll have like so-called best practices for certain companies, I think it'll take them a while. But I'll probably be visited at that point.
What's your favorite technical book? So I'll give you two. It's one of the philosophy of software design. I just love this book. It's still today's the only book that actually compares architecture approaches between like groups of students and what we can learn from that. I wonder with AI if we
can not replicate this, like have agents like build different software, but it still wouldn't be the same. But it's just a really nice written book. I really like the ideas of modules, shallow modules, deep modules, and so on. And then I also enjoyed Ken Beck's tiny first book. It's a really thin book, but I just like how Chris, every single idea is, even though like that book might be a bit less relevant when you're writing a bit less code, but I just like the thinking that's behind it.
Besides craft, what are some of your favorite software tech products? I really like granola. Or for meetings, it not only takes all your notes, it fills out your notes. And it's just such a delightful example of what like an AI added product could be. Like I'm happy to pay for that because I get more value and it's easier not taking less issues with it not having to think about that. I wish actually that I could see like more products that are like that. And I also,
I still really enjoy perplexity search functionality, especially the deep research. Every product has rolled out deep research, but perplexity is still the one that seems to be the fastest. It's like it's, which was what Google would do for search. And again, it's something that I pay for and I have like no affiliation for it. And this is specifically a search I don't like their new push for like computer or any of that stuff. But like again, like from the beginning, like I feel there's some things where like
AI can really add just a new experience. I'm like, Oh, I didn't know this could exist. Forget what changes. What's one thing about software engineering that you're about will be the same in five years? I think there will be just as a big, big demand. I hope a bigger demand for
professionals who care about the craft and who are true professionals. And in the sense, professionals that you, you know where the industry is at, you know what the tools are, you've used them, you use most of them, you know what their trade offs are, you have no ego and you just choose the right one for the right job. And right now today, this will involve like, okay, what kind of tools I use to write code with? How do I test it? How do I deploy it? How do I verify
the correctness of the system? And as a professional, you care about the things that the average person would not like if, if I'm a building architect, I'm not one, but I would imagine that when I look at a building, I see all the things that as a pedestrian, I don't really care about them. Like, Oh, it's beautiful glass windows and your, and the darkness probably thinking how it holds up, what kind of characteristics, what about earthquakes, what about this, what about that. And I think that having a software professionals who can look at that would suffer work with it and change it, be unafraid of changing it with high confidence because we have the tools that the tools, you know, sometimes again, with buildings, you sometimes you put a scaffolding to make some changes. Sometimes you don't need to, you just like, do a quick job. I think that will be a lot more in demand. And I hope
that we'll have more people who care about this and AI is not going to scare them away, or maybe I just scared the way people who never really cared about a software they just always cared about, you know, like making a quick buck and like just it, but it was never about the industry. Yeah, so these are all the questions. Thanks, Gary for the very interesting conversations. We appreciate it. Thank you. It's a bit weird to sit there because usually that's my line that
you just said, but gigs, this was awesome. Thanks so much. Thank you. And thanks everyone, of course, who submitted questions? Well, this was a different format. And finally, it was nice to not be the
one asking the questions for once. Leave a comment to let me know how you like this one. Thanks and see you in the next one where we're going to return to usual setup.