Transcript: Dedy Kredo (Qodo): Developers as Orchestrators and the Future of Code in the AI Era

Dedy Kredo: [00:03] There's some very big shift in the role of developers. They're becoming a kind of orchestrators for agents, and basically every developer has their own set of agents, that maybe they'll even bring with them from one workplace to another. The working methods, how you get agents to work properly, to be online, how basically strong developers of ours, I'm seeing now, are learning to adopt these methods of working with agents, and truly becoming several X more productive. So basically the shift that developers are going to go through is becoming more orchestrators and reviewers of the agents' work, and through that also becoming much more productive,

Eyal David: [00:53] and also basically doing things that maybe weren't possible before. Hi, this time our guest is Dedy, CTO of Qodo AI, an entrepreneur and a smart, humble product person, for a conversation about the future and present of software development in the AI era. And of course about the early spotting of the opportunity that led the company already back in 2021, and about the focus on code quality and not just delivery speed. A super fascinating conversation with tons of insights. Enjoy. Hi Dedy, what's up? How are you? A pleasure — you walked here, you made it? Yeah, yeah, in our nice weather here. Cool, with us is Dedy from Qodo AI. So Dedy, tell us a little, in a few words, about the path that brought you to Qodo.

Dedy Kredo: [01:44] With pleasure. So basically the worlds of coding are something that fascinated me from a very young age. I started — the first language I think I programmed in was QBasic, in 8th grade, with some 286 computer, and basically my first job was with my co-founder Itamar, who was in high school back then, and he'd founded some company, it was a kind of startup building websites, late nineties, when the internet in general was just starting to arrive here in Israel.

Eyal David: [02:14] You pulled his hair, huh? What's that? Pulled his hair. Yeah, yeah, apparently.

Dedy Kredo: [02:20] And yeah, I was basically his first employee at the company, and it was the first startup he founded. And basically fast forward many years ahead, to VMware — I got to VMware through the acquisition of a company called Digital Fuel, and there I moved into a product role; basically I knew that entrepreneurship was what I wanted to do. The way I thought to get there at that stage was through product. And I think that when you're a product manager, you basically think like an entrepreneur, you're responsible for your area, and even if it's some product line within a big company, then maybe you have autonomy in your area, and that's the best course you can get before entrepreneurship, in my opinion, and that basically led me there. Before we founded Codium, I passed through a company called Explorium.

Eyal David: [03:15] You called it Codium because like... Right, right, it was... Right, right, it's... yeah.

Dedy Kredo: [03:21] Not Codium, Codium. Right, so basically the company is called Qodo, we founded it as Codium, and basically at the stage of... the last role I did before, was basically a VP of Customer Facing Data Science, a kind of Applied Data Science, you can think of it that way. I basically recruited a team of 20 data scientists, got into these worlds of data science, of models, of predictions, and basically I saw the area of LLMs beginning, and I understood that here there's now a shift. Crazy — is that what we're talking about? It was 2021 when I understood that there were going to be very, very big changes, and that there's a significant opportunity here, that there are going to be many significant opportunities. That was the stage when I basically decided to leave Explorium, and to found Codium, which later became Qodo.

Eyal David: [04:16] Yeah, crazy. Look, usually at this stage we move on to talk about the company, and that means, let's stop for a second — this is a hot field, an interesting field, give us a minute, if you can, your real take on the Future of Code, referring of course also to vibe coding.

Dedy Kredo: [04:30] Yeah. If you see it that way. Yeah, I think that developers, basically, there's some very big shift in the role of developers. They're becoming a kind of orchestrators, of agents, and basically every developer has their own set of agents, that they might even bring from one workplace to another. The working methods, how you get agents to be, to work properly, to be online, how basically strong developers of ours, I'm seeing now, are learning to adopt these methods of working with agents, and to become truly several X more productive. So basically the shift developers are going to go through is becoming more orchestrators and reviewers of the agents' work. And basically through that also becoming much more productive, and also basically doing things that maybe weren't possible before at the scale of organizations, of small teams or startups.

Eyal David: [05:36] Wow, that's crazy — set the reviewing aside for a second, how do you even manage a bunch of company-owner types like that now, right?

Dedy Kredo: [05:43] Yeah. It's unbelievable. And let me take you for a second — so really, if we talk for a moment even about vibe coding, where does it look to you like it's going? Look, I think vibe coding, which caught on like this, like wildfire — on one hand it's a concept that's very magical, I mean, there's something very magical about working in iterations against some AI, giving it instructions, and you basically flow with it somewhere, like seeing, you know, say something isn't working, so you tell it fix it, notice, this isn't working for me, this isn't working for me, and just continuing like that in iterations until it works. I think the danger in this is that behind it, in the end, you can build some kind of Frankenstein, and there can be some monster behind it that really doesn't — maybe it kind of works, it works, like, but the question is whether it's sustainable, whether the code is in a state where in real code of, you know, code that's, or mission critical — even not mission critical, the code of a serious company, that you want to now live for a long time and be maintainable and sustainable, I think this concept doesn't hold up, and I think what you need to do is find the processes that make this thing hold up, so that on one hand you can work in iterations against an agent, and on the other hand also make sure the code is aligned with best practices, to identify when you have two agents that clash and are now doing the same thing, or fixing the same bug seven times in different ways,

Eyal David: [07:26] duplicating the code across the whole repo. How do you get into this? I think it's something — sorry to interrupt — is it something you already thought about in 2021?

Dedy Kredo: [07:35] During COVID? I think we did — if you look, say, at our seed deck, then I think it has aged very, very well. Look, it's not that everything worked perfectly for us and that we did everything perfectly, in terms of the path we've traveled since we founded the company, but I do think that conceptually what we said holds water, and I think that's also an interesting point about how you sustain companies in the world we're in, you have to have some point of view, you have to be opinionated, you need it somewhere to hold water. And from the start of the road we basically identified that — it was in early 2022 that we started ideation, and basically we understood that software engineering was going to change a lot. We saw that the big thing back then was code completion. GitHub Copilot came out and started to catch on, and basically we saw that these LLMs know how to write code, or at least manage to complete code — that's what it was back then. And at this stage we understood, we said this is going to keep improving, and that it will succeed in doing more and more of the tasks that developers do today, that these LLMs would manage to do end to end, and then in such a world we basically tried to think, okay, what would actually be the big challenge. And we also bet that the whole thing would be this challenge, which we call code integrity, which is basically how I make sure that the code written by these LLMs, first of all works according to the requirements, according to the spec, does what it needs to do, and only that, and there's no edge case, and it doesn't add a thousand and one more things it didn't need to do, how I make sure it's according to the organizational standards, how I make sure it doesn't break anything, and how I do this at scale. So yeah, I think this did hold water from that standpoint.

Eyal David: [09:39] Yeah, along the way you've actually already gotten us ahead to the problem you started with, and not yet what the company does, but the direction it's heading, which is also interesting because, surprisingly — not really surprisingly — you're connected to the Development Life Cycle, and you're really sitting on the company's money, if I think about it. Because on one hand, as you say, let's be efficient, let's make sure the feature, or the product, does exactly what it needs to do — of course — but also, as you said, I'm producing some Frankenstein behind the scenes, I'm not producing a moat, and then again, we talked about a changing environment, right, that I need to move fast, so that's a weak point, that's a significant Achilles' heel in the company, in the product. Right. Cool. Just before we continue — click the link to the Product Builder WhatsApp group in the episode description, and join the community. There you can continue the discussion, ask questions, and be the first to get updates on new Tachles episodes. We're waiting for you there. And now, back to the episode. Tell me, what in your view — and of course this has changed a bit, so I'm curious — what in your view are the most important product skills in product people today? Yeah.

Dedy Kredo: [10:47] I think if I had to say one thing, it's adaptability — basically the ability to hold on one hand beliefs, and to be opinionated, to have some point of view about what needs to be done, but to be able to fail fast, and to pivot fast. Basically all the time you need to, there's a lot, a lot of noise — first of all you need, through some noise-filtering mechanism, as a product person today, to pay attention and understand what's happening in your world, to filter the noise from the signals, and to understand what the signals are, and basically to understand what, how you actually manage to penetrate organizations, or basically to do something that's unique, and that will hold up over time. Over time. It's not simple. If I were to say the most, most significant thing, it's the ability to cope with changes. Say, to build some roadmap, for a quarter, or I don't know, for two quarters, and wow, to be able, mentally, to make very big changes to that thing, because of the things that happen. I feel like today you're measured, no, today basically a lot of people, or until today, a lot of people were used to this, to define some roadmap, and as much as possible to try to stick with the roadmap, and basically to give developers, to give the organization the ability, the calm, to work on things. I think the strong organizations today, are organizations that know how to move very fast, both to execute very fast, and also to change very, very fast. I'll just say one last thing — I think you do need to know how to strike this balance, between a track, a fast track, where you do iterations and change things very fast, and throw things away, and swap in new things, and directions that, you know, strategically are important to you over time, and there you put some set of resources, that, well, you don't move them. They work on that thing. And also, basically as a product person, it's also some skill — it's important to identify these things: these are the things we're now putting resources on, and we're working on them now over the long term, and on the other hand here, these are things we need to know how to change very fast. Like, we're talking here to product people,

Eyal David: [13:06] but it's surely true for every entrepreneur, at the start of the road, and later on it's completely the situation now — so let's break down all these things for a second, there's something that really stuck with me — you've repeated it a few times, about a fast track versus the fast one, right — if once a company or organization had an innovation department that ran on x percent of the initiatives, today that's changing a bit, right? It's also even before that, still at the level of the initial discovery, like, of what the business does and what it needs to focus on, so maybe take us to what you did — maybe that'll be an example we can look at. Yeah,

Dedy Kredo: [13:39] so maybe I'll, let me lay it out for you a bit, basically, at Qodo we basically focus on helping enterprise organizations make sure that everything of theirs is aligned, both with the requirements, and basically with their best practices, that it's tested properly, basically helping them do review, of code at scale, in a world where every developer becomes 10x more productive — really, I actually see developers of ours today, it's crazy to see it. And basically that's the challenge we come to solve. So we basically have a platform very focused on the areas of review, we have a product called Qodo Merge that's a leader in these worlds of code review, and basically we also help developers create coding agents, to build coding agents that are very specific to their needs — custom agents — that basically fit all kinds of stages in the development process. I'll give an example: we have an agent that generates our release notes, that has a really specific prompt, specific tools we defined for it to generate our release notes, and if you go to... the release notes of our CLI tool — we also have a CLI tool called Qodo Command — all the release notes are generated automatically; it knows how to identify all the latest PRs since the previous release, to basically understand the changes, update the release notes, update the release notes website, and it does this in a very stable way. It does this in a very stable way because it's a custom agent, specific to this use case. So I really do believe in a world where there are agents that are specific to all kinds of tasks, maybe at the core they're very similar across different agents, but you define for the agent the permissions, the tools, the triggers, specific to its use case — so basically the two main things we do on our platform: one is really the whole world of code review, and two is basically creating, the ability to use, to build custom agents, and basically to connect them — maybe we won't get into the how yet — I'm glad I did this,

Eyal David: [15:46] because we really ran fast, so maybe we'll talk about the how, about how you understand context, but let's stick for a second — so we understood the context, and I was thinking meanwhile, okay, as an employee I'm producing all these agents, and then they're stuck inside the enterprise company, and I can't continue with them, I don't know — okay, let's set that aside, maybe we should think for a second about whether there's a way.

Dedy Kredo: [16:04] We already have a solution — so we have a kind of marketplace of agents now. That's exactly what I was about to suggest, let's skip it,

Eyal David: [16:10] so take me for a second, really, like moving fast, moving slow, like thinking fast — how do you actually how did you actually deal with this, do you have some story around it? Yeah, so basically

Dedy Kredo: [16:22] so if we think these are now areas — all these things, what's common to them is that you need very, very good organizational context. So we basically have a context engine, that lets us index in certain cases thousands of repos, and bring all this context to the agents, both for review and also for generation, by the way. And basically our example, say, is an area of context — we invest in a much longer horizon; basically there too we're going to launch a product called Qodo Aware, whose launch will be really soon — maybe it'll actually be after this episode is released, but basically it's basically part of a product that knows how to expose all the context, and it's basically agentic context, and it's an example of something we say — this is a track we're constantly investing in, developing, it's long-term, and basically it's one example. Maybe let's dwell for a minute —

Eyal David: [17:16] but basically you approached it, you defined it as a core product initiative, no matter what, because it produces a moat, and more than that — it's probably accompanied by the understanding that those enterprise organizations won't go and use AI companies to give them the context. That's an excellent point.

Dedy Kredo: [17:34] Right, but like, if we need a minute to formulate it, so... Right, it's an excellent point, because basically in a world where you're competing with the LLM providers — we saw this now, we also provide code generation, basically — and we see that this world, the whole world of code generation, is exploding with tons and tons of players, and the LLM providers too spotted the enormous potential there and entered it. So basically today the big LLM providers, which are OpenAI, Anthropic, and Google, have solutions in these worlds and are entering them, and then basically if you think — you take a step back and you think about it from the enterprise side, really the enterprise side, they're the giants of the world, some of whom are already our customers — so I think, okay, today whatever, Claude, Sonnet 4 is the best model for coding, tomorrow, who knows, suddenly it's some other one from another provider, and presumably the LLM providers too will want to build basically that memory layer and that context-collection layer, but I think that if I were sitting in the chair, in the cubicle of some enterprise, of the big banks, of whatever, I'd prefer to have more ownership over this thing, to have some kind of vendor-neutral thing, so to speak, that manages this, and it's very easy for me to switch models, when tomorrow suddenly, I don't know, some new provider comes out with a model, or one of

Eyal David: [19:04] the existing ones with a new state-of-the-art model. Cool, I think that's a really good example, we wanted

Dedy Kredo: [19:10] to bring out one more example. So one more additional example, which I think — if it interests you — with us, is the review; basically in the end we're measured in the world of code review, you need to give suggestions, and you have here some signal-to-noise ratio, that's very, very important. You don't want, on one hand, you don't want to miss problems — now we're basically, the way it works in our code review, in Qodo Merge, the moment a PR is opened, it goes into action, it scans the PR, brings more relevant context, tries, connects the organizational best practices, and tries to alert now either about potential bugs, or about areas where the code deviates from the best practices and things like that. And on one hand you want there not to be a situation where there's a severe problem you didn't catch; on the other hand, developers have pretty much 0 tolerance for irrelevant suggestions, so it's something we're constantly investing in, really the quality of the suggestions, reflection techniques, how you work with these models, how you give it the relevant context, but not too much — and the measurement done there is simply whether the code was implemented or not.

Eyal David: [20:16] Beyond that boolean, what else can you actually tell? How do you actually close the loop?

Dedy Kredo: [20:22] So closing the loop here — first of all, this is really the most significant thing, right? You implemented our suggestion, and that's what we measure, we do some tracking basically on the percentage of suggestions that get implemented, which afterward doesn't have to be crazy, it's not that you need to get 80 percent of the suggestions — I have to say, if you give, per PR, say 2-3 suggestions, and you build your UX in such a way that it's very easy for the developer to scan it in a few seconds, this yes, this no, this yes, this no — it's fine, so it's not so terrible if not all the suggestions get implemented.

Eyal David: [20:52] Let's clear something up — why is it fine, for a moment? Because I heard somewhere, I don't remember, just for context, today a million dollars buys 5,000 development hours for enterprise, in two more years it'll look like 90,000 hours at the same cost, so it's just not normal — so as you say, 100 percent doesn't matter, the 80 percent doesn't matter.

Dedy Kredo: [21:10] Yeah, what matters is catching the bug that will cause and lead to the next CrowdStrike.

Eyal David: [21:16] And basically bringing the world to a near halt — do you want to maybe say a sentence about it? Maybe not everyone knows.

Dedy Kredo: [21:22] Yeah, so basically CrowdStrike — when was it? I don't even know anymore, time flies, it was more than half a year ago maybe, something like that, it could be it was even last summer, I'm trying to remember exactly when it was — basically there was some PR that wasn't checked properly at CrowdStrike and introduced there some bug that basically brought down pretty much half the world — mainly Windows servers, that it turns out tons and tons of top companies are built on, and all kinds of critical services are built on this thing, and basically installing software that they push out, it basically brought it down for — I don't remember if it was already a day or more, and basically canceled flights, it was like a very big pain that was felt, but that's not the only thing that scared us in a bug that could happen — right, right, that's it.

Eyal David: [22:16] I think we've now also closed the second example, if I'm not mistaken — I think from there it's great, and like, what else we wanted to touch on in this aspect — because it's what you said, basically, we talked about innovation, about the team, about that part of taking bets — let's say it's now so different from classic product management, or even founding a venture, which is classic, right,

Dedy Kredo: [22:40] that you take a bigger percentage and now start throwing out bets, basically — yeah, basically it's kind of either win big or fail big, you basically have to have some point of view — I started talking about this a bit earlier, you need some strong point of view where you know how to say okay, I believe I have some insight here, that it's something people don't pay enough attention to today, and this is going — our example, say, was in the worlds where all the excitement was around wow, I give a prompt and it generates code for me and suddenly I have a working website and that, which is amazing — yeah it's really amazing — but we understood relatively early that for it to work in enterprise you need other processes, you need to be able to define best practices per team, per organization, you need to have definitions of what you're willing to deviate on and what you're not willing to deviate on, you basically need to have the ability to let developers take ownership of AI generated code, which usually is an unsolved problem — even today the X of the thing is how it's going to happen. You need to have some strong insight that you believe in, you need to go with it, and you really need to make big bets in order to do this. Another area that's like very interesting to think about is basically — if you're building a product that's GenAI — constantly thinking, basically, where this is going, meaning what the models will be able to do a month from today, three months from today, a year from today. It's very hard to predict this, but you need to have some point of view about this thing, and you need to find this balance between, on one hand, as a startup, you have to build things that deliver value today, right, you don't — you don't have the luxury to wait to build something for another year or two, and you need to find this sweet spot — I think this sweet spot is usually some few months ahead, meaning you don't want to invest the bulk of your resources on something that will be obsolete in two more months, right, and on the other hand, if you build too far out, it just doesn't work in the world of a startup, so finding this balance is hard, it's hard, and it's even harder, I think, because also the expectation

Eyal David: [24:58] it's hard, and for products of much higher quality, right — and what you're describing, let's run as fast as possible, let's build something, not that it should be scrappy, but that it delivers enough value and looks normal enough if it has some visibility, or that its results are good enough — combining between these is an innovation asteroid, I really don't know,

Dedy Kredo: [25:16] the whole company needs to rally to it — that's how it sounds, yeah, yeah, totally, it's some kind of mode where on one hand you move very fast but on the other hand you create some kind of mode where say, for example — I'll just give an example of what we do now — we put out a pre-release, meaning it's a kind of, like, you can think of it as a beta or alpha, that we release already and customers can, say, move to it, use it — we know it'll be a bit more fragile, more brittle, and they can go back, they can switch and go back to the stable release, easily. Also in your definition of success you need to be much more brittle, basically, you understand that it's either, like, up above it worked really well, or there's probably something in the middle. Now the question is whether it's worth investing in this thing, because it'll now be in a few more months, right?

Eyal David: [26:04] Yeah, yeah, yeah, yeah, yeah, yeah, so like it's really interesting — tell me, how did it work out for you, basically — very hard to predict what — you have more resources, probably, that you'd rather fit into your X and beyond that, but

Dedy Kredo: [26:16] how do you know how to predict today, really, what the companies, the models, will do? I can tell you the things I do: so first of all, I constantly try to listen to the folks who are there. So like there are all kinds of, you know, specific podcasts or things that, places where you can hear them, both at conferences and other places — I think it's important to listen to the researchers who are at the state of the art, at the big labs, as they're called. So I try to stay up to date with that, though of course they don't share everything. In the end you do need to constantly listen to all kinds of diverse information sources, it's both the researchers themselves, it's also people in roles at these companies, and it's also keeping up with research that comes out, finding some balance in your day-to-day, where you also dedicate time to staying updated on what's happening in the market, what's happening in research too, and from that extrapolating what will happen next. I think a good example, say, of what's happening now is basically the scaling being done for reinforcement learning, and I think we're still at the beginning of this process, which for me means we'll see rapid advancements in the capability of the models in verifiable domains. Coding, obviously, is one high use case for that, verifiable, and basically domains like coding, like math, and maybe also other domains that today are a bit on the edge, we'll basically see there very, very rapid progress in the models' capabilities, and in the models' ability to also do tasks that are more on longer horizons, and basically the more that models can work — say if a year ago, you wouldn't have imagined a model, some agent that works for you for an hour, and doesn't break along the way, right? And now you see it, right? Like, so once they know, the models and the agents that use them know how to run over longer time spans, that also enables much more advanced capabilities.

Eyal David: [28:31] To solve problems of a different kind, and maybe of course of a different — yeah, I hope. What it also made me think is that companies have probably started to hire someone like that full-time, whose whole job is that. Like, this reverse engineering, to aggregate all the alerts from every direction, and to start drawing conclusions from it, and to think about where it's going, to run models on the models, and economic models, I mean — it's really interesting. Yeah, either to hire, or maybe it's a balance, maybe an idea for a startup, what you're suggesting. It's possible that many financial bodies, that's what they're doing now, but within a company, you want to understand your ecosystem, and very fast, and to move and all that — you need some kind of intelligence person, or there's such intelligence, right? Interesting, maybe a service. So let's actually take and talk about Qodo, because we haven't talked about Qodo yet. So like, we understood the world of the problem you started with, tell us maybe where it unfolded, tell us who the customers are, who exactly the person in the organization is. Yeah, so really our focus is helping organizations,

Dedy Kredo: [29:31] mainly enterprise organizations, but not only. We basically have significant customers, and I think that, once it becomes an organization of a few hundred developers, it already starts to be a problem, that once developers start adopting agents, they start becoming much more productive. The question is, how do you actually make sure the code is aligned, first of all with the organization's best practices, and also, how do you even help developers understand what the agent did? For instance, there's a task, or you connected the agent that connects to your Jira, or your Linear, it now starts doing some task, and it could be that different teams also work with different agents, and developers here prefer, or one works with Cursor, and here Claude Code, and here who knows what. How do you build the frameworks around this whole thing? And first of all help developers understand what the agent did, and then also make sure there aren't any significant bugs here, that there aren't all kinds of agents that did similar tasks, and override one another and affect one another, and how do you actually handle the hygiene of your codebase over time? That's broadly the world of problems we come to solve. We have today a product called Qodo Merge, which is basically a leader in these worlds of code review. We recently released, not long ago, a CLI tool called Qodo Command, in which we also enable code generation, and we also enable creating custom agents, that tailor the use case, they're very specific. I touched on this a bit earlier. And it can be in the areas of testing, it can be in the areas of production error analysis, or release notes, or really creating a custom agent, that are tailored to all kinds of processes in the SDLC basically — software development lifecycle — organizational. And we also have a plugin for VS Code and for JetBrains. We basically believe in meeting developers where they are. That's, like, generally speaking, and in Qodo Command, we're now really also building this review experience, of basically how I onboard code written by an agent, and understand what this code does, and basically catch the problems there too. And basically a very big area that we — a very significant area that we invest in — is the whole context area, that we talked about. So we basically have some context engine, called Code Aware, that connects and does indexing, and connects for you to, I'll say, GitHub, GitLab, Bitbucket, whatever you use, does indexing of both the codebase and the documentation at scale, and basically knows how to connect this codebase, both to code generation, whether ours or others', and also to answer questions, and even to plan features end-to-end, new features. That's kind of the high-level platform.

Eyal David: [32:36] I really love that you told us basically what the company does, and along the way, underneath, a ton of challenges — I identified a ton of challenges, maybe you'll list them too — there's also the matter that you really work very close to the user, who is also themselves a developer, right, but you basically want to manage — not manage, but to be a very significant chunk that helps manage the life cycle in general, there's a ton of stuff here, how do you even get into an organization like that? How many customers do you have today?

Dedy Kredo: [33:02] Yeah, that's an interesting question, because we also have open source — meaning it's basically built on open source called PR Agent, which has tons and tons of usage, every time we discover another, whatever, big bank, or who knows what, that uses our open source, and we also basically have a freemium version, and a version that's kind of self-serve — I think this is very important for every dev tool company like ours, basically that you have the ability to try out the product, in a way that's really self-service, PLG is the classic term, it seems so important to me for a product where the time span is that low, yeah, and then like, basically one of my turn-offs, when I try to give something a chance, and I see contact us for demo, and I have no way to try it out, so that's why we really believe in self-serve, and in the ability to try things out — so yeah, we have dozens of enterprise customers, and hundreds, or even thousands of organizations, the smaller ones. And how fast do you manage to show value now to that developer?

Eyal David: [34:01] I don't know — who exactly are we after — after the developer, or after basically whoever orchestrates this in the organization? Excellent point,

Dedy Kredo: [34:07] basically our strategy, at the go-to-market level, we call it middle out, basically we have both top down and bottom up, basically with us we're both bottom up and top down, basically bottom up is what we just talked about, self-serve, and you can convert, and use, and put in a credit card, and like teams — and we see, several times, dozens, a team of a few dozen, that simply put in a credit card, and use it, and on the other hand, many times in these organizations, and especially there's so much rush, in the market that we talked about, which is one of the challenges — the really big ones — how you stand out above the rush, so it also helps a lot to come in top down, and indeed a lot of our messaging, and the value we deliver, is very resonating, with more the management layer of the organization, or the organizational architect, so basically, what we do see, in many of these organizations, is a kind of rise of platform teams, or developer experience teams, that are getting more and more power in the organization, especially since there are all kinds of agent teams, autonomous ones, they basically decide which vendors, which agent team they bring into the organization, and there's suddenly an enormous impact, for this group, of the platform team. So in the big organizations, the big enterprises that work with us, we usually sell to this group. Sometimes they do try it out, they do kind of, say, come to us after they've tried the product a bit, gotten in, we suddenly see we have some new user, from some big company, and then they reach out to us, usually, yeah, it's like, they tried it a bit, say, on their own, it seems interesting to them, and then they reach out to us, okay, now we want to do some proof of value,

Eyal David: [35:56] what does such a pilot look like, they pull back the curtain and then like, what does such a pilot look like in the morning?

Dedy Kredo: [36:02] Usually we try to do it relatively — it's concise, of course it varies from organization to organization, so we also have organizations where at the start there was some small POC, and then there was the pilot of a few months in the organization, and then in the end they bought some big license, for the whole organization, for thousands of developers — we have a few of those. But usually the standard is, basically they make contact with us, we have a kind of group of, we've now built a whole go-to-market — really amazing — that's based both in the United States and in Israel, and includes both the sales engineers and the salespeople, and basically such a pilot usually starts with some kickoff, we try to understand what their success criteria are, basically which metrics they look at, and of course we come with our recommendations, and with our best practices, usually it's at the level of a few weeks, we define some group of developers, success criteria is also, by the way, a very interesting topic, many times with these tools, they look at more of a kind of vibe, as I say, a kind of vibe benchmark, the developers like it, don't like it, it feels, doesn't feel, and I think here too there's some shift, that as time went on I see more and more, that it's more important to them to see tangible metrics. So for example, what percentage of the code, if it's code generation, then the classic metric is what percentage of the code that developers wrote, was written by an agent. If it's basically in the world of code review, then it's basically how it affected the review cycle, or how many problems it caught along the way.

Eyal David: [37:41] Yeah, like coming back to regular metrics, basically, of moving the business. Okay, wow, fascinating, and tell me, tell me, if you now had to bet, for another three months, from now, some, starting, not getting into your strategy, what would you go with in this world? Wow, amazing question.

Dedy Kredo: [38:05] I'm mainly focused on our areas, but I'll answer to the best of my ability. This world isn't specific to Qodo's world, but really all code writing. Or in general, I think it's very interesting — areas where, I really believe that the future, it's, the future of the whole combination of AI now in organizations and especially in enterprises, it's basically, how do I manage to do customization, of the output of the models to my specific processes. Personalization and customization, basically, more personalization — maybe that's a more correct word. I believe the future is personalized. And, I think that, ideas that are in the areas of doing things that are very personalized for me, it can be for me as a consumer, it can be for me as some kind of, I think the worlds of data are very, very interesting. I think there are things there that aren't, aren't solved, in everything related to Analytics — those are very interesting areas. Trivia, the — event. Yeah, I think that, or in general, how I explore my data, how I basically understand the data structure in the organization and create interfaces that are very easy to... I think there are very interesting topics to get into, and I think that really, the focus is to understand, okay, the models' capabilities keep improving, they know how to do more and more complicated things, but, how do I harness them to specific use cases and do it in a way that's very personalized? Cool, good answer.

Eyal David: [39:53] Okay, we talked about leaving this for the episode, but still, let's dive in for a minute, because it also basically gives an indicator of the temperature right now of things — let's talk, like, in a few words, about what happened with Windsurf, and really, around the acquisition that didn't happen the second time, and what unfolds afterward — where is it going, in your opinion? Are you hoping specifically about where it's going with Windsurf? No, I mean specifically about the fact that the giants are trying very hard to get in, for example, Apple in general, I don't know where they are, like, how is it going to look, in your opinion? Yeah. Yeah, I think the giants are totally getting in, we see everyone kind of, coming with their coding solution, and when you approach this field, you need to basically think in terms of which things, or which capabilities, you, as the enterprise organization, want to take from the big ones, or to give freedom of choice to your developers, to choose for instance which coding agent, one or another, and which areas you basically want to keep more under your control, or under the control of some vendor, that you trust to be more agnostic around the... I think it'll go in a direction like Microsoft and Office, and you know, that dominate these organizations, in the end, like they hold the whole setup, the question is, like, does it look to you like OpenAI made one such slash, or alternatively, that it'll really be a mix of small companies that won't necessarily be important, or a setup, like organizations will know how to work with all these vendors? I think it's... basically there are here two different worlds. If you look at pure code generation, basically I'm now giving some agent some very specific task, and all the relevant context, and now implement this — so I think this area is becoming more and more commoditized, and there are basically quite a few players, both players that are more like LLM-vendor agnostic, and also LLM vendors themselves. In some cases too, say Google is an example, they're both an LLM vendor, and also a kind of cloud vendor, right? We see that Amazon too is getting into this world. And of course OpenAI and Anthropic. But I think that if I'm in the cubicle, if I'm wearing the hat of an executive of an enterprise, who knows that right now the model today is Claude Sonnet, it's state of the art for coding, and tomorrow it's suddenly GPT, whatever, GPT next generation, and the day after it's Gemini 3, who knows. So I want the flexibility, to switch between providers, so I do think, because it changes so fast, so I think it's a bit different from the cloud worlds, where usually most organizations make some bet on one central cloud provider, although there are quite a few organizations, especially the big ones, that have several cloud providers, right? Several cloud vendors. Say they work with both AWS and GCP and also Azure, say, as an example. But because here it changes in an instant, suddenly there's now, suddenly tomorrow some new model comes out that beats everyone, and wow, I want it now, in the coming two months, it can, literally, affect my productivity, the ability of my development organization to work in the best and quickest way. I want them to be able to use this thing fast. So I think that... in these areas, I'd want to give my organization the choice. Meaning, developers could use the CLI of, whatever, of Claude Code, or tomorrow the Gemini CLI, and the day after Codex, and suddenly some, I don't know, xAI suddenly puts out some Grok or something that will be amazing. So I think, but on the other hand, the whole layer of basically my context, and my verification, and how I verify that... how I basically help do the code review, how I verify the code, how I basically use this at scale, and it makes much more sense to me, when I keep this closer, as an enterprise organization, keeping it closer, I think that means, either to use some vendor that also lets me do some self-hosting on my side, just as an example, or to use some vendor that I'm simply happy with, that we know through which I can use all the providers. Okay, that's probably a wow that you also identified in '21, right? Yeah. I have a question that came up for me along the way, let's continue — do you today enable companies to do benchmarking between models? Yes, we also enable choosing models, you can really choose which model you run both for the... both for the... and also for the... and we also have our benchmark. Yeah, but is it exposed to me, basically as your strength — now I run Gemini X and I run another model, whatever, and the user doesn't know it, and I look at which one, cross-company, performs better? Can I? That's a really cool idea. We don't have that today, meaning, the user sees what they choose. Really, in Qodo Merge they don't see, but it's some regional choice. In the area of generation, which is basically either agents that are either for coding or for specific processes, custom agents, then there you basically choose, the user really chooses. And we have our benchmark, which is a very interesting benchmark, that basically we have four hundred pairs, and we basically run models one against the other, and basically use some technique called LLM as a judge, but it's all real world examples, and basically every time a new model comes out, we can very fast see whether it's hype, or whether the excitement, you know, the Twitter trend of ten incredible unbelievable examples of what this can do. Whether it's really real, or whether it's just hype. Really, and you can also break out trends per company size, per whatever, gorgeous. Really. Let me run for a second — so basically we really talked about, we wanted to touch more, but we basically talked about building products today, or businesses today, where basically the how is very — it's much easier, basically. Yeah. You know a lot of your doctrine, it was gorgeous. Let's leave something also for the next time you come back here, I won't dwell on it. I do want the closing part called — let's touch on some thing I asked you a moment before recording, I asked you, but why did you come here, basically, right? And innocence. But I'm really glad to expect something. Yeah, so look, I think that if you look at this whole world strategically, I think it's very important that we have here in Israel companies that are at the state of the art of these worlds of AI, and I think that also specifically in this world of coding — basically the progress there is what will take us in the end to what you call AGI, super intelligence. Once models know how to do AI research, once they know how to write code at a very, very high level, that will really propel us forward, and I think that in Israel we need to be pioneers in this world, we need to lead this world, and yeah, I think there will be more entrepreneurs who engage with the state of the art in these worlds, whether it's coding, whether it's data, whether it's anything that's Deep Tech in the worlds of AI — I think it's very important for us strategically, also as a country. Really, really, that's it. I think, if it's hard for them to even get into it afterward, it's probably much harder since we finished a degree, um — and a recommendation, a final one, do you have some recommendation? Any recommendation I put you on, but feel free. Yeah, I think that, like, stay curious, all the time if you basically want, whether you (m/f) want to be entrepreneurs, or to get into the worlds of product in organizations, or in existing ventures, and basically to lead the products — so basically you need to constantly explore, to constantly think outside the box. I think it's very easy to reset to the hype today. There's so much noise. And my recommendation is, like, challenge things, try to think outside the box, try to think about what things aren't being thought about enough, or aren't being talked about enough. And... and I think whoever succeeds, in these worlds, is whoever manages to connect the dots, to understand, okay, this is where it's going, this is how it'll look a year from today, and... and these are the things, that if it's going to look this way, then people aren't thinking about them, and aren't solving them. And then basically, or that... we'll go back to her point, basically, that kind of succeed big, or fail big, like, if your prediction turns out to be right, you'll be in a good position. If not, then not, but that's what a startup is. Yeah, that's the hand you were dealt. Really. Great tip. Um. I had a really great time — this'll turn into an excellent episode, full of things. What a joy that you came. Of course, thank you very much, thanks for the invitation. It was fun. Bye bye. Bye. Hi, friends, thanks for listening. If you found this podcast valuable, you can subscribe to follow us, of course, for more episodes, on Spotify, Apple Podcasts, or any other app. Of course, if you didn't find us on some app, I'd love for you to write to us. We'd be very happy for five stars on any platform and for you to follow us, so that more listeners can be exposed to us and find the podcast. You can also find the previous episodes on any app, or on the YouTube channel, we have links in the description. Until next time, come on, efficient description, and bye bye.