"Everyone wants velocity. The difference between velocity and speed is that velocity has a direction."
Is your team's AI strategy tailored for a fast-moving startup or a high-stakes enterprise?
The answer could determine your success or failure. We're rejoined by Itamar Friedman, co-founder and CEO of Qodo, to break down what separates engineering teams that truly thrive with AI from those that are just experimenting, explaining why the path to success is fundamentally different for a startup that needs speed versus a large enterprise that must untangle bottlenecks.
Itamar reveals his vision for the evolution from "vibe coding" to a more mature "grounded coding" that relies on structured workflows and rich, automated context. He also points to the trend of dev platform teams as the future "agent keepers" who will own the holistic and safe implementation of AI. Itamar provides an actionable playbook for leaders: map your current processes, identify your biggest bottleneck, and find a specialized AI tool for that specific problem.
Show Notes
- Learn more about Qodo qodo.ai
- Connect with Itamar Friedman on LinkedIn
- Follow AI thought leader Andrej Karpathy
Transcript
(Disclaimer: may contain unintentionally confusing, inaccurate and/or amusing transcription errors)
[00:00:06] Andrew Zigler: Welcome to Dev Interrupted. I'm your host, Andrew Zigler, and today we have a special guest with us because my normal co-host Ben, isn't joining us for today's new segment. But instead, I brought along one of my close friends in the DevRel world. I'd like to introduce you to my pal, Rizel. Rizel, why don't you introduce yourself to our listeners?
[00:00:25] Rizel Scarlett: Hey y'all, like you said, my name's Rizel and I'm a tech lead at block focused on open source, particularly focused on an AI agent slash CP client called Goose. Um, and I'm so happy you have me here.
[00:00:40] Andrew Zigler: Oh, we're excited to have you here. Uh, we talk a lot about Goose on Dev Interrupted. That's not the first time our listeners have heard. Uh, so definitely be sure to check out Rizel's work, uh, after you listening to this podcast. Rizel's publishing stuff constantly on Goose, and I'm learning from her all the time. Uh, but today, Rizel is, uh, locked in with me. We're doing the news segment [00:01:00] together, so we're gonna cover some of this week's news. We're gonna go ahead and dive in Rizel. Okay.
[00:01:04] Rizel Scarlett: I'm ready.
[00:01:05] Andrew Zigler: Alright, so first up, uh, we're talking about the acquisition of Windsurf.
[00:01:11] Andrew Zigler: This is an interesting saga that hit the, hit the news in the last week about the next chapter of Windsurf. and this comes after windsurf, um, acquisition as, as you might have. heard they were gonna get, they were going to get purchased, uh, but then there was like a second purchase.
[00:01:29] Andrew Zigler: There was another agreement. And uh, to be honest, Rizel, I had a hard time following this
[00:01:35] Rizel Scarlett: Yeah.
[00:01:35] Andrew Zigler: in terms of who got what from windsurf and what was left and what's the deal. So what's your read on this?
[00:01:42] Rizel Scarlett: Yeah. First of all, like you, I'm confused because. I didn't, first off, wind surf is a hot commodity. First. I remember Microsoft or Open AI wanted them and then Microsoft came in and was like, nah, I don't know about that. And then we got Google. I heard that they got, they [00:02:00] acquired them. And then this week they're talking about, Uh, cognition. Got them. So I'm very confused on like who got what, and this is just not like the traditional type of acquisition. I'm wondering if more acquisitions would look like this in the future too.
[00:02:13] Andrew Zigler: Yeah, I kind of think it's becoming the norm of
[00:02:15] Rizel Scarlett: Hmm.
[00:02:15] Andrew Zigler: this talent that's acquired. We're gonna talk a little bit more about talent acquisition later. Uh, there's a new segment on that as well. Uh, but there's, there's just something to be said here about the, uh, like founders and these early product leaders like getting acquired or like.
[00:02:29] Andrew Zigler: Uh, more like an acquihire, right, from these larger companies who want their brains of how they organized and built this application. Some cases you see them immediately turn around and leave, or then go to another organization. It becomes this new norm where people are kind of ping ponging between these huge players and in the tech world. and so this windsurf, this wind surf acquisition by cognition who you may recognize as creating Devon, It, you know, that's gonna be an interesting new chapter, uh, in terms of how they're going to combine. Uh, Devon itself is a loaded, [00:03:00] uh, brand name in like AI agents and AI code, uh, review in general.
[00:03:04] Andrew Zigler: So I think that they're probably gonna get a lot of mileage out of the windsurf. Name and ip. That was probably a huge part of the acquisition. Uh, but definitely go check out the LinkedIn video, uh, from the, from the founders talking about together. We're gonna make sure that gets linked for y'all. We'd love to hear what you think. Uh, 'cause uh, I, I'm kind of left scratching my head a little bit. Like Rizel.
[00:03:23] Rizel Scarlett: And uh, like you just said, um, I do think it'll kind of impact Devon's brand. 'cause I think Devon's real enterprisey right now and windsurf is, is less so.
[00:03:32] Andrew Zigler: Yeah, I think it'll be interesting to see how they kind of mold together.
[00:03:35] Rizel Scarlett: Yeah.
[00:03:35] Andrew Zigler: so diving into this next article, uh, this is a note from, uh, a past guest on Dev Interrupted, someone that we know and love very much here, Brigitta Boler, um, at ThoughtWorks. And she, uh, has been covering how teams are working with Age Agentic, ai.
[00:03:49] Andrew Zigler: There's another article from her on this topic, and it's called. I still care about the code and I loved this article and that's why I'm definitely gonna include it in our show notes and why we're talking about it [00:04:00] now. Um, ever since, you know, coding assistance kind of came on the scene, there's a lot of people that have been saying like, oh, since AI can generate it for me, like. You know, I don't really care about the code anymore, but TTA takes a really strong stance on why caring about that code matters more than ever, and what it really means to care about that code, what that means from beginning to end, then creating it, but also understanding the purpose of it. Uh, this is like a really salient article.
[00:04:26] Andrew Zigler: Um, it's not even, it's, it's not super long. It captures the idea perfectly, so I really recommend you check it out. Uh, Rizel, what do you think of this one?
[00:04:34] Rizel Scarlett: Yeah, I loved it, especially the part where she called out being on call. Um, when you're on call, you gotta move. Fast. You wanna be res a responsible developer and solve these problems. Sometimes just relying on AI alone isn't the best idea. I've always been like, uh, an advocate for like, if you're gonna vibe code, do it responsibly.
[00:04:55] Rizel Scarlett: Use version control. Still still use critical thinking just 'cause we're using AI [00:05:00] to code or any type of AI assisted coding doesn't mean we just like shut off our brains and just let AI do all of it.
[00:05:07] Andrew Zigler: I couldn't agree more. It's like, uh, by caring about the code, you build your domain expertise in it. Uh, and you can also, both of those existences can ex can coexist. You
[00:05:18] Rizel Scarlett: Yeah.
[00:05:18] Andrew Zigler: create. All of the code and still care about it
[00:05:21] Rizel Scarlett: Yeah.
[00:05:22] Andrew Zigler: about the intentions of the code and where that code came from.
[00:05:25] Rizel Scarlett: Yeah.
[00:05:25] Andrew Zigler: a really important distinction and one that, that brigitta draws here. So be sure to check this out. and moving on now to our next article. Uh, so this is an interesting one. I'm gonna just, I'm so r result. I'm just gonna read the title and I think it's gonna pretty much summarize for everyone what happened here, and then we're gonna talk about it.
[00:05:43] Andrew Zigler: So here's the title from Wired. McDonald's AI hiring bot exposed millions of applicant's data to hackers. Who tried the password? 1, 2, 3, 4, 5, 6.
[00:05:56] Rizel Scarlett: There is so much to unpack.
[00:05:58] Andrew Zigler: There's so much [00:06:00] to unpack. So, first off, this is like, whoever, whoever master crafted this title over at Wire Bravo.
[00:06:08] Rizel Scarlett: Yeah.
[00:06:08] Andrew Zigler: is a great one. So, uh, let's unpacked the beginning. So, McDonald's has an AI hiring bot. Uh, this is interesting.
[00:06:14] Rizel Scarlett: Yeah.
[00:06:14] Andrew Zigler: read a lot of stories about their hiring bot independent of this article.
[00:06:18] Andrew Zigler: And how frustrating applicants have found it for actually applying for jobs,
[00:06:22] Rizel Scarlett: Interesting.
[00:06:23] Andrew Zigler: a barrier, uh, for folks trying to get hired at any McDonald's location. Uh, so that's an interesting use case of AI that I'm, I'm not sure, has been like a lot of success. Now, on top of this, the AI hiring bot. Having collected all of those applicants, uh, it, it seems like there's a hacker that was able to access it by just using the password. 1, 2, 3, 4, 5, 6. So someone, someone unfortunately on the McDonald's team deployed this system and left an insecure password on it, so became a perfect storm.
[00:06:52] Andrew Zigler: We're talking about McDonald's, one of the largest employers. People in the United States, one of the biggest, uh, in fact one of the biggest fast food chains on the [00:07:00] globe. Uh, I would say that if you think of fast food, McDonald's is one of the first that come to mind. So this is a big, this is a big problem for them because they have a lot of people's data and so, uh, they could be culpable for a lot here. Rizel, what do you think of this one?
[00:07:15] Rizel Scarlett: I'm concerned, why are we making passwords be 1, 2, 3, 4, 5, 6? Like, this is like, like basic knowledge that even if you're not a developer, you know, this is not like a secure pattern that you should do. Like the everyday person knows this as well. And I am concerned about like. what data do they have?
[00:07:34] Rizel Scarlett: Like you said, this is a global company. There's McDonald's in Jamaica and Japan and like is the, do they have
[00:07:42] Andrew Zigler: too. Oh my gosh. Do
[00:07:43] Rizel Scarlett: Oh
[00:07:43] Andrew Zigler: have the, that information too? Like, oh my gosh. Like I, there's
[00:07:48] Rizel Scarlett: yeah.
[00:07:48] Andrew Zigler: somewhere to learn about how widespread this is, but this is a, just a good reminder to everybody, uh, to use a secure password, uh, and follow best practices for your password.
[00:07:57] Andrew Zigler: That, uh, practice will never, ever go outta [00:08:00] style. Well, until we become, we become totally passwordless. Uh, but until then, you know, buckle up. Yeah. Do we have that? Uh, so moving on, uh, to our next story here, uh, I can do is start this one with a si Elle. So for the last several weeks, we've been covering this saga, uh, this absolute epic here on Dev Interrupted about AI leaders poaching talent from each other. And we're not talking about, oh, you hired the product leader, or you hire whatever.
[00:08:32] Andrew Zigler: We're talking about ripping out entire heads of labs. An entire established machine learning teams within some of the largest companies in tech. And this has been happening between Meta and OpenAI and Apple and Google, really anybody who has access to huge amounts of capital. And that's a huge need and incentive to build AI as fast as possible with the brightest minds on the planet. And here at Dev Interrupted. It's exhausting to cover what [00:09:00] happens on all of these poaching wars from week to week. Uh, folks who tune in, they know this from me. Uh, you know, what are they gonna get this week? They're gonna get another news segment on the AI poaching war. So here we go. Ah, meta has taken another AI talent, uh, this is from, uh, OpenAI.
[00:09:15] Andrew Zigler: They've actually taken two lead researchers now from OpenAI. So I think, I think now if we were to tally it, that takes the, uh. Number of sci uh, scientists taken from OpenAI by meta to 10
[00:09:27] Rizel Scarlett: Wow.
[00:09:28] Andrew Zigler: Uh, and we're talking about million dollar compensation and signing bonuses to bring all of them over. Uh, so this is another interesting development in the Super Intelligence labs coming from meta, uh, Rizel.
[00:09:41] Andrew Zigler: Have you been following the story? What's your take on it?
[00:09:44] Rizel Scarlett: My take on it jokingly is hire me. I'm.
[00:09:52] Andrew Zigler: You're so right. What am I doing? Poach me guys. Poach me. Poach me. Yeah. Poach us. Poach us.
[00:09:58] Rizel Scarlett: Right.
[00:09:59] Andrew Zigler: Rizel and I will [00:10:00] split the salary even,
[00:10:01] Rizel Scarlett: cheaper. Yeah.
[00:10:03] Andrew Zigler: cheaper. We'll, we'll take it. Uh, gladly divide and conquer on that one. So, mark, hit us up. I know you're listening, mark, because like, you're, I know you're tuning in.
[00:10:12] Rizel Scarlett: These two did this podcast. Definitely.
[00:10:16] Andrew Zigler: but yeah, so it's like this has been going on for weeks now.
[00:10:20] Rizel Scarlett: Yeah.
[00:10:20] Andrew Zigler: it right.
[00:10:21] Rizel Scarlett: Yeah. I have, I have, and on a serious note, I'm curious on like. Has this really been benefiting them as much or is it really like, maybe like more of like a scare tactic for other companies? Like, yeah, we got the best, get your, get your stuff together. I don't know, I'm, I'm curious of the mentality.
[00:10:38] Andrew Zigler: Oh, no, that's a great way of, of striking it. And in fact, there's a lot of emotion involved in these conversations. When I read these articles that have quotes from these founders, they use language along the lines of being robbed. And, uh, of them even used the terminology. It's like someone came into our home.
[00:10:55] Andrew Zigler: And so there's definitely very personal feelings, I think, between these [00:11:00] ai le uh, company leaders, uh, about the acquisitions of these hires from right underneath each other. I do think that. Because of that, there's some level of intimidation involved when we think of AI and this, and the hyper scaling wars and all these companies trying to get as big as possible and how much capital it takes to do that.
[00:11:19] Rizel Scarlett: Yeah.
[00:11:19] Andrew Zigler: you want to be the loudest drum and you wanna be beating the hardest. And so, uh, it makes sense that they're going to, you know, between these, uh, leaders in this space, there might be a lot of like posturing and like kind of mental games going on with these acquihires. Right. It's kind of crazy.
[00:11:33] Rizel Scarlett: Yeah,
[00:11:34] Andrew Zigler: I, I do think that there's something to your. To your hypothesis there.
[00:11:38] Rizel Scarlett: I'm curious to see the end result and when Will, when will they stop hiring
[00:11:43] Andrew Zigler: I don't know, but hopefully, hopefully not before they hire us though. Hopefully we scoot in there. Yeah. Yeah. And then, and then when we come back, uh, you know, next time people tune in would be Ben doing this new segment. He can cover the AI poaching wars, and he can talk about how we got acqui, uh, acquired by meta.
[00:11:58] Andrew Zigler: Uh, again. [00:12:00] That'd be great. Uh, and don't worry, listeners, we would bring you along for the ride as well. Um, and you know, before we, we start wrapping up our new segment, and this has been a lot of fun. We're gonna have more of these like this where we bring in other folks to share their opinions on the news. But before we, uh, move on a little bit, I just wanted to plug something happening at the end of the month. You see my friend Elle here, she is the host of the Great Goose Off, which if you've not watched it, is one of the best new, emerging game shows, for engineers. Uh, not like it's a vibrant category, but she's definitely dominating it.
[00:12:31] Andrew Zigler: And so, uh, on the great goose off, you have folks go head to head using Goose Coding agent. Uh, on wacky projects on a real time live stream. Uh, it's a totally fun experience. Uh, and Elle, why don't you tell us a little bit about it?
[00:12:45] Rizel Scarlett: I mean, you, you really described it well. Um, the next one that people should tune in, we'll have you on July 29th, but I basically have two people going head to head and I'll tell them something crazy, like, build a login form that you can't log [00:13:00] into and we'll see who can and make the morals chaotic. Yeah.
[00:13:04] Andrew Zigler: Oh, okay. So you're gonna break my brain. Is, is what, is what I'm hearing. So, so, so definitely be tuning into that. That'd be at the end of the month. Uh, I'll make sure the link gets included so that y'all can go check it out. Um, now that I've had Rizel here for a new segment, now she's going to put me to the test in a live vibe coding challenge.
[00:13:22] Andrew Zigler: So be sure to come check it out. It's gonna be a ton of fun. let us know as well what you thought about today's news segment so we can be doing more like these. And now that we're coming towards the end of our new segment, I do want to intro our upcoming guest. I'm gonna be sitting down with Eda Mar Friedman of Kodo. We're gonna be talking about, uh, how you can tailor your. Teams AI strategy, uh, whether you're a fast moving startup or a high stakes enterprise, we're gonna talk about what it means to go from vibe coding to grounded coding, to building structured workflows and beyond.
[00:13:55] Andrew Zigler: So this is a little more of a technical dive about how to reinvent your engineering team in the world of ai. Be sure [00:14:00] to stick around for a listen
[00:14:01] Ben Lloyd Pearson: Phew.
[00:14:05] Ben Lloyd Pearson: What if you could cut code migration time in half, just like Google did. Discover how Google used AI to automate 75% of their migration changes recently boosting developer productivity by 50%. Our Dev X guide to AI driven software development breaks down Google's real world strategy and gives you the frameworks to adopt AI thoughtfully across your SDLC. Don't settle for the hype. Learn how to do it the best. Download the guide today and lead your team into the future of AI power development.
[00:14:39] Andrew Zigler: why I'm really excited Itamar to have you here today, because you know, you have so much insight on how technology's evolving right now and how teams are using ai. and for our listeners, you know, Itamar Friedman, he's the co-founder and CEO of Qodo.
[00:14:52] Andrew Zigler: It's a company tackling AI driven. Software development and it's tackling it heads on, Itamar works with teams that are figuring out these [00:15:00] challenges in our industry right now and what separates an engineering team that thrives with AI from those that are just kind of experimenting with it.
[00:15:08] Andrew Zigler: um, as a guest. Itamar has been on Dev Interrupted before, um, and has actually talked about AI back before a lot of people on here were talking about ai. So we're, we're, we're kind of revisiting a source of some of initial thoughts that we had on the show.
[00:15:21] Andrew Zigler: and we're specifically we're talking about how teams are transforming, how they work to build high quality software. that is safe and reliable and scales. So Itamar, welcome back to the show.
[00:15:33] Itamar Friedman: it's really a pleasure to come back here, really high quality episodes and, and, and it's also fun. I like the way you conduct, uh, the discussion. Thank you for having me here.
[00:15:43] Andrew Zigler: Amazing. I love to hear that. Thank you. Well, I, I look forward to having a blast with you today, and we're gonna kick things off with, I, I like to kinda ask this of folks who come in and talk about ai, kind of like setting the stage initially about how teams right now are actually implementing these tools because everyone is under some sort of [00:16:00] mandate or pressure to start implementing them, but that looks different for everybody.
[00:16:05] Andrew Zigler: And to kick things off, what would you say is the minimum bar for a team to say, you know, we're using AI effectively right now.
[00:16:15] Itamar Friedman: I don't wanna complicate my answer, but I do think it's important to put it on a table that there's a very big difference between a startup with a team of three people, like a company with, 200 developers. which for some might sound like a lot. For some it will be small. And with a Fortune 500, with 10,000 developers. Like, there's so many different things about what matters, there, uh, what is the cost of a mistake? For example, probably the cost of moving slow is much bigger for the team of three people. Rather the cost of, uh, sorry, fucking up. And, if you're a bank is, is, is really, really high. So I, I think like that also leads to. the managers or even the [00:17:00] developers' decision on, for example, what to adopt first and what to adopt second, and how do they measure success of actually implementing ai, uh, uh, et cetera. for example, let's say that one AI tool will help you write much more code faster, let's say that there's a metric for that. Let's assume that this, they're gonna be the same amount of bugs per line of code. it means that you're gonna also get much more bugs, right? So, but, but still, if you are a small startup or a beginning startup, you wanna just move fast.
[00:17:34] Itamar Friedman: I'll, I'll move fast and break things. Then, the first thing you probably wanna see is like, let's say speed. Uh, uh, I'm intentional not using the word velocity for a second. Like, you just wanna see that you're, you're moving really fast and probably also vibe coding. Like, like actually works pretty well for you. if you are, uh, like an enterprise for example, then what you want to see successfully, [00:18:00] happening in, in my opinion is, untangling bottlenecks.
[00:18:04] Andrew Zigler: Mm.
[00:18:04] Itamar Friedman: That's so different what I just said on, on the first part. Like it could be that your bottleneck is, how quickly you, you wrote write your code. Uh, but it might not be, I just came back. Uh, I was, uh, I really apologized. I'm gonna say it in front of everyone. I was like three minutes late to this meeting because talked to someone that he pitched me. He has like hundreds of developers. That actually senior developers care very little about writing code.
[00:18:31] Itamar Friedman: That's almost easy for them. What they care about is thinking about the right engineering and the right best practices and how to do things right. And then actually how to use ai, to do that is a question. but I think like if you ask me how people, these days that are thoughtful looking on it is basically via the bottleneck. Where, where is my bottleneck that I can untangle?
[00:18:56] Andrew Zigler: So splitting it then into two camps of you have, you know, teams [00:19:00] that maybe just care more about that speed aspect, and then teams that worry about the bottlenecks and the engineering complexity of what they're working on. And Ima and understanding that their needs for starting with a tool like AI are very different.
[00:19:13] Andrew Zigler: Uh, what are some ways that teams can identify where they fall on that spectrum and help identify what actions they should take? Is it just based on size or is it based upon the complex, like your industry? Are those like the main ones that you're, you're alluding to?
[00:19:31] Itamar Friedman: Yeah. So, uh, the reason I didn't use, speed before, where, where I said that startups want speed. Because I think you can't find, you can, you can find something in common between all of them. Everyone wants velocity.
[00:19:45] Andrew Zigler: The difference
[00:19:46] Itamar Friedman: between velocity and speed is that velocity has a direction.
[00:19:49] Itamar Friedman: And I think like, it's a really good question to ask what velocity means, means for you, right? for example, how to measure it could be how quickly [00:20:00] am, am I, uh, like releasing features. Okay. Uh, for other words, how quickly I'm releasing features and I'm not creating tech debt how quickly I'm releasing features, without additions ofof bugs and issues. for others it could be how quickly I'm doing, new frameworks of versioning or how quickly I respond to RCA, like to root code analysis or, you know, so like, like basically I, I think like the, the first thing to think of is, what's most important for me as, as a dev team?
[00:20:33] Itamar Friedman: What's what's expected, I think for everyone, like it's maximum value and minimum time, but this is like, so generic. it is, I think the most generic, framework for engineering. Maximum value, minimum time. But then you have to think. What is the value I expected, uh, of for me and it will be different.
[00:20:53] Itamar Friedman: And I think there is correlation between, how big is the business and how much [00:21:00] risk it could suffer. And then there is correlation to how many developers, but that's the first thing. Like
[00:21:06] Andrew Zigler: Hmm.
[00:21:07] Itamar Friedman: meta Facebook. I think even when they were big, like relatively big. I think they still had this moto move fast and break things. Um,
[00:21:17] Andrew Zigler: Right.
[00:21:17] Itamar Friedman: had some accidents quite a lot like just a few years ago and they were already huge. Right? While if you're SpaceX, for example, you probably are not advocating for move fast and break things. Also again, not if you're a bank. so I think the differentiation is more like, I do agree with what you say, that's like, what's your application? and I don't want to bash like, say anything about anyone, so please take what I'm thinking with a grain of salt. But I, I did hear from a company, very known company in the, observability cloud visibility. They said, Hey, if we're 15 minutes down, it's okay. Right? it might sound wrong because hey, observability helps others. to have that downtime, but if [00:22:00] they have
[00:22:00] Andrew Zigler: Right.
[00:22:00] Itamar Friedman: minutes, they feel relatively okay. Well if you ask a bank, are you okay with 15 minutes downtime? This is an example. Then, then not. Okay. So extracting what's important for you and then, you can point out what velocity means for you.
[00:22:14] Itamar Friedman: Is, is what you just need to think of.
[00:22:17] Andrew Zigler: So once a team identifies this, you know, let's, let's zoom in a bit on the playbook. there's lots of ways that teams can bring in AI to either. increase their speed or to reduce bottlenecks. but along the way, how do you think teams, um, are like, how should they best measure the impact of those things and prove it, to their own organization that like, Hey, we're implementing ai, not just because it's ai, but because it gives us this and this.
[00:22:47] Andrew Zigler: How do you see successful leaders navigate that?
[00:22:50] Itamar Friedman: Yeah. So look, let's let, let's admit it, I think in most cases right now, or maybe in early 2025, late 2024, the main [00:23:00] metric is, look, my developers are much happier. They're asking for it, et cetera. I think that it was true until just recently, like, early 2025. but I think like, I'm not the only one that's saying what I'm, what I said right now.
[00:23:12] Itamar Friedman: Like, think about your what. What matters for you, and then this is how you measure. And if you want me to be more specific, it could be, for example, time to, close, uh, a open pr and, uh, and how many, critical bugs are being introduced pair PR retrospectively. How much of your work is repeatedly, uh. what the percentage of work that you're doing to fix, bugs and issues. For example, you're starting a sprint, right? You're starting a sprint, and now you have, you, like, let's say you're using the framework for story points. How many story points are taken to develop new features and how many are to. Fix bugs or refactor, et cetera. It [00:24:00] takes some time. Sometimes, like it's, it's hard to really map, but over time, if you observe it over a week, 2, 5, 2, 3, 4 months, using AI tools. can be, you, you are able to map these and then, and then see if what you want to do is verify that you're developing a feature, but in high quality,
[00:24:18] Andrew Zigler: Hmm.
[00:24:18] Itamar Friedman: in addition to you will mix and match like a few of these metrics.
[00:24:22] Itamar Friedman: like how many features are, how many prs and features are, how many features are coming up, and how fast it is from a Jira ticket or a Figma until it's out and to the customers. But you'll also check. of, for example, how many, the percentage of, of user stories that are on new features over time and not fixing bugs that are because of vibe coding or something like that. is specific, but I do come back to my main point before that. It's very important that you sit down and, and think you match those metrics because there could be quite a lot tho those that are fitting what you define as velocity.
[00:24:58] Andrew Zigler: Yeah.
[00:24:59] Andrew Zigler: That's [00:25:00] really a good, really great insight for those listening too, to, to know that like for the, especially like larger companies and companies that are maybe in those more sensitive, uh, industries, for sure. Um, they want to be tracking and showing the success of this and, and part of tracking and showing the success is, you know, getting everyone on board.
[00:25:18] Andrew Zigler: Um, you know, there's a saying like it takes a village to raise a child. Similarly, it takes a team to implement ai. And so Itamar, in your opinion, who owns the success or failure of, of a rollout? And I obviously it depends on the organization, but what are some typical patterns you see?
[00:25:35] Itamar Friedman: I feel that we're already like 15, uh, minutes or so into,
[00:25:39] Andrew Zigler: I
[00:25:40] Itamar Friedman: and I didn't give like a futuristic prediction, so, let me give you one and then I'll connect it to your questions or your question. Okay.
[00:25:48] Andrew Zigler: Yeah.
[00:25:49] Itamar Friedman: let's assume. That in five years. I'm not saying if it will happen or not. Just assume that in, in five years we get to a point where [00:26:00] agents are, agents are writing and reviewing and verifying most of our code who is going to be there? at the dev organization, uh, guess it's the dev platform team. They're the one, the those that I see. it's not like that for all companies. It's like that. But the majority of that, like the, the, the organization that gets the responsibility to implement ai. Like you say, take a village, take a team that gets the ownership of implementing AI is the dev platform team. in some smaller companies, it could be just one or 2, 3, 5 people in some organizations could be. 5, 10, 15, or, or more. It's actually a growing, department because of what I'm seeing.
[00:26:48] Andrew Zigler: Yeah.
[00:26:49] Itamar Friedman: like in some companies we're seeing like it grew in one year from five to 25 in a, in a big company of 10,000, uh, developers. So back to my point, like I claim that [00:27:00] that organization is going be the zookeeper, if you like, or the agent keeper the, the manager of all those agents, they are going to be in charge of, steering, verifying, approving, uh, uh, I'm talking about like, not necessarily just the targets be actually, uh, credentials and things like that.
[00:27:20] Itamar Friedman: In, in addition to steering, other things. I just wanted to give that as an example. So assuming that's, that's the future, and maybe you could claim that only they will exist in five years from now, they're just gonna, not gonna be five people. They're gonna be much more, I actually see this. Uh, uh, already, uh, signs of it happening right now and I see them like thinking when there is such organization, usually I see much stronger, more, fundamental, more established, uh, implementation of, of ai. Otherwise, usually what you get is like developers saying, oh, I really love this code generation tool.
[00:27:57] Itamar Friedman: Let's install that. You know, and that's it.
[00:27:59] Andrew Zigler: [00:28:00] Right.
[00:28:00] Itamar Friedman: have this sta the platform team, or, or someone as in charge giving, then they're starting to think, oh, but how do we implement, how do we verify the best practice? How do we like also measure what's the real productivity? And there there's no like tech debt that is being accumulated.
[00:28:16] Itamar Friedman: I saw this like funny tweet, like two developers doing 20 x works, accumulating 50 x tech depth. so, so how you like that? That's, that's, that's what I'm seeing today that. that you're, you're seeing, a dedicated group. Usually the dev platform give giving the key to ownership. And they work in collaboration with the other engineers, other developers.
[00:28:38] Itamar Friedman: But they do, look at it a holistic approach of how do we do the, we do this software development with ai, but with confidence and, guardrails, et cetera.
[00:28:49] Andrew Zigler: This prediction is really important, I think, and especially since it's backed up by things that you're seeing, the growth of platform engineering teams, developers that are concerned about that experience from end to end, you see these [00:29:00] teams growing faster, than other teams on average. And it sounds almost like how you describe it and like the five year out scenario, you know, it's almost like a rising water and the platform engineers are at the very top and so.
[00:29:11] Andrew Zigler: You know, double clicking into that a bit, for our listeners who definitely see and perceive this shift happening, what are some advice, what's some advice or strategies you would give them to help them move upward into that platform engineering mindset? Are there upskilling opportunities that you see?
[00:29:30] Andrew Zigler: I'm kind of curious to get your opinion.
[00:29:33] Itamar Friedman: Yeah, so, so the very straightforward, answers put someone in charge on implementing AI across the SDLC a thoughtful way. Actually, a podcast and I think actually a lecture that one of the top suggestion right now in the market, forget about developers is take. let's say in sales, take your best [00:30:00] sales person and tell that person, that person actually the one that is excelling in sales. Stop selling. And now try to think how can you implement AI in sales to help all other AEs, other sales person. It's really hard to do because you're not, you don't want take that person out of the
[00:30:23] Andrew Zigler: Right.
[00:30:23] Itamar Friedman: holders or whatever the term is, quota carriers, right. but that would probably lead to the best, um, amplification of the sales team with, with ai. So, back to us, to developers. I think it's the same thing. Like try to think who is the best person, maybe the best, technically the best architecture. If you don't have right now someone in charge of implementing ai, I'm not just saying, okay, let's buy this code generation tool, but actually think about it thoroughly. then do that. That's the number one. Maybe trivial, but that's my recommendation. And then, [00:31:00] the second one is like, just continuation of this is like if I try to round up, wrap up everything we said so far. put a roadmap or. Or like a, a map of not, not a roadmap, sorry, a map of, all the processes that you have right now, probably just doing that will be helpful.
[00:31:18] Andrew Zigler: Mm-hmm.
[00:31:19] Itamar Friedman: put, put that role map and, and try to estimate where, where your bottleneck and then try to learn there and then try to learn about tools that are in addition to co-generation tools. That are specific for that, that's the uplift that I'm talking about.
[00:31:34] Itamar Friedman: There's so many AI tools for developers right now. I
[00:31:37] Andrew Zigler: Yeah.
[00:31:37] Itamar Friedman: that probably for each point in your software development lifecycle, don't know, RCA, uh, uh, management or, or just like root code analysis, uh, handling. There isn't a few even competitors on just in that sub-market. and some of them are, are awesome.
[00:31:56] Itamar Friedman: So you basically like. Nail that and, and then find a tool. [00:32:00] It could be very different than what you would read on Twitter or something like that.
[00:32:04] Andrew Zigler: Right, right. It's not definitely identifying what you need specifically. 'cause the market is going to become very vast with very specialized point solutions. And the better you get at identifying the wins that your organization needs. The more strategic you'll be about adopting them. That's good insight.
[00:32:21] Andrew Zigler: And what you also touched on about taking your best sales person, and, and throwing them into, you know, the, the AI mix and getting them to be like, how would you translate your sale expertise into AI tools or workflows? And let's help, you know, help everybody, take advantage of your velocity, your, that you have, in the way that you work.
[00:32:38] Andrew Zigler: Whether it's with AI or with not, you're taking that domain expert and you're helping, you're bringing them into the conversation to make these changes. and that's something that we learned recently from JJ Tang of Rootly. When he came on the podcast with us, he talked about, instead of going really hard on putting AI into all aspects of the product and productizing it, the biggest, uh, uplift they did was kind of build [00:33:00] it into their executives, by,
[00:33:01] Andrew Zigler: making everyone really curious about ai, having weekly meetings where people come together and share things, learnings and curiosities things they're trying out. And this is across the company, right? You get non-technical folks in these conversations and you become surprised at how fast they can move or the ideas that they can execute on.
[00:33:18] Andrew Zigler: so this sounds very similar to that as well, of making sure that you bring everybody into that conversation. And you know when teams are, when teams are doing that and there's like a lot of chefs in the kitchen, there's a lot of people working on it. You know, there's gonna be mistakes and people are gonna have to like realign.
[00:33:33] Andrew Zigler: And, uh, I loved how earlier you gave us like a five year prediction about like what might happen with, with developers eventually becoming fully kind of an automated process. Obviously that's like a very. elongated kind of, uh, dream about how that could evolve.
[00:33:47] Andrew Zigler: And, and so I, before we kind of wrap up, I wanted to, maybe zoom in on like the most upcoming part of that prediction, that timeline, as we, we continue through 2025 and come into next year. What do you think [00:34:00] the immediate future holds for developer teams? And what advice would you give a listener right now on this show to, get ahead of that curve?
[00:34:09] Itamar Friedman: so and, and I will use, we mentioned vibe coding. I think only once or twice. That's like a. A,
[00:34:15] Andrew Zigler: That's rare.
[00:34:16] Itamar Friedman: At, at, I, I think like basically we're going to see, vibe coding evolve to grounded coding and I'll explain.
[00:34:24] Itamar Friedman: Um, and, and, and after I explain that, I think like you, you'll realize and all the listeners, uh, if, if what I. I'm gonna say makes sense that you would need to, you would need still to know what development mean, what engineering means,
[00:34:40] Andrew Zigler: Yeah.
[00:34:40] Itamar Friedman: need to invest in grounding the, the AI in order so you can move fast, like with, with five velocity. So, to, to elaborate on this, um, hard to imagine that that vibe coding, like was. by carti, [00:35:00] just uh, a few months ago, and it didn't happen two years ago. Actually, like I would say cursor explosion. a lot because of, uh, that Kurt party tweet about vibe coding, where he mentioned specifically Cursor and Sunnet. if you remember that moment, that Sunnet 3.5 came that was really good for, for coding, right?
[00:35:20] Andrew Zigler: Yeah.
[00:35:20] Itamar Friedman: and then we see it is just a few months ago, like, around like early 2025 and I think that, It took only two months that we are starting to see backfire by
[00:35:33] Andrew Zigler: Mm-hmm.
[00:35:34] Itamar Friedman: the engineering.
[00:35:35] Itamar Friedman: Like the more, let's say it, it's still not like, it's not all the professional developers, it's not all the enterprise developers saying that, but we saw early signs of people saying, Hey, if you. doing vibe coding and enterprise software, you're going to have trouble. and that tweet of, uh, um, one, two engineers doing 20x productivity, but 50x tech [00:36:00] debtthat famous tweet that exploded was resonating for with many. And then interestingly, just a couple of weeks ago, we saw Karpathy tweeting saying, Hmm. code that I pro, I'm almost like quoting one word by word code that I professionally care about contrasts vibe, code, right? And so it's not like the pushback only about these, those developers.
[00:36:25] Itamar Friedman: Karpathy is a pretty good developer,
[00:36:27] Andrew Zigler: humbly
[00:36:28] Itamar Friedman: saying, then, and he's saying that, uh, like means something, but. He didn't stop there. He elaborated about how he thinks that needs to be sold, and then he said 1, 2, 3, 4, 5. And I wanna, elaborate, touch only two points. One is that noticed that he's saying 1, 2, 3, 4, 5.
[00:36:50] Itamar Friedman: This is a workflow
[00:36:52] Andrew Zigler: Right.
[00:36:52] Itamar Friedman: thing, and one thing that you need to do, is despite our immense, excitement about letting [00:37:00] AI prompt and just letting it do whatever it it wants to do, which is an agentic workflow, amazing agentic like flow, giving it a workflow. is like very, very useful.
[00:37:12] Itamar Friedman: And there are tools, shameless, shameless plug, also Qodo, that enables you to guide with a specific workflow. and then like, it's a guardrails, for the, that agent to work in a more specific way that you engineered. the way, as Coda, we, we have a open source tool called Alpha Coded, and, and, Coda is formerly known as Code Dium ai. And what we did there is we engineered a specific flow, to compete on coding competition. And it does much better than even the best model in the world to, to compete on coding competition, just letting it run free. Okay? So that's number one. And the second thing is that his point number one in this workflow is [00:38:00] gather all the relevant. An exact context
[00:38:03] Andrew Zigler: Yep.
[00:38:04] Itamar Friedman: that is really, really hard. So the second thing I recommend, in addition to like in 2025, like as a tech lead dev platform, or, or, or just an engineer like sync, what tools and what processes, et cetera could help you. to bring, you know, the right context, but doing that at scale.
[00:38:22] Itamar Friedman: Because if that starts be, it is, that's beginning to be like 99% of your work, then that's cumbersome. And here again, shameless plug, this is where Qodo excels, You know, the large code base scenarios, if you have more than 200 developers, probably you have more than 200 repos and with millions of lines of code, if not tens of millions above. And that's what we do. Like, uh, like in, like even there is a talk with Nvidia and GTC that they checked us and found that we're like doing 30% better than baseline that they can find. So that's like, uh, the, the, the two things that I would invest in 25 and how do I steer. The, the, [00:39:00] the A agent work, whether no matter where in the SDLC, it's doing code generation, code testing, code reviewing, and the sec.
[00:39:06] Itamar Friedman: And, and, and part of this steering is like around best practices, et cetera. And how do I automate like the, the context, uh, uh, fetching?
[00:39:14] Andrew Zigler: That's amazing insight. I love that you're able to, you know, dive a little bit into where Vibe coding's going to evolve. And so maybe we'll have to check in on that prediction later on as, as, as the conversation continues. And, you know, Itamar, this has been a really incredible conversation. I, I love how you think about AI and it's really insightful for me and, and our listeners.
[00:39:30] Andrew Zigler: And, uh, you gave us a lot of actionable takeaways too about how we should look at it within our own orgs. Um, but before we wrap, uh, where can our audience go to learn more about Qodo and the work that you're doing?
[00:39:41] Itamar Friedman: Yeah. So obviously, first of all, Qodo, uh, Qodo.ai. By the way, Qodo stands for quality of development.
[00:39:49] Andrew Zigler: Oh.
[00:39:49] Itamar Friedman: we don't say that often. Yeah. Um, and so that's the first thing. But, but obviously like within the website there, there is a block section and we actually worked hard. [00:40:00] to sector, like making it with sectors that are more, technology specific.
[00:40:04] Itamar Friedman: We share a lot of, for example, on rug, uh, retrieval, augmented generation, we share a lot. Like it just talked about context being as very important. so that's like, uh, uh, 1, 1, 1 thing I recommend the blog, like we, the differentiate section also, I do think that Andre car party that I just talked about as a good influencer to, to follow on tropic or really good in releasing high quality content. we also at Qodo have webinars and, we usually keep them really, really technical. We, I had a webinar with, on Tropic and we're having, a webinar with Google and, and, and that's a very, very like, technical about these topics. So that, that's, that's, these are my recommendations.
[00:40:43] Andrew Zigler: Amazing. We'll, we'll make sure to drop those in the show notes so our listeners can go and check out all of the community building work that you're doing, but also the resources you're gathering. It sounds really useful. and to you, our listeners, you know, if, if you made it this far, then like I always say, then you really like this conversation clearly.
[00:40:58] Andrew Zigler: Be sure to. Subscribe, [00:41:00] share the episode. Uh, reach out to us on socials. Itamar and I are both on LinkedIn. We'd love to hear what you thought about our conversation today. Um, I also really love how Itamar posts and has videos on LinkedIn. You should definitely go check out the content he's sharing. It's really great.
[00:41:12] Andrew Zigler: if you're only listening to this podcast and you're not reading the Substack, then you're missing like half the story. you have to go there and subscribe so you can get all the cool insights that we're gonna drop with this episode. but that's it for this week's.
[00:41:23] Andrew Zigler: Dev Interrupted. We'll see you next time.