The Changelog: Software Development, Open Source - Storytime with Steve Yegge (Interview)

Episode Date: July 20, 2023

This week it's storytime with Steve Yegge! Steve came out of retirement to join Sourcegraph as Head of Engineering. Their next frontier is Cody, their AI coding assistant that answers code questions and writes code for you by reading your entire codebase and the code graph. But, we really spent a lot of time talking with Steve about his time at Amazon, Google, and Grab. Ok, it's storytime!

What's up? Welcome back this week on The Change Law. We're talking to Steve Yeagey. Story time with Steve Yeagey, as a matter of fact. Steve has been at Google, Amazon, and now he's at Sourcegraph focusing on the future of their coding assistant called Cody. Steve shares stories of making Jeff Bezos giggle, how Jeff Bezos also is able to freeze people, the possible outlook for Google.

We talk about sociopathic leaders and how Cody is cheating as a developer. A big thank you to our friends and our partners at Fastly and Fly. Our podcast got you fast because Fastly is super fast all over the world. Check them out at Fastly.com. And our friends at Fly help us put our app in our database across the globe right in front of our users. And they'll do it for you, too, with no ops. Check them out at Fly.io. So I'm here with Ian Withrow, VP of Product Management at Sentry.

So Ian, you've got a developer-first application monitoring platform. It shows you what's slowed down to the line of code. That's very developer-friendly. And it's making performance monitoring actionable. What are you all doing that's new? What's novel there? Traditionally, in errors, what's the strength of Sentry is we've taken not a stream of errors and said, hey, go look at this, like all these error codes are flowing into sets. We actually look at

them, we try and fingerprint them and say, hey, we've actually grouped all these things. And then we give you everything you need within Sentry to go and solve that error and close that out. And that's, I think, driven tons of value for our users. And traditionally, if you look at performance, it's not that thing. It's looking at certain golden signals, setting up lots of alerts, maintaining those alerts, grooming those alerts, and then detecting them. And then maybe you have a war room and you try and look at traces, or maybe you realize, oh, it's this engineering team that owns it. Maybe they'll look at logs,

whatever they have available. Performance is very rotated on detection and then isolating to where the problem may exist. And root causing is often an exercise left to the user. Good performance products provide a lot of context and details that an experienced engineer or DevOps professional can kind of parse and make sense of and try and get to a hypothesis of what went wrong.

But it's not like that century error experience where it's like, here's the stack trace, here's all the tags. Oh, we see it's like this particular segment of code, and Ian did the commit that changed that code, and do you want to fire your issue and assign it to Ian? It's not like that crisp, kind of tight workflow that we have. This is breadcrumbs. Right. And we said, hey, maybe there's no reason why we couldn't do this for performance. Let's try. Okay. So you took a swing. You tried.

Describe to me how that trial works. If I go to my dashboard now and I enable APM on my application, what are the steps? Largely because we kind of encourage you to go and set up transaction information when you set up Sentry. You probably, as a user, probably don't need to do much. But if you skip that step, you do need to configure to send that data in your SDK. And what happens is we start now looking at that information. And then when we see a, what we call a performance issue, we fingerprint that, and we put that into your issues feed, which is already where you're looking for error issues. Right. It's not a separate inbox. This is the same inbox.

The same inbox. Yeah. Now we obviously give logical filters. And if you just want to look at those, we do that. And for newer users, sometimes we detect, hey, you've probably never seen this before. We do things because we know we build for math market that bring your attention to it. But it's the same workflow you have for errors today. So you don't have to learn something new to take advantage of these things. Very cool. So if your team is looking for a developer-first APM tool to use, check out Sentry.

Use our code to get six months of the team plan for free. Use the code changelogmedia. Yes, changelogmedia, six months free of the team plan. Check them out at sentry.io. Again, sentry.io. That's S-E-N-T-R-Y dot I-O. so we're joined by Steve Yegge. Steve, you've had such a career. You're a ranter, a blogger.

You upset people. You move along when it gets conservative and not innovative. Where is the best place to begin for you? Should we just go back as far as we can? What's the fun part for you? Sure, sure. We can start anywhere. Go back to Amazon or my early days in the 90s at GeoWorks, right?

An assembly language or wherever you like. Where do you think, I'm sure that your entire history informs today for you, but where do you think things began for you in learning as far back in your history that sort of informs most of what's happening today? Well, so, you know, I've gone through a couple of like phases in my career where the learning was accelerated for one reason or another. You know how you go, you know, sometimes you go and things are just kind of, you're just getting stuff done, but you're not really learning anything, right? You're just executing. And then you go through these periods where you're just faced with some incredibly impossible challenge. And that's where you learn, go through these periods where you're just faced with some incredibly impossible challenge.

And that's where you learn. Right? You learn when you're being, like, you're in the fire. You're in the fire and you're screaming and you're like, I'm learning. I'm learning. I'm learning. Right? I mean, seriously.

You can feel it when you're learning. And one of the early ones, it's embarrassing that I call it an early one, but it was almost 30 years ago. I was writing a computer game and I was trying to make a massively multiplayer. I had a big vision for it. And I was trying to do it kind of myself. And that's where you learn a lot, right? You're a founder, entrepreneur, startup. And it's where I learned that I needed better dev tools.

It's where I learned I needed better languages, better frameworks. And I realized everything was like a dumpster fire. So that's one example. But I mean, there were others. I mean, obviously, Amazon was an unbelievable learning experience, working for Uncle Jeff. Uncle Jeff. Uncle Jeff.

I thought it was Dread Pirate Bezos. Dread Pirate Bezos. Yeah, he is. That's hilarious. Kind of defies description. You know how they say Jobs had a reality distortion field? I never liked Steve Jobs, but everybody said that he'd just come into the room and bend reality, right? And Bezos would do that, too.

He would just sort of bend everything to his vision, right? And his vision was just insane, you know, to his vision. Right. And his vision was like just insane. And none of us believed it. I mean, we all believed in it at some level, but when he kept walking around after four years, you know, or five years and we're, we're super successful and we've got all these product lines and he's like, it's still day one. And we're like, come on, Jeff. Isn't it day two? This is back in like 2003 when our stock was like $200. Yeah. So yeah, he was right. We were wrong. So did he have the vision for AWS? It seems like he did the services and it seems like he kind of dictated everything's going to be a

service and this turned into AWS. Is that how it worked or am I reading it wrong? So I had an insider tell me after I wrote my platform rant, okay, because it got a lot of attention at Amazon. And there was a lot of truth to what I said, you know, certainly Jeff, we did have extra compute power and off cycles and there were other reasons to do it. But the insider told me, one of the engineers said that basically Jeff was always afraid the company was going to die and that he was going to have to do a DE shot. Remember Jeff came from DE shot, I believe, before he went to Amazon. He was a Wall Street guy, one of those types that buys at companies and chops them into pieces and sells them off.

Okay. And so he designed Amazon to be choppable into pieces. Okay. And the way that you do that is you make everybody basically an autonomous business unit. Every team is completely, you know, sort of self-contained and they have their own API boundaries and they're almost, they're basically replaceable. And that led to AWS actually, in addition to the technological advancements

and the compute power and the other factors. Yeah. One thing that you wrote recently for Sourcegraph on the cheating is all you need is you wrote, did I ever tell you about the time AWS was just a demo on some engineer's laptop? No, did I ever tell you about the time AWS was just a demo

on some engineer's laptop? No, you haven't told me about that time. I mean, you tell it in a paragraph there, but you want to tell that story and flesh it out for us. I mean, it had to be, you know, everything has meager beginnings, but did you know when you saw that demo or when the people saw it was like, okay, this is going to be huge or was like, oh, cool. Let's try it. We thought it was weird. Well, it was so weird about it yeah like why would you why would you make an art a perfectly good rpc call like a high performance binary call over some fat text protocol right right it didn't like look we had we had performance drilled into us amazon was always the victim of scaling because we were always growing so fast.

So we were always crushed under the load and it was like, hey, let's try this slow protocol for our PCs. Huh, guys? And it really kind of went over like a lead balloon, right, at first.

We were like, huh, you know, that's weird. But, you know, as we started to sort of like understand what it was capable of and that was back

during the soap versus rest wars. Do you remember that? I do, yes. Rest was really cool but everyone was like, no, soap, because we're all stuck in our old object i was gonna say why did anybody like soap most people were like ah soap no but every for there was a time when soap was the thing all right so strong man soap like why was it cool give it the best argument for why

soap is like was good it was familiar right it didn't require you to learn anything new it was just it was sort of like corba or any of these other ORM layer or whatever they were, right? You know, these PubSub sort of systems. Yeah. And it was object oriented and that's kind of what everybody was doing back then. And it was XML, right? I mean, it was a bunch of XML. I'm surprised they didn't have XML and inheritance. Yeah. Okay. So people were liking soap. I think, you know, it was familiar. That's why we stick with a lot of things, right? Because that's what we know. And rest was the new hotness. But this text protocol, not only unfamiliar, but not performant with regards to the things that you're saying. So like, what were the virtues? It was weird, but it was weird good, obviously. What

were the virtues of this change, you know, in retrospect that turned into what it is? Well, I mean, for starters, and I've been saying this for 35 years, and I will continue saying it until I'm dead. All right. And I'll keep saying it. I'll come back from the grave. Oh, my goodness. And I'll haunt you with it. You'll run a will.

Yeah, which is that text is king. Okay. And binary formats are wonderful and have their purpose. And, you know, we use them all the time. But text is the ultimately flexible ultimately debuggable human readable it's a it's a currency you know and we're seeing that even more now with large language models right yeah but even back then there were there was two schools of thought one was performance is everything and you know put the burden on the

backs of the engineers to figure out how those bits work. Right. I mean, seriously, we had people at Amazon. I'm sorry. You guys are going to get me ranting. No, that's what we're trying to do. Yeah. I mean, this is our goal.

No, I'm saving it, but a little. I'm sure I'll really get you in a bit. But I mean, there was a, there were guys at Amazon who were like, don't ever use regular expressions because they're slow. And then they would like, they would write a, you know, 2,500 lines C++, you know, crappy parser that was actually slower, you know, and, or, or it didn't matter. I mean, like they had wasted all this time on this, this giant, you know, ball of legacy tech debt, right. When they could have just used a regex, but it was just, it was just, uh, anyway, yeah. The back camp of the bits camp and you know come on i mean like i you know i i resonated with them i worked at a company geo works where we did nothing but 8086 assembly language 8086 not 8386 not 80x86 with their their fancy 32-bit registers no okay we were working with like you know 16-bit

registers and and this awful just you know intel architecture we did this for five years because you know we were like performance is king all right and guess what we died because performance was not king it's not it's important but not king it's not king. It's not. It's important, but not king. It's not king. This should inform us, though, that we look at things like SOAP or REST and this argument, gRPC, you know, over a binary protocol versus a text protocol. You can look at that, too, and juxtapose that to now, where people are nacing, you know, the AI direction we're taking. Basically just pushing down future. Let's stay with the past, was familiar,

and not embrace the future, which is just comfortable in the moment, but long-term short-sighted. A hundred percent. I even saw a comment on Hacker News in a discussion about LLMs. It might've been actually after my first cheating post. And somebody said,

I can't wait for all this LLM stuff to just go away so we can all just go back to doing our jobs. I mean, it was almost like I couldn't have written a better trollLM stuff to just go away so we can all just go back to doing our jobs. I mean, it was almost like I couldn't have written a better troll comment if I had tried, right? It was like, what is he? But seriously, there's people who like they have their heads in the sand and they're waiting for it to all just go away, which is pretty funny because I don't think it's going anywhere. No.

Well, it's going somewhere, but away is not the direction that it's going. Yeah, it's amazing how disparate the reactions are to it. We just had a, we'd play a silly game on our jazz party podcast where we, it's like family feud. It's called front end feud. And so we surveyed the audience and we asked them how they feel about things

and what their favorite and least favorite programming languages are. And we asked them recently, this was back in November. So pre like big change, but it was GitHub Copilot. Like what are your feels about GitHub Copilot? And the answers are all over the board from like excited, surprised, can't wait to like terrible, doomer, like it goes so broad. And then everybody, and there's people all along the spectrum in the middle about what we're thinking about this. Not very many technology advancements have, I guess, that drastic of a reaction. I wonder if this AWS stuff inside of Amazon had the spectrum of support that we see for LLMs.

You're also around when Kubernetes was first being worked on inside of Google. These major shifts, like did AWS have so many different opinions about it inside Amazon? Or was it kind of like, eh, it's weird. But then eventually it's like, yeah, this is the bee's knees. An interesting question. Interesting question. You know, I wish that I could answer that, but AWS didn't really get kicked off until

sort of when I left Amazon in 2005. So I'm actually older than AWS. Wow. And that's funny because for a lot of people, AWS is, of course, part of the fabric of the universe. But I can tell you that at the time that I left, you know, I mean, Amazon was, of course, a company that was just absolutely filled with opinions,

but there was only one opinion that really... Okay. He guesses whose it was. Right. This story you told that was pretty interesting. Speaking about Jeff and I guess a semi dovetail from AWS was that a good day with Jeff. That post I was really enjoying. I didn't get to read every single note.

And I was like looking for the bad guy moments. But one, you're a great writer. So kudos to your ability to write so well. For sure. But then this story to like how you all the details of how you got to this meeting and it's back in like the boathouse area and all the idiosyncrasies of that. Can you share just like a little bit of that? Some of that story?

Oh, tell the story? Not the whole thing, but like a version of like this first meeting Jeff and this figure that we all know so well, but you've met in person and got to sort of see that opinion live. I'll tell you a story I've never written down before. You guys can get the poop. How's that? Nice.

Yeah, we'll take it. That story was cool, but you can read about it. It was just basically, we all went to Jeff's boathouse to talk about AI back in like 03 and it was premature, but it was a pretty fun story,

you know, go read it. But there was another meeting with Jeff that I, it's a story I do love to tell. And I've never had the opportunity to write it down. It was my first meeting with Jeff, actually. And I was a technical program manager. It was in my first year at Amazon. So this would have been 99 because I joined in late 98.

And my boss was Kim Rackmiller. She was the, you know, the group program manager. She's the best program manager on earth in history, actually. She taught me the art of being a TPM. And then we had Jeff Wilkie with us, right? Who's one of the best leaders that I've ever worked for. Two amazing leaders, right? Me and Jeff. And I'm just like, and they told me to shut up they were like steve you know don't say anything at all unless he addresses you all right scary you know stay on your knees keep your head forehead to the ground all that stuff right spoken spoken to yeah so we're like we're we're in this like tiny room in the you know this transylvania looking castle building that we were in the pac-med medical building in beaconacon Hill, Seattle, right? You know, it's a big, big, intimidating looking, and he's in the top corner wing in this corner office, real, real small office, you know, corner window view of

the sound. And we're sitting in there like at this tiny table and Jeff's back is to the, to the windows and right. And, and we're pitching him on my project, right? Which is one that he asked for. It was a, it was a team, what they call a two-pizza team, right? Jeff came up with these two-pizza teams, these tiger teams, basically, right? And it was for reducing customer contacts. And he wanted fitness functions. Now, this is a great way to run teams, having an objective function, right? I mean, really, seriously, if you can find a function that you can drive up into the right, then it's a great way to run a team. But finding the right objective function is a really challenging problem for some problem domains. But for mine, it was really simple. It was like, we're trying lot of people had been presenting Jeff with these really complicated mathematical functions of many, many terms, weighted terms, right?

That are going to be this is how their team's performance is tracked. And he would argue with them and throw it out and curse and scream and whatever. And right, not scream and curse. He was always very quiet. But you would feel him cursing and screaming at you like somehow in your mind. I don't know. It was very intimidating. And it was mostly because he was so smart that he would just pick holes in everything that you showed him right so we knew

we knew that we were going to bring this to him and he wasn't going to lie there was going to be something wrong but they were like steve don't talk don't talk okay we got this okay so jeff listens to the pitch and he asks one question and he put them into a state of complete paralysis have you ever seen when you draw a line in front of a chicken? Oh, my gosh. Yes. Yeah, and it thinks it's a snake, and it just freezes, right? So they both, they both of them, sorry, Jeff and Ken, amazing readers.

So you have a chicken analogy here. They froze deer in the headlights because Jeff asked, he asked something that nobody had ever heard of before. He just, out of the blue, he goes, every fitness function has a yin and a yang. What are they for this one? And they, they, they literally turned to stone. It was the wildest thing I've ever seen. And, and, and I'm looking at him and I'm looking at Jeff and I'm like, you know, and I'll see,

you know, my, my, all, all the algorithms don't talk, you know, that, but they weren't saying anything. They were just sitting there like, right now, at least like five to seven really uncomfortable seconds past maybe 10. I mean, it was like long enough that I finally said, I know what the answer is. Okay. And they all looked at me and I swear to god i swear to you okay they looked at me like they were drowning and i had thrown them away okay like the exact same thank you for saving me right and jeff of course kind of slowly turns to me like oh it talks right and now now all the pressure's on me and of course i like had had a total of five seconds to assess the situation, parse his question, come frame an answer.

And so now I'm on the spot, right? First time I've ever talked to Jeff. And I told him, I said, look, we could have done some fancy, complicated function here, right? That will track it more accurately. The problem with this function is if we just say we're reducing contacts, then we can just route you to an automated system that takes care of your, you got a free replacement issue. We'll replace it for you.

Just self-service the contact goes away. It's not counted by finance because no customer service person ever talked to it. That would be the wrong thing to do because we're not fixing the root cause. If we just give them a new sweater, then we haven't fixed the problem at the upstream, the distribution center or the supplier or whatever that caused them to get the wrong sweater in the first place and he started nodding and he goes yeah that's it right you know i said the yin is oh my gosh right and you could

see them like defrost and the conversation started to proceed again and boy it was wild though it was wild he could he could take the most seasoned brilliant leaders and just and freeze them huh wow well congrats on having an answer so i right not being one of those chickens but uh i'm so dense i may not even follow the yin and the yang so the yin is that this it's really simple but the yang is like well we can game it is that what you're saying like it can be gamed without actually fixing the problem is that what you're saying? Like it can be gamed without actually fixing the problem. Is that what you're saying? Well, I mean,

I was, I'm over here on the answer. I'm trying to understand. What's the logic of the question? Yeah. That is right. That, that was approximately the level of,

like, I think he might've winged the question. I don't know. That sounds like a question that's like made up. The way I would interpret that really is the way that yin and yang works is that it's two fish in perfect harmony because they're chasing each other's tail and they can never catch each other.

So they coexist forever in their pursuit of their pursuit. And so the same here with, if you're just trying to reduce contacts with customers, that's the one, the yin or the yang, pick one basically.

And the yang is the upstream of how did the problem exist in the first place? And so that's what he was asking. Well, that's really beautiful, Adam. I wish you had been more. Well, I had more than five seconds, Steve. So I can assimilate that.

And none of the pressure of Jeff Bezos staring at you. I mean, that's just got to- I'm sorry. I have to do this, Jared. I have to do this. Ring the bell. My answer really comes from watching Silicon Valley. Yeah.

Yeah. So I learned more about the term yin and yang. That's a joke. Missed it. Yin and yang. A lot of these guys come in here and they can do all the engineering stuff and they get all hung up on technicalities. They can't just tell you what their vision for the company is.

It's like you need both halves of the brain, right? The jobs and the Wozniak, the Ying and the Yang. Oh, I think it's Yin. Yin? Like Yin and Yan? No, like Yin and Yang. No, it's Ying and Yang, they're opposite. So Pied Piper, drop it on me.

What is it? Because of Silicon Valley, that was explained in that TV show. Not to that detail, but the concept of yin and yang, which I knew already, but it was like just explaining to him like I'm five explained in the TV show. So there you go. Well, I just want that power to just be able to make people freeze and, you know, crap themselves in their desire to please me with some sort of an answer. I mean, that would be a superpower that I guess would be kind of fun to have. Don't you think, Steve?

I certainly saw it wielded a lot. It was wielded frequently. I'll tell you another story. I mean, seriously, every time I met with Jeff, it was kind of like an interesting experience. For example, did you know that he had a survival keychain? No. In case he ever got trapped key chain? No, no.

In case he ever got trapped under earthquake rubble or something. He had some really fancy earthquake rubble. Yeah. Tell me more. What's a, what is a survival key chain? It's kind of like, I don't think he ever showed it to me. I just heard rumors about it. Okay.

And he took one photo every day, which was like a big deal back in the nineties when, like, I don't know, where would you put him? I don't know. People thought it was, there was a time when people actually thought that was like a big deal. Yeah. But the story I was going to tell was I had to do a presentation to him. It was called the Fundamental 50. It was a proposal that I had made to make our engineers

kind of more broadly generalists, which is something Jeff always wanted. Okay, Jeff's dream was that engineers were just chess pieces, not even chess pieces because those are different. Checkers. Checkers pieces, right. That he could just move around, right. You know, fungibly from project to project because that's, that's the easiest way to plan, but of course never worked out that way in practice. Right. So anyway, I was trying to level everybody up with this, this sort of effectively a training program and, and, and they liked it enough that they, they decided to put me in front of Bezos. And so I'm sitting there, but this time it was in a larger conference room and I had about 10 or 12 of his lieutenants sitting around the table, right? All

his cabinet basically. And I don't want to throw anybody on the bus here, but I was a little disappointed because I gave my entire presentation. All right. And Jeff was very engaged. He was sitting right next to me, very animated. Right. And I could, the way you give presentations to Jeff is you write a five page presentation because he doesn't like PowerPoint. You write five pages of text and then you randomly remove like 30% of the paragraphs.

All right. To let him fill in the gaps. Right. Is that right? To make it harder to read? Do you like put the blanks there? Or do, is it like there should have been a paragraph no no you just like literally like skip stuff and just cut stuff and let him let him make the leaps because like if you're not it's like like fran's list he used

to be so good at sight reading that he would like on the first the only his first read through he would play it correctly and after that he'd start embellishing right and jeff's mind is just like think about it the dude he's smart to begin, but he also sits in meetings all day long and basically gets opinions and perspectives. And you do that for a few years and all of a sudden you've got a general's view that nobody else has. And so they can spot things. And so we prepped and prepped and prepped and prepped for the stupid presentation. Right. And, you know, we were excited about it, you know, but I went to probably 12 principals and engineering directors and said, vet this for me. It's going to Jeff. We all vetted it.

Right. And I showed it to Jeff and he looked at the list and said, so where is machine learning? And I bust out laughing. I couldn't help it. I laughed off my head off. And everyone stared at me in complete shock. Okay. His jury of cabinet members, right, all these VPs and senior VPs that were sitting in complete shock. Okay. His jury of cabinet members, right?

All these VPs and senior VPs that were sitting in this room. Okay. They refused to give any facial expression until Jeff gave a facial. It was disgusting. They want to be on the right side of history of Jeff's history. And they were, they did not want to commit to liking my proposal or not liking it until Jeff had committed, which was the biggest pile of bull that I've ever seen.

So to your superpower, right, the one you wanted, which is make people freeze themselves trying to please you, you better be right. Yeah. The thing is, Jeff was usually right. So that was good, right? That's a good combo of powers. Yeah, but if you're wrong, then you're just going to be like super wrong. Who's going to tell you? And so this group finally, like I laughed and I said, you got me. I said, 12 principals and directs vetted this list and we forgot machine learning and you,

you got me straight up. And I, and I congratulated him and he like, he went, ha ha ha. And then we moved on. It was no big deal. Right. I said, I'll get it into the list. And everybody in the room relaxed and started smiling and clapping me and you know clapping on the back after the meeting come up to me going i've been in a lot of music with jeff and they never go that well and you know whatever right and it was like yeah thanks for the support before he smiled This episode is brought to you by our friends at Drada. Automate and accelerate your SOC 2 compliance, your ISO 27001 compliance, and many, many more compliance frameworks. With a suite of more than 75 integrations, Drada easily integrates with your tech stack through applications

such as AWS, Azure, GitHub, Okta, and CloudFlare, and countless security professionals from companies including Lemonade, Notion, and Fivetran have shared how crucial it has been to have Drada as a trusted partner in the compliance process. They have deep native integrations that provide instant visibility into a security program and continuous monitoring to ensure compliance is always met. see all their controls, easily map them to SOC 2, ISO 27001, and many other frameworks to gain immediate insight into framework overlap. They are the only player in the industry to build on a private database architecture from day one, meaning your data can never be accessed by anyone outside your organization.

It is time to say goodbye to manual evidence collection and hello to automated compliance by visiting Drada.com slash partner slash changelog. That's Drada.com slash partner slash changelog. They are bringing automation to compliance at Drada speed. KOMMENTARER How much do you attribute to your career success your ability to think on your feet and respond in the moment? Because it sounds like it was pretty key. You're pretty good at it so far. Two for two. Yeah, I'd say it's a pretty important survival skill, right?

Yeah. Especially when things get heated. I mean, it's a leadership thing, right? I mean, like, you're always going to wind up in situations where two people have pretty heated opinions that are diverging or not aligned. And you've got to be able to navigate those things and that solution space really fast, really fast. So, yeah, quick thinking is good. What's your analogy to that? Do you think it's like, be like water? Navigate those things and that solution space really fast, really fast. So yeah, quick thinking is good. What's your analogy to that?

Do you think it's like, be like water? Because water takes on any form its container is and can go anywhere, really, if necessary. I suppose heat might change its ability to travel because it changes state. But what's your analogy to advice for how to do what you do? Oh, how to think on your feet? Yeah. Like be like water. I don't know. Is that something I'm trying to, trying to fancy a Steve? Is that, is that accurate? No, it's fire. Put your feet in the fire. And then say I'm learning. Right.

See, I expected Steve to have a good answer to that one. I've been in a lot of fires. Yeah. Well, you, you wrote that they call you the closer at Amazon. Maybe this lines up with that because you close sales. What was the idea with the closer? No, I closed interview candidates. So we would frequently get candidates who had really fat offers from Microsoft or Google or whoever. I see. And Amazon didn't pay well at all and probably still does. Right. Because Jeff is cheap. And so we'd be dealing with, you know, super, super big comp packages. Gotcha. And honestly, like, you know, the perks too, right? Amazon didn't have any perks. Your perk was you get to work at Amazon and pay for parking. Right. I mean, it was like, it was bad.

So how would you close them then? I mean, it sounds like a terrible comparison. That's the thing is I got every single one of them to come to Amazon, every single one for, for my, my whole. And then when I went to Google, I was a great closer too. And I had a, I had a track record that went for, I think eight years of complete success, like 50 to 70 people. I had closed complete rockers. Yeah. Batting a thousand. And then, wow. And then one day somebody said, no, nah, this is what happened to Amazon too. Did you just like try and pay them off? Like, listen, I've got a perfect record. You're ruining my perfect record. somebody said no nah this is what happened to amazon too did you just like try and pay him off like listen i've got a perfect record you're ruining my perfect record what's the situation here no no i was just like i was like why why like you know like there was one dude i was trying to

get there was he was really good and i went golfing with him right him and him and our mutual buddy and i tried to and i and i even had a deal with him that if I beat him on this one hole then he had to come work for me and uh dang it was a par three and I got like two feet from the pin and he got foot and a half from the pin it was nuts so he went to Facebook uh but uh that only happens in the movie Steve that's not how it works out in real life it was it was why it was why we were both trying really hard I guess he really wanted to go to Facebook. Yeah. I mean, he could have just disagreed. I mean, it's his life, right? But I mean, look, I come up with compelling arguments, right? It's, you know, make a convincing story. You gotta be a, okay, here it is. You gotta be a storyteller. Yeah. Okay. This is something that Jeff believes in, I believe. I'm pretty sure that was in,

I think it was in the Working Backwards backwards book potentially just like the the concept and ability to tell a story because that's kind of i think what they did with like some of the reverse the working backwards with the prime the the prime membership you know was like you got to tell a story you can't just put the press release out there first kind of aspect like tell the story what actually happened all that good stuff because if you just sort of like put some kpis out there and some goals and you achieve it it's like well did we really do it well like how would this actually be to a customer in the end and that's always seemed like a storytelling an important thing for that organization for amazon to storytelling i mean it's important yeah i mean of course right i mean every company should be mean, it's important. Yeah. I mean, of course. Right. I mean, every company should be doing it. It's your, you know, your user, your story for your users. It's your

user story. And, and we often make the mistake of putting something out there without a story, but it's like a lot of cool flashy tech or this or that. And, you know, Jeff usually didn't make that mistake, but I will say, I will say that Jeff was not always right on his first try. Jeff would often like, I've written about this, but it's so important. Like the difference between Jeff and Google is that Jeff would continue trying after he had failed the first time and Google never would. Never, ever. Google was like, oh, it didn't work the first day. Kill it. Right. Because they were going for like basically a shotgun approach trying to get a billion users. Right. It was really interesting. Eric Schmidt was the CEO when I joined. Right. And he was, you know, he's a good CEO. I really liked working for

him. And he was, he was the let a thousand flowers bloom type, right? You know, he wanted us to, you know, sort of be an innovation factory, basically put a bunch of brains in a jar and see what they come up with. There were problems with that approach. One was that, you know, there weren't truly enough of us, you know, and we were also kind of non-representative of the engineering population. So our story was different from the story of the average dev. And so Google wound up building a lot of stuff that was really only good for Google as a result. Jeff would put stuff out there and it would be awful. It would be awful.

We would like, did you ever read about the time the guy got evicted from his apartment because we shipped him a book? No. Doesn't ring a bell? No. Would you like to hear that story? Please tell us. I want to now. So I'm working.

So the most eye-opening thing that you can do is go work the customer service queues at your company. It doesn't matter where you work. That's part of the story because you're going to get the tail end of the story from them. And Bezos was obsessed with this, and he would have a weekly meeting that I was in because i ran customer service tools for two and a half years at amazon and built their arizona system their web-based system we switched from command line tools to a

web-based system yeah and he would review every week what were the what was the histogram of customer contacts right where did these stories end and they were they happy endings or not there was one bucket for happy you know the was positive feedback. And that was usually the lowest, but it was in the top 10. And then the number one was always, where's my stuff, right? Bezos had this idea, which I think every leader and every company should emulate, which is get your people... Grad had this too. It's called go to the ground. Get your people out there using the product, talking to the customers, dealing with the people, feel their pain, feel their pain, okay? Whether it's in the distribution center or the call center

or wherever, right? And so we've got on these field trips and one day, like, and every single time I went out, there was some like crazy emergency, which suggests that they're happening all day long at Amazon. So like, I'm reading this mail from this guy and he's like, dear Amazon, you know, a long time shopper, first time, you know, caller. Just want to let you know that, you know, I ordered a book from you and and it came and I opened it up and it was filled with ants. And they went to my bed and I can't sleep there anymore. And and yeah, I just want to let you know. So there was that that story. And then on the same day, there was – so that one actually led to me walking by a room with a bunch of VPs with the word ants written in giant letters on the board, right? Yeah. The reason for that one was that somebody had been eating food while they were packing the box and food would fall in.

It would sit in the post office and ants would go in the box, right? They put food away. This was the beginning of taking everything away from the workers. It's horrible. But the other thing that happened was this guy goes, I mean, the long story short, man, we had an auth bug with the bank where we went into a retry loop where we would re-auth. Like, you know, an auth, they put a dollar charge or whatever on your card. Right. And then you don't charge until shipping time.

But you auth, and it does actually does actually like freeze up a tiny amount. Right. But if you get stuck in a loop that, and the auth at the time was probably more like three bucks or something. And so this guy got like just enough in our loop that, that he was using his debit card. So he bought a book. He got off the bunch of money that wasn't ever going to get charged, but it froze up the funds in his bank account. And all his checks started bouncing and his rent check bounced and he got evicted.

Okay. Okay, and he called us. And the weirdest thing, guys, is that the people who get shafted the worst, okay, the ones who get really, really, really hit hard by our bugs, they're always the nicest. And the ones who are screaming at us over a nickel

with froth coming out of their mouth, it's just so interesting to watch the dynamic there. But yeah, it was one of the super nice ones. And so, of course, we got the Blackhawk helicopters and flew out to the bank and basically held them. You know, I don't know what. Don't say a gunpoint. You almost said a gunpoint. You thought so close.

And to make sure that they got this dude back into his apartment. So you save the day or proverbially the you, the Amazon you save the day. Well, they also caused the problem. They ruined it in the first place. Yeah, exactly. The key is they did this over and over and over again on purpose because you'd call Amazon and you'd be like, you guys screwed me and I'm mad. And then we would say, oh, we are so, so, so sorry. And we had a pyramid of professional ass kissers that to deal with, you know, journalists and stuff that would

get, you know, screwed by our bugs. And we would say, we're going to, we're going to make it up to you. We're going to give you gifts to keep that one. We'll send you another one and we'll give you a gift certificate and a direct line to blah, blah, blah. Right. And you come away going, they were so nice to me. They really get me. Right. And then you've turned this pissed off customer into a loyal, a brand loyalist for life. Right. It was like, and it was because we had no QA. It was like, what was he thinking? And yet it worked. Right. It was just the most bizarre loyalty creation thing. But he was, he was right by being customer first. It meant that if they had a problem, he would fix it.

And sometimes the problem was pretty terrible. Yeah, I think getting evicted from your apartment because you bought a book is pretty terrible. That's a bad outcome. Ants and a frozen account? I mean, that's a double whammy. Yeah, I mean, we would do simpler stuff. In the early days, the distribution center was just a big hash table, right? Because you have way more SKUs than you have space. And so, you know, you have random items sitting next to each other on shelves.

And so inevitably, statistically, some religious family would get pornography, right? You know, or whatever, because you just sleepy, grab the wrong thing. This is in the early days. So a lot of people got the wrong thing, but sometimes you would get the really wrong thing. But, you know, I mean, like stepping back, I mean, like, you know, the way Bezos ran the company, it's, you know, I've talked about him over the years. He did so many things right. And then he did so many things that I found just personally, just morally objectionable.

You know, it's really kind of a, you know, it's nuanced, right? Just same thing with Bill Gates. I feel like, you know, Bill obviously, you know, did a lot of stuff that even the government sued him and won you know that was naughty he was naughty but you know he spent a lot of a lot of time trying to i guess sort of maybe sort of make up for it and clear his name like you're curing malaria and stuff but the way that the way these folks led their companies you know you kind of have to look at it and you have to say well are there parts that we can take that that led to the success without being jerks? Right. And that's honestly been kind of the crux of my, my leadership struggle. Like, you know, cause leadership is a struggle,

right? Leadership is always, always hard. And it comes down to how can I be as effective as, as Bill and Jeff without being an asshole like Bill or a tyrant like Jeff. Right. Yeah. It's tough because especially when you do it, if you do it publicly or, you know, in the quote unquote town square, you have a lot of, when you speak with nuance and you speak highly of certain aspects of a person like you are, right? Like this was really good for the company, this aspect of him. There's nuance there. Like you said, it's not, you're not just with a broad brush saying, therefore Jeff Bezos is awesome in every way. And I love him and I want to be just like him in every single way. But it's hard to get

that across sometimes. And you can receive criticism as if you're lionizing the man, because you're zooming in on something that was net positive and saying, like, let's emulate this. It doesn't mean that you're, you know, vouching for every aspect. But so many people will just read it in that binary fashion that can be. Yeah, you have to be binary on these decisions, Jared. That's how it is these days. You have to, you know, if somebody's a villain in one way, shape, or form, villain in all the ways. You know, there's no coming back.

We do tend to get that right i mean yeah and we'll the herd herd mentality for sure yeah to what you say though steve about leadership and being hard i think that to the level that bill gates to the level that elon musk jeff bezos steve jobs you pick your well-known fANG organization leader. And at some point in their journey, they transition or already were sociopathic, right? To be in the meetings that Jeff's been with, like these, like you said, seasoned veterans, et cetera, that just crumble in his intimidation factor. You only hit the balls you swing at, right? And to hit that kind of ball, those kind of balls at that level, you have to take a lot

of swings. And I think at some point you skew towards, if you don't have the right grounding, into sociopathic behaviors. And that's kind of what you probably struggle with is like union concerns with Jeff in particular. I've seen you write about and the bad, naughty things you mentioned about Bill Gates. I'm not sure which particular things you're talking about because there's so many things that you could mention, some true and I'm sure some not true. But I just wonder if like at some level, at that level of leadership, you to get to that point, what it takes to get there is somewhat sociopathic behaviors. Well, we can find a counter example, maybe disprove the theory. Like, is there a, down-to-earth, rational, well-balanced person

that's the head of or founded and built a Fortune 50 or Fortune 100? Benioff? Maybe. I don't know. I never met him, but I've heard good things, right? Yeah, maybe. Or the dude that founded Twitter originally, I think. Maybe. I'm not sure. I don't know.

Definitely Benioff, right? I mean, there are examples. Yeah. So it's possible. But it just seems like it's more rare. Benioff is Slack, is that right? Benioff is Salesforce. Salesforce. Okay. I thought so. I was like, Slack.

I mean, unless I'm wrong. Yeah, they own Slack, so you're in the same work there now. Sociopathic? I mean, maybe you certainly have to be thick-skinned yeah right i mean like because empathy you know too much empathy has been my my historic failing and and you really have to kind of dial it back a little bit you have to have empathy but you can't let it ruin you because too much empathy is gonna gonna stall you so so i don't know anything about sociopaths other than that they exist, and I just don't understand how they work.

But I do know- Here's a definition to give some context. All right. And sociopathic doesn't mean you have sociopathy. It just means you have a, it's like an itis. It's like a version of it. It's described as a mental health condition in which a person consistently shows no regard for right and wrong and ignores

the rights and feelings of others, which kind of describes Jeff in some cases, not in all cases. Honestly, that would be a really useful life skill, I guess. Something to strive for. I guess, but I mean, like, seriously, when you mentioned the word that you used that really resonated with me, Adam, was grounded. They're not grounded and that's why they, maybe that's why they become the way they are, right? Because of course, over time, anybody can become jaded to other people's problems if they're not feeling them, right?

Well, if you have a cadre or a jury near you that's well-seasoned veterans that only facially express, when you facially express, those are yes people. And if you surround yourself in no grounding and no people, people that push back on, that really know you and can push back against you, then you're not grounded. You're in your own la-la land that you've made up. And it's easy to skew towards sociopathic because you don't have that grounding of regard of right or wrong or someone else's right because you just are so far truth. That's right. That you can't empathize or react in that way. Right.

So you just naturally hurt them even if you don't want to. And this is why, this is one of the reasons, one of the many reasons that diversity at a company is so important, right. And getting, getting, you know, a diverse set of voices and backgrounds and experiences, you know, to help ground it, you know, to help ground it, you know, because I was talking about Eric Schmidt and his failing with a thousand flowers. We weren't representative because we were all super homogenous, you know, Google engineers. Right. And, you know, it was, it was actually a lack of diversity. It was holding Google back from,

from innovating, I think in the early days, right. We had a few minor wins like Orgit, but most of Google's successes were actually acquisitions maps mail youtube i mean i might be wrong about one or two of those but i mean like a lot of them started docs was rightly they bought it it was in c-sharp or it's a java right you know spreadsheets all that stuff how long have you been outside of google oh gosh i've been gone for about i guess 2017 so what five So what, five, six years now. Okay. What's your assessment now in terms of innovation? The company, it seems like they've had so much a lead on AI stuff internally with DeepMind

and with all, I mean, tons of effort going into it. And yet, at least from the outside, it seems like they were caught very much off guard by ChatGPT specifically, but just by the general LLM movement here in transformers. And they're like, you know, having all hands on decks meetings and trying to like come out with Google

bard. To me, it seems like they're very much threatened right now. And I mean, search their, their moneymaker is threatened by this. What are your, what's your thoughts on it?

Yeah. I mean, I think that, I think that Google's never had to innovate because they found a moneymaker early on. That's so good. I mean, such a moneymaker. It's so good that they never felt any pressure. They wanted to innovate because they were like, well, we better come up with something just in case.

But that was about the amount of passion and energy they put into it. And so most of Google's focus forever has been on basically their ads monopoly. And yes, I did say monopoly, and you shouldn't edit it out. I mean, I do not think that Google has a monopoly on most things. Having been there, I can see that we're desperately behind in most places, cloud and so on. And anybody can, in theory, write a search engine to compete, but the ads ecosystem is unbreakable, and that is a monopoly. But that meant that they didn't have any pressure to innovate. Amazon always had pressure to innovate because the margins were so thin.

Right? In books, music, video days, they had high margins. But Jeff came to us one day and said, that's all going away. And we're going to have to sell other stuff because of digital. And nobody's going to buy a CD anymore. He told us in 99. We were like, what? We love CD.

I mean, like, I'm sure. I'm an old. See you again. Holding on. You probably us in 99. We were like, what? We love CD. I mean, like I'm shit. I'm an old, see you again, holding on to that. You probably still have your collection. You got a collection? Yeah.

And so Amazon was forced to be innovative and forced to try many, many, many things. And so Amazon got good at innovation.

Google never really got good at it. It never did. Now, LLMs have kicked them in the pants. And I, I was going to say,

do you think that monopoly is really at stake? Well, honestly, the bigger threat to the monopoly is the EU, and I hope they're successful in breaking it up a bit, right? Because we need to have other players in the ads ecosystem. I don't think ads should ever go away. When you see an ad that's perfectly done, you don't know it's an ad. Oh, yeah. In fact, if you go on Amazon's mobile app and right, you know, and just shop, like 80% of the stuff you're seeing there is ads.

It's hard to do placements, right? But the ones that I'm looking for, like I search for things, there's a little video there. I welcome those things. Like, give me a great demo. If they know what you're looking for. If I see a guitar ad, I will always be happy, right? So, like, I don't think ads are going away. But, you, but the way ads are done today is very much dictated by Google.

Google and Facebook, those are the two networks for slightly different reasons. Amazon is, of course, actually passing them both. Again, so maybe Google doesn't have a monopoly on ads because Amazon is making up so much ground, but still. Well, even Amazon has competition now too with TikTok shop feature. They do, they do. Like a lot of people find things via social and just want to go and buy it. The creator almost misses out on the attribution.

It's hard to say, hey, use my links and I get a little bit back. But like they literally put a shop button there on the TikTok video and you can just engage right then and there. Using money. Yep. It's disruptive.

And every time we think the giants are done and they've won, they get disrupted and Google's getting disrupted right now. And it's by the LLMs and it's that they invented. And it's quite ironic and amusing, but I hear from folks that I talk to at Google all the time that they've pivoted the whole company to AI and every team is being asked to train models.

Is that right? Deal with AI and work around there. Yeah, and it's kind of a dumpster fire in that you can't just take some random engineering team and have them fine tune a model because it's a very delicate operation and apparently it's not going very well. So they're in panicking, scrambling mode again,

but they also have incredible infrastructure. They still have incredible engineers, and they have a lot of power to get stuff done. They've typically been their own worst enemy. They block themselves a lot, but I think maybe now with a fire lit under their butts, they might move a little faster we'll see you got a prediction who's gonna win well do you think open source

no easy button the problem with open source is it's it's not the model it's the data that makes the quality of the ultimate thing right so opens you know open source whether it's open or source or not doesn't have as much really bearing as whether you've got really, really, really good curated training pairs. And right now, the big expensive guys all do. You know, Anthropic and Bard and OpenAI, right? Right.

Let me ask more specifically, where do you think Google will be at given this ad monopoly that you've described? The change that can happen there, the shakeup up the reaction give me a five year where would google be at five years from now what will we will we be looking at a company that's clawing its way to relevance remaining relevance what's going to happen i think though i think they'll be fine right i i don't think that we're going to see

this search market overturned um i do think that in a 10-year horizon, that's going to look really, you know, the old adage about how we overestimate and underestimate, right? But five years is pretty soon, but 10 years is pretty far. It's a little weird, right? But in the five year horizon, I see Google being fine. I don't have any hot takes there, right? They invented this stuff. And what they did was they immediately stopped publishing. They've been going back and forth on this for many, many, for decades. Should we publish papers? Should we share our, our discoveries with the world? Right. You know, should we open source things? Yeah. So, you know, TensorFlow, which was originally a Google, Google brain, or maybe it wasn't

TensorFlow was not called TensorFlow originally. And Jeff Dean wrote it at Google and it was, you know, the new hotness for training and inference. And Jeff published an internal memo at Jeff Dean. We're talking about Jeff Dean. Like Jeff Dean is probably the greatest programmer of our generation, right? You know, of our time, the most influential. I mean, you could point to and maybe Linus Torvalds,

it would have to be between those two. Who else would there be besides Jeff Dean and Linus, right? I'm trying to think of somebody here. I mean, who do you think, right? That's the thing, it's hard to pick individuals. Things aren't solo efforts anymore. Everything is a team effort, right? Look at the paper. It's like Ken Thompson, but that's not

this generation. Yeah, exactly. Ken Thompson, Dennis Ritchie. Dennis Ritchie wrote the first Go compiler at Google when he was 72 years old. So yeah, but that was previous generation, right? Jeff is a generational talent, and he had to write a memo internally saying why we are open sourcing Google Brain. I remember it was actually called Google Brain because he said why we are open sourcing it? Which became TensorFlow. And he had to basically mandate sort of by fiat, which he had never done before, that we're going to do this

because there was a big resistance to sharing stuff. But Google was falling behind because you lose thought leadership. What Jeff outlined for us was that we kept inventing stuff, publishing a paper on how cool it was and how cool we were. And then the open source world would go,

wow, that sounds really good. Thanks for not freaking giving it to us, Google, but thanks for showing us how you did it. So we're going to build our own. So we had MapReduce and they had Hadoop and we had this and they had that. And all of a sudden Google was the proprietary assholes for everything because of our policy of not open sourcing. And Jeff was like done with that. And so we open sourced TensorFlow and it was very successful done with that. And so we open sourced TensorFlow and it was very successful, right? So the pendulum has swung back now, right? And, you know, no more papers, no more papers,

because like they had somehow managed to- We're giving them our ideas. They're eating us from the outside. Well, Transformers was Google's idea. So was Burt. I mean, most of it, it was only up until Sam Altman like crazily said we're just gonna shove it out there and see what happens because google was too conservative with it

right all the big names they got they're they're trying you know but they have to be they have to they can't they can't write an offering that's fused the users really badly this is why source graph has been able to kind of like do something cool that's been harder for microsoft to do i'm not saying they can't do it and they very well may, but we went hog wild when the LLMs came out and we said, we're going to live with the occasional hallucination and the wrong answer and whatever, because like if you've got a metal detector that can find, you know, rubies and, you know, some of the time it's finding, you know,

dog poo that you step in, but whatever, right? You clean your dog. Cancel find dog poo. Right? Whatever. You're still finding rubies once in a while, right? A lot of it depends on your tolerance, your threshold for, your pain tolerance for how, you know, it's like using regular expressions, right? Once you start using one, you've got two problems now, right?

Mm-hmm. right you know once you start using one you know you've got two problems now right and so so some people it is for some people and it's not for others but we decided to just just open it up and say look this is a developer tool we think you're really cool and microsoft and and the bigger companies are kind of unable to do that because they need to curate curate curate and make sure the experience is really polished or else their users will be you know jarred and oh whatever and it makes it harder to innovate and be bold because we can go out and try things, right? And see what works. And this, so again, this is, you know, I would say I learned a lot more from Amazon than I did from Google when it comes to leadership and products and finding product market fit. Because again, Google wasn't forced to innovate and Amazon was. And so I watched Bezos try and try and try. He wouldn't give up.

He'd be like, I know there's something here. Think of, I mean, you can count probably on one hand the projects. You know, the Google graveyard is thousands of projects, right? But Amazon's is pretty small because Bezos will keep changing the shape of it until he finds the fit, right? The only really big one I've seen him fail at was the Fire. Oh, the Fire Stick. Not the Fire Stick, but the Fire Phone. Yeah. one i've seen him fail at was the fire the fire the fire stick not the fire phone yeah yeah that was a fail because the market was just too crowded or whatever i heard they're divesting uh in alexa and that and echoes and stuff i'm not sure if

that's true yeah there's no money in it there's no money they i think the idea was they're gonna sell more retail but the idea i mean look i was talking to some Amazon insiders recently and they were making making fun of the poor Alexa team right because how do they monetize whenever Alexa says you have a new notification would you like to hear it what she's really saying is I'm running out of money would you like to help

that's an interesting one because there's so much application for Alexa that I just wonder why they haven't been able to land the plane because – Well, and they got penetration into the households. They got in there. They just had to give so many away. I mean, look, I'll have to imagine that the ML, the sort of step function improvement in ML, they probably were not using large language models before. They were using their own models.

And with LLMs, they're probably revisiting and wondering. I mean, Alexis, you know, Siri's and Alexis of the world are probably going to have to take a big, I think they're going to get dramatically better. If they don't, then there'll be like one on your desktop that you can talk to that's dramatically better. They're going to get replaced if they don't get better at this point.

This is why I was saying Apple needs to do something with Siri. I mean, it's so embarrassing at this point. I thought they would announce something at WWDC because everybody's just talking about theirs and they're so much better. Siri is so embarrassing and yet not a peep at WWDC. They're just not doing anything with it. It's just like, what?

I'm sure they are internally working very hard. They have to be. They probably are. Here's a question for you, Jared. We just had this conversation. It was kind of to you and Steve, you can participate because you're here. Thank you for being here, by the way. Yeah, gee, thanks. Steve, you can also answer if you and Steve you can participate because you're here and thank you for being here by the way yeah gee thanks

Steve you can also answer if you want no you can totally answer well they call it AI I mean they resisted calling it VR with the Vision Pro like they literally danced around the buzzwords will they use the AI will they say AI I don't know I'm not going to

answer they're so particular with the way they say things and name things to a. I'm not going to answer. They're so particular with the way they say things and name things to a fault. I mean, the way they refer to iPad, it just bugs me to death. You know, like they won't say the iPad, they'll just say iPad. Like that stuff's all very intentional and calculated. And to me, that makes me not like them as much, you know, even though I like them for lots of reasons, it's just like, there's something weird about that. So the way, I don't know what they're going to call it because they probably have meetings upon meetings about what they're going to refer to it as.

They did say transformers a few times. They stopped saying machine learning. They started saying transformers. So I feel like that's kind of a nod to us nerds to be like, what are you using there on their autocomplete or something? So that's my take.

I have no idea. Steve, you probably have a better take than that. No idea. Yes. I tied him. I tied him tied him they tied him same exact response i want to go back to this google thing for a minute i thought maybe this is a better way of casting it which might be uh enjoyable to think about so 10 years cultural significance like we can talk revenues and we can talk monopolies and stuff but you know today it has been for

probably 15, 20 years, you Google it, right? Did you Google that? Just go Google it. I'm starting to hear, did you ask chat GBT? Do you know, like, will Google it be culturally significant in 10 years as it is right now? Or will there be something else that's more like asking AI? That's like, that's what we do now. We don't Google it anymore, really. Or if we do, we don't say that. Now we're saying this other thing. What do you think is that might be a proxy for their success over the next 10 years? Well, I mean, the pundits have been saying the search bar is dead, right?

Yeah. And it's any dead yet, but I mean, I guess it's mortally wounded at this point. So I expect that Google will look very different in five to ten years right it'll still you'll still google things but it will be a much friendlier experience and by the way i don't think they like it's funny i mean of course it'll never change but i don't think they like using googling as a verb for this whatever the trademarking reasons you know what i mean so it's pretty because people will be like i googled it on bing i googled it i know but i

it's kind of like, sorry, guys. I hate being the de facto standard that everybody refers to. Right. It's like, oh. Here's the world's smallest violin. Tie it again. Well, for this call, I did not chat GPT your name. I Googled your name and dug through links because that's what I wanted.

So contextually, I wanted a list of links. I did not want a dissertation of who Steve Yeagy was. Because I'm sure they can be... It can maybe give me resources and sites and things, but I've never actually had it give me a link that I'm aware of, so I could be wrong. The thing to keep in mind here is the models don't know everything. They've looked at a lot of stuff. They're very good at pattern detection and completions and all the other things, but models don't know i mean they don't know everything they've looked at a lot of stuff right very good at pattern detection and completions and all the other things but they don't actually like have perfect memory and recollection of everything they've looked at it's almost like

taking a person and letting them read a bunch of books and they'll write stuff about the books right but they won't even always be as accurate if you see the gettysburg address printed in 100 different books then they'll eventually memorize the text of the Gettysburg Address, right? But for the most part, they're just kind of – so what they need is they need other data stores. And Google's is like the biggest, right? Google has all of these really bespoke, custom, fine-tuned backends for every possible knowledge topic. And so it's really just a rejiggering of the form factor for google right that you know putting a more ai sort of like i don't know interaction model in front of what what already exists is very mature search infrastructure yeah that's why i don't feel

in any trouble i'll tell you what i know exactly how they win if they're listening they should do this this is exactly how they remain relevant 10 years from now. Let me tell you, their cash cow is the ad monopoly. And the way you win there is with relevance and quality. When I search, I ignore because it's not relevant or quality. If they increase that relevance and quality, the way that Amazon has, the way that Instagram has, or even TikTok in some cases, they're not ads. They're just creators being paid on there. They make a living on there. So they're kind of ads. They're kind of like walking ads or living ads. Relevance and

quality. If they become Bezos obsessed on relevance and quality in their ad department, in every way, shape, and form, they can win. Interesting. I bet they've never thought of that before. I guarantee they haven't. You know why they haven't? Yeah, because- Because it sucks already. Relevance and quality is not there, ad-wise.

My hypothesis is that because there's a third dimension that they've decided to optimize for, which is money, which is another form of winning, I guess. But yeah, it would make for bad ads. I would pay more as an advertiser if you let me deliver relevance and quality that also equate it to money for both of us. I think the advertisers all want to pay less, though. You're probably unique. Maybe you actually pay less in that way.

You force relevance and quality. The people who want to be low quality and low relevance, who need to be there, literally need to pay to get views or eyeballs or whatever you call it. They'd probably pay a lot more than those who truly are relevant and truly are high quality. So maybe you optimize for those who really need to claw to get there

and the ones who really deserve and have that kind of quality maybe they pay less. I don't know, but relevance and quality is the levers. I'll tell you, quality comes at the expense of privacy in ads. It does. I don't know. Well, how does is a levers. I'll tell you, quality comes at the expense of privacy in ads. It does. I don't know. Well, how does Instagram do it?

How does Amazon do it? They trample your privacy like sociopaths. Right. And there's an uncanny valley of quality where when it gets too close to home, which is too good, specifically Instagram, now you start having people say, I think they're listening to me. I think they know. They do. Yeah, it's 100%. You can talk about some random topic in front of Alexa and you'll start seeing ads for it. So there you go. I mean, that's quality though. Cause

it's exactly, it's so relevant to your life. Like what is, cause quality means intent really. It means like, you know, is relevance, right? They're all, they're all related, right? It's, it's intent if you search for something. Right. Yeah. We all want relevance and quality. And, of course, Google wants to give you and all these advertisers want to give you the relevant, you know, high quality. But we don't want creepy. So, like, there's this weird rub there, right? That's the problem. That's exactly the rub, man, is that, you know, they need to know a lot about you, potentially through mechanisms that you don't really want them to be using in order to show you something that's relevant to you. That there is the crux of the problem.

How do we bake? I mean, this is not a show on search relevance, but how do we bake in opt-in relevance then? Because I would certainly opt into some relevance, like at my own discretion. Don't be sneaky about it. Give me a chance to tell you what's relevant to me. And then I would happily take. I mean, opt-in relevance is called a click. Yeah. I mean, you're right right i mean like like click the ad click the ad or search something right if i search

something or if i engage with something that's kind of like if i knew that as a user like hey when you engage in my platform with x y or z that's opt-in relevance and i'm going to give you more because you you're going to want this this is how the platform works so get used to it if you're here and playing, you opt by these rules. You play by these rules. I can't believe you guys got me talking about ads. I've done it twice in my career and said no, never again, and then done it again

maybe three times. Here's a good transition off of ads then. Five years ago, sorry, 2017, you joined Sourcegraph. Did you come out of retirement to join sourcegraph or was that just a kind of a cool way of saying it you actually retired yeah you retired so after google you're like i'm done yeah i was working on my house mostly okay building a house

i built a really i built a really nice studio i'm sitting here i can see nine guitars from where i'm sitting nice piano and ukulele and a mandolin and yeah all right so i was living the dream so the dream turned out to not be so hot i mean how can we come out of retirement then why don't you just live the dream i guess it's possible to start climbing the walls even in your own private like castle with all the powers you ever want to climb the walls okay i missed people i'm a bit of an introvert but also you know i miss miss with people. I miss solving great problems. I know, and I know, I know a bunch of folks that did exactly the same thing I did, right. Retire for two years. And they're like, I get to get back into this, you know, honestly, truthfully, and this is just my, my personal

opinion. And it's like, it could be just, you know, complete astrology, but I think that there's maybe a correspondence between like, you know, having a purpose and, and, you know, and work can give you that purpose. Of course, you know, there's, there's other things that can as well, but I don't feel like I'm ready to sit down and just start writing novels all day long yet. And that purpose will, will I think help you live a longer and fuller life. I really do. Retiring at the age that I did might've, might've actually like wound up aging me prematurely. I were, I worry. Who knows?

Who knows? In any case, I'm having fun. Those two years that you were retired? Yeah. And how long have you been back in it? Gosh, I think it's since late September. So, yeah, going on 10 months maybe. How old are you?

How old am I? I'm 54. Okay. And after these 10 months, you're in it. you're back in the game, you're working. Do you feel more satisfaction? Do you feel more fulfilled? Do you feel like you're, do you feel better? Yeah.

I've lost weight. I've, I've been, I've had more energy. I really, it's usually retiring turned me into sort of a video game drooling zombie. And I didn't like that and i was kind of reminded of uh you know borodain saying you know he has to spend all of his time coming up with stratagems to keep him from turning into that guy that sits on the couch and plays video games right and so i was kind of inspired by that that's that's a part of why i came back out of retirement that's crazy makes sense it's just funny timing we just had a show with kelsey

hightower who just retired from google at age 42, I think he is. And he's like literally in week two now at this point. And so he's like, I'm going to see how it's going. But you know, we'll see. I, when I retired, I was like, man, I am done. Right. And you know, my buddy Mark Porter is like, I give you two years. You drilled it. He's retired twice. And both times his wife wife said you need to get out of the house yeah for real i mean having something to to truly push up a mountain is fulfilling you know we we like struggle struggle is necessary for life you need you know some people get upset when they have fear or they have doubt those are just feedback mechanisms you know or even anxiety is a feedback mechanism for your body pain even you know it's just a

feedback it's not bad you don't remove pain you want to remove fear or even anxiety like these are things you need it's calculus when a car goes down a road it doesn't not feel friction against the wind it needs that it's aerodynamics it holds it to the ground and it gives it ability to hit the turn hard or whatever it might be same thing for human beings and their purpose like you need you need that push back to life you know you sound like a great leader adam you need fear fear you need i work you too that's what i tell myself i lead myself i i somewhat lead jared to some degree i mean we co-lead each other as you've said before i'm kind of kidding though because i mean like it is it is important basically training data for your brain like

it's the whole when i said feet to the, you learn with your feet to the fire. Right. So we don't want to remove it though. They want to remove these things. They went, I don't want fear. Right. I don't want anxiety. Those are feedback mechanisms.

Leverage them. Right. When, when they disappear for two years, they're replaced with Cheetos. Well, like you said, Steve, at the beginning, like those moments when you're learning, like there's when your feet are to the fire, right? Like you actually growing. It hurts. It's scary. It's unpleasant. But you're growing as a on Netflix. The stories of your life happen at the uncomfortable, outside of your comfort zone, weird times. Those are the ones you want to tell your kids later. So it's all part and parcel of the same thing. You're back in the game now.

Okay, so why Sourcegraph then? I mean, you probably could have went anywhere, right? You probably could have got a job. Literally why and literally how? How did you get wooed back to the game? The exact story. I mean, there wasn't much to it. My buddy from, you know, Temporal?

Or Cadence? It used to be called, I forget. I think it's Temporal now. Really cool system. The CEO, a buddy of mine, introduced me to Quinn and Beyond. And we chatted and found that there was a lot of alignment and overlap and interest. I mean, they're just cool, right? I mean, Sourcegraph

is just super cool. Honestly, once I realized, man, I got to get out of the house, right? And get out of retirement, Mark was right. Then I started talking to companies and I talked to 21 or 22 different companies. And I'm talking about like talked to 21 or 22 different companies. And I, I'm talking about like, like in your interviewed with 21 or 22 different companies, like, you know, and had the conversations with them, you know, for a bunch of executive positions, heads, VPs and blah, blah. A lot of them were banks, oddly enough. And a lot of them were ads related, which I really didn't want it. And I just, I was trying to find something that I could be passionate about.

Something that I would just like, it wouldn't be work. You know what I mean? It wouldn't feel like work. It would be work, but I mean, it would be like, it would be rewarding enough, you know, and I don't feel like working at a bank would do that for me, honestly, even the cool one, like Capone, and because they're very tech savvy and they use AWS and whatever. So, ultimately, I mean, I took a look at the options and I was like, yeah, I mean, God, I mean, they're pre-IPO, you know, Series've got money in the bank, they've got an incredible culture, they've got really cool tech. I looked into their tech and it had finally reached the point after 10 long years because they focused on search first, which is a huge, hard problem. And then eventually they came around to code intelligence.

That's when they started talking to me when they had that stuff off the ground and that's the stuff i did at google and it dovetailed really nicely and right after i joined i joined in september did some sort of you know kind of like cultural maneuvering to make sure that everybody knew that we were going to be a high performance culture going forward but without the tyranny and and everybody actually wound up being super excited about that we wound up with a really cool, you know, high functioning team. And then in November, LLMs landed. ChatGBT landed on November 30th, right? At which point we like kind of looked at each other and because we'd already been working with AI, with, you know, natural language code search. And that set us up, you know, that actually set us up to basically build Kodi in a couple of weeks, you know, V1. And so,

yeah, it was just, it's just, I was lucky. I was lucky. It was unbelievably lucky, like fit for, for I think all.

Really? Yeah. Seems so premeditated. I mean, based on your history. Well, I mean, they said,

that's the prepared, right? I mean, I taught 21 companies, so maybe it wasn't luck. Maybe it was hard work and hard searching, but I'm grateful that I found Sourcegraph. And I really, really, really excited to be on this journey with them because it's gotten crazy this year. Your selection criteria is good though.

The pre-IPO, Series D, great relatable founders. We both know Quinn and Byung. I've known Byung for years. I've known Quinn for at least a year or so. And I've had great conversations with both of them and they're down to earth, easy to talk to, just not strange people. Not sociopaths? No. That's a good point. I don't usually list that reasons for why Sourcegraph, but I did talk to a lot of C-suite people in my journey. And Quinn and Biang are definitely a different cut from the rest, right?

Like you say unbelievably down to earth they're just just like us they're just like they just want to get cool stuff done and and hack and have fun and change the world and that resonated a lot with me so we clicked the three of us and that that that had a played a huge role where is this going now that you've built Cody in two weeks which was probably six months ago, seven months ago now-ish. Kodi 2. Hasn't Kodi 2 just come out or something? I saw a blog post.

Yeah, we did just launch, well, it was really kind of Kodi 1, and so it's the first time anybody could just, like, use it. We have to pay for GPUs for you, basically. So it was a bit of a logistics challenge to actually get Kodi out to everybody. Next up is we're going to make it an even more lovable experience. I mean, like I said before, a lot of your experience with any AI coding assistant, including Copilot, is going to come down to your tolerance for its making mistakes. So we're putting together an unbelievable world-class AI team that's pairing up with our code intelligence, our code graph, which is the thing we've spent 10 years building.

It's the external store. Remember I mentioned Google's going to be fine because they have all this incredible search infrastructure that the LLM can use? We're in the same boat. We've got fantastic technology, so we can start improving Kodi's quality. Because ultimately, look, Kodi's great right now if you have a question about your code base, about some stuff you haven't seen, it's new code you haven't seen. But for day-to-day where you already know the code and your IDE's helping you and stuff, you know, maybe Kodi doesn't come to mind as your very first stop.

And so my goal is, you know, in very short order, we want to make it so that Kodi comes to mind first when you're thinking of, you know, a much broader class of problems. And we do that by making it really, really good. You would be shocked at how much low hanging fruit there is. I mean, like you look at Kodi and it, you know, it does make a fair number of mistakes, but I mean, like the ability for us to fix these things is it's extraordinary. Everything that we've tried has made it better. So in other words, we're starting with the kind of the worst possible implementation of Kodi. It's open AI, generic sort of Wikipedia embeddings. It's not, I don't even think they're code oriented. You know what I mean? So yeah, it's going to get a lot better. Where are we going with it? I mean, long-term, look, I want Kodi to

be your first stop, you know, when you're coding, like if you've got, look, one of the things that I was always jealous of, truly like envious, right, Was about, about VPs with, with, you know, executive assistants, right. As I was, as I was going through my career was that they were always like a team of two people and they were always incredibly productive and they, and the VP never had to, never had to go to the printer, right. Never had to do any of the, you know, the expense reports and stuff that engineers are always having to do and the VPs were always freed up, right? I mean, honestly,

that's what coding should be for a developer, right? Is like, oh man, I got some stuff I need to get done that I really need to get done or that I'm struggling with. Let me just have the AI help with it, right? That's honestly, that's where I want to take it. Right. And, you know, it's so early days right now.

I mean, it's really raw right now. It has a lot of value if you play with it, but still, I mean, you probably find that it also has a lot of mistakes, but the way, the way I see it, I mean, this is inevitable. This is an inevitability. It's going to happen. It may take a year. It may take five years before we're all using it every 10 minutes, but it's inevitable. If you got to give us a standout snap, you're going to say it's inevitable.

You can't put that word out there and not snap your fingers. Snap your fingers. I'm inevitable. I forgot about that. I am inevitable. Yeah, that's, that's really, I mean, I do agree. It's, I think the reason why we, and I say the re is a real we, where it's not really my opinion, the pushback against it

is that i never thought would happen in my lifetime i guess and so it seems so sci-fi all you can do is you really have to suspend your disbelief this is what you do in sci-fi films that's what makes them magical is you suspend your belief to believe that star wars is possible that you can shoot lasers at each other have these you know laser swords like that's not gonna happen maybe it would happen actually i have to suspend my disbelief on that one you know i really like that framing i really do that's the problem is everyone is having that trouble it's like i never thought we would truly have what is considered buzzword or not artificial intelligence in my lifetime and the fact that

it's here and we have to explain it on podcasts and be seasoned engineers that have done major things and have innovated and it's here it's here in our lifetime like that's kind of unbelievable and i love the way you framed it i love this adam sorry go ahead what's the end that's that's it really just like i'm just in disbelief. That's why it's so hard to believe that it's really here. But now that it is, the point is, it is inevitable. It's here. It's not going to go away.

The boat has landed on the land, and all the trolls or whatever you want to call them, and this boat, this analogy of a boat, they're coming off and they're getting on land, and they're going to start invading this LLM invasion. That's what's going to happen. Like it's inevitable. We're going to have a future that's going to have artificial intelligence in it. It's going to be the case. It's not going to be true.

And it'll be polished, right? Look at your journey on this show from Skype and mailing USB sticks around from your games to Riverside today, right? Stuff gets better. And a lot of engineers are young and they don't see that the world used to not have AWS and Kubernetes. And they don't see that that makes it harder for them to suspend their disbelief. I love your framing, Adam, right? Because right now, if you look at it, you'd be like, well, it's wrong sometimes. Like that was a huge stumbling block for IDEs in the early days. A lot of Java programmers wouldn't, any strongly typed language

wouldn't take an IDE because they weren't accurate all the time until they finally evolved to the point where they were. And during that period, there were a big war between the command line folks and the IDE folks for, you know, five, 10 years before finally the IDEs were good enough and everybody was using them. Coding assistance is exactly the same. Okay, it's early days. We're basically getting the early adopters bleeding edge, but it's going to improve in quality and refinement until it's pretty obvious that you have to use it

or else you're behind, you know? And the current state of affairs is kind of like the future's here, but not evenly distributed. And the distribution is like your ecosystem, right? So like if you're coding Python, if you're coding Java, probably, things that are well-established

and have tons of public resources, you have a better experience today than if you're coding in a niche language, a less popular language. I know this personally, because I'm an Elixir programmer and they're not good with Elixir. They just aren't.

I see things in other languages. I'm like, oh, that's cool. I wish I could have the right answer for me. So many of the answers are wrong. I don't want to be the skeptic who's like, this will never be good for me. But I do want to be realistic and say, for me today, it's not there. Like, it just doesn't provide very much value. I can see the potential value and I can see the demos on the social networks where it's like, that's awesome, but it's not evenly distributed. I'm not writing that mainstream language. And so that stuff will obviously everything will catch up eventually. That's right. But there, I, I believe you that there's so much low, low hanging fruit. It's gotta be fun because low hanging fruit is the best, right? It's easy to pick. It has the biggest impact.

It's the funnest moment. It's like, i can make huge advancements with not very much effort there's there's been magic times i've i've witnessed and i've rarely been in one myself but i see them like borland in the 90s was that they were the kings of programming languages compilers and it was a magical time because all the best compiler writers in the world you know the kind of not all but probably a good like you know good big chunk of them kind of congregated at Borland. Then Microsoft bought them all and was like, ha-ha. Then they all went to Microsoft and there was this innovation magical time then when they built.NET and C Sharp and that whole stack. Those don't happen very often. There were a few at Google.

Things get built by small teams in a hurry. And then there's just a lot of chipping and polishing and deployment and engineering and stuff. But it's hard to get those magical innovation loops. You know what I mean? Like everybody's always too busy to like innovate. And so when you can catch them and get people together and there are low-hanging fruit in the space and you're in this innovation loop, it's incredible. Because every week it's demo, you know, everybody's demoing to each other it's exciting you know you're seeing you know meaningful incredible advancements you know like week over week that's where we are today

that's where we are today at source graph and it's man it's it's just incredibly incredibly fun and gratifying but it's also we're turning the ship you know the car the vehicle we're in so hard that you know we're right starting to strain the chassis a little bit. But it's been a lot of fun. Well, I'm excited now. I don't even work there. But just the fact that y'all are doing this and you're not the only ones doing it. I mean, there's people in the meeting.

There's tons of excitement right now. Yeah, I love it. And it's going to be goodness coming out on the other side for everybody. So that's exciting. Anything else we didn't ask you? You're like, man, I've been waiting an hour and a half for these guys to ask this question, but they're too dense. They didn't even think about it.

No, this has been a lot of fun. I love that you guys let me tell some of these crazy old Amazon stories. Are there any? Do we miss any? Are there any other stories where you haven't told them yet? And you're like, I'll just shove this one in at the end. I'm here for it. Oh, gosh.

Yeah. Let's see if I can find a good Amazon story. You know, I've made, I've made Jeff actually laugh. You've all heard Jeff Bezos's laugh, right? That's actually curated of course, right? It's part of this, part of this brand. Is it part of his thing? Yeah. Cause I've made him giggle. Oh, he has a giggle amazing oh nice so that's breaking news it sure is if you had a sound bite of that you can put it against that picture of him with his shirt off

you know the one that's popular now where he's like all ripped and jacked and you could have just him giggling to counteract that I actually remember the joke where I made him laugh too I was in a meeting with Jeff and like I don't know, 40 or 50 people. Like there, it was all of his leadership and I was giving a presentation and I, and I stopped and I said, let's talk about the elephant in the room. And suddenly everybody was staring at me, including Jeff. And I said, I'm not the elephant.

And he just lost ited it nice winging it just winging it yeah so yeah I don't know you know I could probably pour more

more customer stories of people who got completely screwed but I don't know that could be its own podcast you know

it's almost like it could be a recurring podcast I think somebody should to be honest because I mean somebody from ex-Amazon because there are

so many stories that are just going to be lost to the annals of time that would just be wonderful to share about, you know, things, things that can go wrong that are just unexpectedly, like almost emergent, you know, bug behaviors. And then how do you clean it up as a business and then monetize on it? Right. Cause like Grab didn't learn this lesson. I'm serious. I went to Grab. We didn't talk about Grab. I'm serious. I went to Grab. We didn't talk about Grab. Grab was an amazing Southeast Asia adventure. It was a lot of fun. And they emulated Amazon and very successfully, they are kind of the Amazon of Southeast Asia, but they never really got that particular lesson.

We've imagined a podcast of postmortems, you know, where everybody could learn from the mistakes of everybody else after an incident or after something happens and it's always kind of been stuck at well who wants to come on the air and like air their dirty laundry of these porous motors they don't it's a you know what i tried to do that at grab when i first got there and i was like they were like we should do a public they had a horrible horrible outage because of etcd it had a huge meltdown and it's terrible and to restart and they actually grab didn't know how their service dependency graph, they had circular dependencies and they didn't know how to restart their whole system if it went down. So it's hours and hours. And of course, in Grab in Southeast Asia, when the system goes down, people can't get to school, they can't get to work, they can't get to church, and they start throwing rocks.

And then the governor calls and then the president calls and it's a big deal, right? Yeah, I could tell you all kinds of Grab stories. When Grab makes changes, people come and throw bricks through their windows. Oh my goodness. Yeah, very different. That's high stakes. And so they were like,

we should do a post-mortem and tell everybody what happened here. And so I wrote a really, really good, entertaining, funny post-mortem about what happened with NCD and the whole, the history of like, how we were trying to diagnose it. I was ready to publish and they were very uncomfortable with it. And the leadership was like, we can't, I just, I don't think we should. And they, they eventually pushed

back and they didn't want me to, and I was new in partly an Asian thing, but also partly a people thing, right? They, you know, there are definitely tech culture differences that we butted up against from the Westerners as the outsiders. Sure. Right. And they're not big on vocal self-criticism. Okay? That's a core value at Amazon, and it is an anti-pattern in Asia. Okay? Do not be vocally self-critical. And I was basically being vocally self-critical with this post-mortem post,

and they were like, no. Not doing it. The world never learned the lesson. This is a problem. This is just like the problem of genomics and medicines. They can't make good medicines because they can't get trial data because of privacy, right? It's always a data problem working in the back here. Well, I have one more question about Cody. You got a couple, I think this might be one of the, yeah, this is a second episode.

I couldn't find episode one. Cody is cheating. This is from Cody is cheating. Which explains to me, because you said in this episode, I'll explain our moat. And these are, you know, we talk about moats. What's somebody's moat? How's Google lose their moat? How is this going to lose their moat? What moat do they have? You know, it's all about the moat.

I feel like Seinfeld right here. What's the deal with the moat? How exactly Cody is differentiated. This is what you say over all the other coding assistants. And you said, hey, they're going to be here. they're not going to go any way away they're inevitable and you go on to say including co-pilot and why we will always be the leader those are strong words uh you see in episode three you're gonna come up with something else but then you said this you said did i mention i'm lucky well this is the world poker tournament i love the word you

use when you write by the way this is the is the World Poker Tournament of Software Showdowns. This is the big one. And I got pocket rockets because there's truly no comparison to Cody. We're so far ahead. It's blowing people's minds at co-pilot shops. You'll see soon enough. Give us that soon enough. That was two months ago, basically, this post. The soon enough?

Yeah. Yeah, the soon enough. What's, you know, you say you're so far ahead. Software does take time to build. So you're going to see us iterate towards this. But what I was talking about here is it's funny. It's sitting in a tab right next to the one that I'm in right now. It's a thousand page research paper that just got published on the intersection of knowledge graphs and LLMs.

Okay. Our knowledge graph. We have a knowledge graph is what we have about your code and not just your code but the metadata around it like ownership and security vulnerabilities and a bunch of other stuff source grab that's what that's what source graph does is they just learn everything about your environment right and that is the defensible part because it took a very very long time for me to build that at google and i knew that it was going to be almost impossible to build it for the world because the world uses git and perforce and SVN and they use different languages more than Google does, different build systems more than Google does. So Sourcegraph tackled an extremely difficult

problem. Quinn and Beyond went after what I wanted to do, right? When I was back at Google, which was to offer Grok and CodeSearch to the world and Quinn and Beyond did it. And kudos and respect because they had to handle all the enterprise scaling and all the buildup of that. Now that's what they have. That's our moat. Because what you can do

is you can use that graph data in many, many, many ways. This is what I was talking about, about that low hanging fruit. You can inject it into your embeddings to improve their quality very trivially. You can inject it into your prompts for in-context learning.

You can generate few shot examples with it. You can validate the outputs of the LLM. It goes on and on and on. That graph is an amazing asset that Sourcegraph has. There are other people in this space like, you know, I don't know. I don't really follow them, but there's a bunch of really cool, you know, cool looking, slick coding assistants. But really all they are is a proxy to the LLM. They don't have this independent, authoritative source of compiler-grade metadata about your code base that Sourcegraph has that the LLM can also use. Cody can go look up stuff in the graph.

It knows GraphQL. In fact, I don't even need to know GraphQL. I asked Cody to do the queries for me. So that's our mode. I hope that makes sense. Yeah. You said in quotes, YWO will always be the leader. Cody to do the queries for me. Right. So that's our mode. I hope, I hope that makes sense. Yeah. You said in quotes,

why we will always be a leader. That's, you know, that's an absolute always. Then you're staking the ground. Well, so of course, leaders,

of course, another, another one there could mean various things. Sure. Not the best. This comes back to a point that I made earlier, which is that big companies are necessarily, because of the entire Gordian knot that gets created around the investors and the sales and the shareholders and the quarterly earnings reports and the expectations,

the commitments and the quality that people expect, they are stuck. It's called the innovator's dilemma. We cannot be as aggressive as we can. It'll actually like, it'll be problematic for them in a number of ways. Now they may try, I'm sure they can try, but my prediction is that we're always going to be a little edgier, like going out with things that are a little more raw, a little less polished, but still very powerful. Right. And that's what I mean, right. Is that we're always, you know, I think that we have more, we just, as a smaller company, just like we have to watch out for the 10 people startups because they, on UI and stuff, they're iterating faster than we are. If you're smaller, you can iterate really fast. And thank goodness, right? We have this 10 years of, you know, code graph,

you know, behind us that we can use to building blocks in in this race which is truly going to go on for the next 5 10 you know 20 years it's begun the race is on right it's just started just firing that just went off okay so given that let's imagine now i'm let's just say i'm prompting you okay i'm gonna this is why i tell chat bpd sometimes yeah i'm telling them sometimes imagine act as if you're a doctor of programming, right? You have a piece of paper in front of you. You're going to write a prescription for all developers. And from this podcast, they're listening.

They're on the edge. They're like, man, okay, Steve's going to tell me what to do. This is inevitable. It is the future. You may very well be, you being Cody and Sourcegraph, be a leader in this future. What should developers do today to ensure they follow the letter to this inevitability from this podcast? Is it go play with Cody? Is it go play with everything out there? What's the prescription you can write for the majority of software developers to not be left behind today?

Well, I mean, obviously, Cody is the easy button there because you can just download it and use it. It's free. But really, I have written this, Cody's the easy button there because you can just download it and use it. It's free. But really, I have written this prescription for several people recently. Okay. And I know what the prescription is. Okay. The prescription is it's time. Learn AI.

Okay. I put it off for my entire career. I learned linear algebra and a little bit about neural nets way back in the day and then went for 20, 25 years ignoring it. And you can no longer safely ignore it as an engineer. Okay. It is going to start overlapping and intruding into your life more and more. And you're not going to be able to separate AI from engineering here pretty soon. Look what's happening at Google. All the engineers are being told to learn AI. Okay. It will happen to kind of everybody at some point. AI is like a new, you know, it's like it's been mixed into

the atmosphere and there's no getting it out. Right. So it's like it's been mixed into the atmosphere and there's no getting it out right so you it's like salt and pepper in a shaker you know in the same shaker you can't separate the salt and pepper you can't unmix them but it's different than it was 20 years ago okay the set of it's leveled up it's just like today you don't have to learn assembly language you don't have to learn you know a machine know, code or whatever that stuff is below assembly language. GraphQL. Yeah, you don't learn GraphQL either, right? So learning this stuff is easier, but everybody, every engineer needs to familiarize themselves with the foundational concepts that are in play with LLMs today.

Because, first of all, it will make you more effective. You'll just be better at, like, pulling off-the- shelf tools, including the large language models for your own work. But second, it's, it's just, right. I mean, like, look, the broader form of my prescription is learn, keep learning, right. I've said this my whole career. I don't think it's anything new, right. You should, you should be always reading papers, reading books, catching up. Like college was not the end of learning. College taught you how to teach yourself, right? So now everybody should just be continuing to take time to teach themselves. And I took it upon myself to finally, I told you I'm 54. My brain ain't exactly as springy as it used to be, right? And it's been 30 years since I took a linear algebra course. And yet I forced

myself to learn as much as possible. And I continue to force myself to learn this stuff and become an AI guy. Okay. Because, because it's time. It's time. It's, it's here. Right. Don't ignore it or you'll be in last place. That's my prescription.

Okay. Let's get a second prescription to this then. Once they learn it, what should they do then? Go play with Cody. Once they don't know, you'll'll know you'll know because it's you you'll just know you'll be like oh wow i could do that that's what you'll say the reason why i ask this because i'm talking to like companies and people and they're like well i gotta in i gotta put ai so i have lots of conversations behind the scenes with would be in potential sponsors and

partners and different things we do and more recently recently, I talked to Abhinoda, who is founder of DX. And he knows he needs to put artificial intelligence in it because it's essentially surveys to improve developer efficiency. Sourcegraph may even be a customer, I don't even know. But the point is, it's all text, it's all learning. But it's so hard to like parse that. You want to take this large survey that everybody gave back and you want to inject AI somehow well then you've got to take your prescription say okay i'm willing to learn artificial intelligence i'm willing to go to the ground with it and get it all but then it's like how do i apply and everybody sort of needs to inject or embed ai into their product some way shape or form and you say you'll know from this learning i was hoping for more of a silver bullet

if you had one like uh specifics well i mean at that point like ai is always the same like ai is a very well established even with llms it hasn't fundamentally changed the basic game which is you establish benchmarks with human evals that say this is good this is bad let's teach the computer what's good and bad okay that basic these are the foundational concepts you need to learn because then you have an idea. Let's say you have an idea, Adam, you're like, you know what? My product is a photo browser. I'd love it if the AI could like make recommendations based on my photos of whatever, right. You know, some, something I should buy or something I should eat or whatever, knowing the foundational concepts that I just learned as a middle-aged, you know, old dude, right. You know, this-aged old dude this year, because the resources are unbelievable.

It's not like it was 20 years ago or 15 or even 10 years ago. You've got YouTube and you've got all these amazing tutorials and visualizations and all this stuff. It's very accessible now. Once you learn the foundational technique of whatever, you're just going to, you're going to, whatever it is you're doing, it's basically going to be evaluation and experiments. And you say, did it work? Did it improve my benchmark? That's it. And you just iterate your way. And so, and it's so quantitative that there are leaderboards for it. You guys know Hugging Face, right? Literally, it's a giant, it's like video games. Like literally it's a giant it's like the it's like video games like literally everybody's like their leaderboards ai has been a big gamified competition for 20 years and so once you're in ai

you realize how it works you're like oh okay this is what we need to do we need to make our product better in the following ways we're going to set up the following experiments and we're just going to keep picking the one that works the best that's it there's nothing to it you know it's just learning the tools and the names and stuff cool nothing to it but to do it and the time is now so i know i didn't subscribe the prescription i know it feels like a bit of a suppository but it it really is it is something that i i strongly recommend that people do i know everyone's kind of lazy and kind of doesn't want to do it but seriously and it's fun too right i mean like i didn't seriously i ignored ai i like purpose even at google for so long and and and i'm telling you now you don't have to do

you shouldn't ignore it anymore you guys both sound exceedingly disappointed by this no no what my silver bullet to be you wanted a silver bullet no no it's not that really not i think any answer is great you got a lot of wisdom. And I think, you know, your prescription was spot on. Your second part of it was mostly there. But we'll improve it over time, right? Yeah. No problem there.

Iterate. Come back and answer it. Give us another prescription. Come back. Give us another one. It's really been fun going through your history and like hearing this experience. Because I think. For sure.

One thing that I think Jared and I both love about this show is we have a diverse set of people on this show from all sorts of different backgrounds young french dudes who make the first or an oculus rip open source version of it like that kind of thing from folks like you that uh have a lot of experience in history and not just years but like diverse i've been here i've been there i've been there and you've been bold enough to push back you've been bold enough to get jeff bezos to giggle and then also leave whenever the you know criteria no longer met your desires you know this no longer innovates this is too conservative or this is not that anymore and that's i think people need to hear stories like yours because that's that's how we all sort

of learn we hear stories back to the storytelling you, that's how we all sort of learn. We hear stories back to the storytelling, you know, that's what's key. And that's what the real magical thing of this show is like, there's no perfect Steve Yegi episode here, or at least the first one I'm, I'm hoping and sure we'll have you back to share more.

Thanks for having me. I, I, I really, I, that's what you're saying is resonating. I really, I really do think that we need to share more.

Like our companies tend to be really closed up, you know, but we're just people. Let's share more. For sure. For sure. For sure. So sourcegraph.com, sourcegraph.com slash Cody. And we're fans of Sourcegraph here.

We've been sponsored by Sourcegraph before. We've been sponsored by Sourcegraph before. We've been fans of Sourcegraph for many years, and I'm rooting for you all. I like your moat. I hope it gets bigger, and I hope it gets fortified. So good luck on that. Well, thanks so much. Thanks, guys.

Yeah, thanks, Steve. This was awesome. Okay, so next week on the pod, Solomon Hikes comes back to talk about Dagger and really we catch up on the days of Docker, really since he's left and all the things that has transpired in his life, really. To get to the point of creating Dagger

and what they're doing there. Hey, we're users of Dagger, so we kind of have a little bit of experience to some degree with it. So come back next week. We'll have that show for you. And of course, there's a bonus. There's a bonus today on this show,

but it's only for our ChangeLaw++ subscribers, our beloved yearly and monthly members who directly support us. They drop the ads, and they get a little closer to the metal with bonus content like today and more. Check it out at changelog.com slash plus plus. 10 bucks a month, 100 bucks a year. Support us directly.

And for that, we will love you. We love you either way, but we'll love you a little bit more. And we'll show you too. Okay, so big thanks to our friends at Fastly, Fly, and our friends at Type Sense. And Brake Master Cylinder, those beats, they're banging. Gotta love them. But that's it. This show's done. We'll see you again on Friday. Thank you. Outro Music