The Developer Skills That Will Actually Survive AI

[00:00:00] Simon: Before we jump into this episode, I wanted to let you know that this podcast is for developers building with AI at the core. So whether that's exploring the latest tools, the workflows, or the best practices, this podcast is for you. A really quick ask: 90% of people who are listening to this haven't yet subscribed.

[00:00:24] Simon: So if this content has helped you build smarter, hit that subscribe button and maybe a like. Alright, back to the episode.

[00:00:32] Guy Podjarny: Hello everyone. Welcome back to the AI Native Podcast. Today we have a really exciting guest that I've known for quite a while here. We've managed to juggle schedules to make it work, and that's Thomas Dhomke, who until recently was the CEO of GitHub.

[00:00:48] Guy Podjarny: And now we'll talk about this and the brave AI dev startup land. Thomas, thanks for coming onto the show.

[00:00:55] Thomas Dhomke: Hey Guy, and thanks so much for having me.

[00:00:57] Guy Podjarny: So, Thomas, it's funny a little bit. As we connected here or reconnected, we've been chatting throughout the time, we talked a little bit about how we met initially.

[00:01:04] Guy Podjarny: It’s not publicly discussed, but there was a moment in which it looked like GitHub might acquire Snyk early in the journey. I guess that's how we got in a room.

[00:01:21] Guy Podjarny: Clearly, that led to interesting paths as GitHub got pretty big in the AppSec space. It clearly didn't happen on the Snyk side, which worked out pretty well. But yeah, those were fun and intense conversations. There was a way to start a relationship there.

[00:01:36] Thomas Dhomke: Yeah. So for the listener to have some context, it was summer 2018 to my memory.

[00:01:43] Thomas Dhomke: Microsoft had already announced that we would be acquiring GitHub. I was part of that acquisition team together with Nat Friedman, who then became the CEO of GitHub, and a few other folks, Mario Rodriguez, who's now the head of product for GitHub.

[00:01:59] Thomas Dhomke: As we had signed the deal, we were thinking what else would be a strategic fit for the GitHub portfolio, for the product, and our platform to be added to. We had a long list of interesting startups in the space and one of them was Snyk. To my memory, we had you over in the Twitter building, funny enough, where back then Nat's office was. In the conference room, we chatted a little bit about how your business is going and how the open source security space is important for GitHub as a platform and obviously for Snyk as a product.

[00:02:27] Thomas Dhomke: That's how we met the first time.

[00:02:29] Guy Podjarny: Yeah, I remember the conversation very vividly. We also dragged Nat at some point, I think it was before that, actually, to come speak at our Snyk all-hands as he was exploring possibilities.

[00:02:41] Guy Podjarny: It was interesting; you were in that annoying window in which you couldn't act, you could just explore because you were waiting for the GitHub acquisition to close. In the meantime, investors swooped in and gave us higher valuation opportunities.

[00:02:58] Guy Podjarny: Anyways, it's been fun times. Clearly, both worked out pretty well. GitHub built a thriving security business and Snyk has grown well.

[00:03:06] Thomas Dhomke: Well, if you could time travel back to 2018 and tell both of us what the valuations of developer tool startups are in 2025. In 2018, everybody thought the $7.5 billion price tag for GitHub was high.

[00:03:18] Thomas Dhomke: And the valuation for Snyk was growing quite fast in a way that a number of other companies felt "overpriced." In today's world, that's almost a seed round. I think there's so much excitement, and we'll get to that later in the pod, but there's so much excitement in the space that people now believe developer tools companies can become much bigger than back in 2018.

[00:03:44] Guy Podjarny: Yeah, I fully agree. I guess part of it is software eating the world, but also just a bunch of more developers and better learnings about how to turn these tools that were just described in technical terms into real business value.

[00:04:01] Guy Podjarny: Maybe it's just the "software is eating the world" trend manifesting in its own inevitability. So I guess you are back in the startup reality of it after GitHub. A lot of the conversation when I hearken back to that moment was on startups versus incumbents.

[00:04:25] Guy Podjarny: You have the large companies coming along that have all the distributions. In the security context, for instance, GitHub eventually acquired and launched Dependabot. Clearly, you can now flip a switch and suddenly all the repos get this functionality, which is this massive power.

[00:04:43] Guy Podjarny: And yet, common wisdom is that startups are where innovation happens. You learn the breakthroughs happen in startup land. The incumbents come along and I was curious about your view. You were in the heart of it from the incumbent land in GitHub, and now you're going back to a startup. What's the view you've built up in terms of where innovation will come from in this world?

[00:05:12] Guy Podjarny: Is it still that startups will innovate, or are incumbents clued in enough that you actually expect a bunch of the innovations to come from there?

[00:05:25] Thomas Dhomke: We see innovation everywhere. It's both the startup world and the incumbents trying to out-innovate each other.

[00:05:33] Thomas Dhomke: I'd love to think about competition and business as very similar to competition in sports. You have a leader. Right now in Formula One it is McLaren, but for the last four years it was Red Bull. That's effectively the incumbent and they have the challenge of staying on top of the game. They get the best people poached off to other teams because they have a career path there and can become the team boss, while at Red Bull they were stuck under the previous team boss.

[00:06:02] Thomas Dhomke: The same is true in our industry. The big ones make a lot of revenue. It's easy to forget when you're in startup land or when you're looking at these two worlds that the big companies fight a very different battle. If you want growth at Microsoft at double-digit percentages every year, that's double-digit billions of dollars that have to be added net new on top of the renewals, as everything these days is a SaaS business or a recurring revenue business.

[00:06:30] Thomas Dhomke: It's a very different game if you have to find $30–$40 billion in net new revenue versus your small startup that is happy when they get to the first one, two, or 10 million in revenue run rate.

[00:06:47] Thomas Dhomke: So I think that's number one: you have to realise they're not fighting in the same weight class. But then when you look into what they're actually building, the products and the features, it's actually in the same space.

[00:07:04] Thomas Dhomke: I think that's really what makes this so interesting. We see the big companies all adding AI features, building agents, and pushing really hard. In the middle are the AI studios like OpenAI and Anthropic, and even there, there are big differences in sizes of these studios, right?

[00:07:23] Thomas Dhomke: From really big like open AI to quite small and somewhere in the middle is Black Forest Labs. Then there are the companies that are completely new and they're building with these AI tools what you might call "AI native" from the start.

[00:07:38] Thomas Dhomke: I think that's a very exciting space because effectively, as a dev, you start your morning not by going into Jira or, if you're more lucky, GitHub Issues (and if you are even more lucky, maybe it's Linear). You go and open a terminal, and you open Cloud Code, Cursor, or Gemini CI.

[00:07:57] Thomas Dhomke: You brainstorm with the agent as the first step. That's a very different way of working than if you're in an existing codebase in a large company where you have a lot of Snyk findings that you have to work through, your product manager backlog, your technical debt, and your on-call rotation.

[00:08:15] Thomas Dhomke: I think both of these sides will innovate in their own ways and it'll be really interesting to see who can change the world faster.

[00:08:23] Guy Podjarny: I really like the lens on the dollar count. It is one of those that gets challenged with AI, which is the level of attention, right?

[00:08:32] Guy Podjarny: I was CTO at Akamai before founding Snyk and as we were building business cases over there, it was immediately like, "Can this be a fairly instant $100 million ARR business within the next couple of years?" because otherwise it's hard to really rationalise it within the system.

[00:08:49] Guy Podjarny: That definitely opens that window for companies to build up and prove out that a business will become that size. But it seems like in AI there's this sense that things can become a billion-dollar business in such a short window of time that the bets are legit to make even within the incumbents. That feels a little bit different now than it did in the traditional startup versus incumbent reality.

[00:09:20] Thomas Dhomke: I think it is and it isn't. The revenue numbers or the enterprise sales motion are also often used as an excuse that you don't focus on the product anymore and you're accepting that the user experience sucks. I won't call out the product, but I was trying to set up SSO for a tool that we want to use within my new startup.

[00:09:40] Thomas Dhomke: I couldn't figure out how to do that. You would think that's an easy thing these days to just connect your Google identity and have SSO for everybody. I'm sure somebody has a reason why it is the way it is. The challenge that we are all fighting is that we do not have enough time in our days.

[00:09:59] Thomas Dhomke: Everything, whether it's Netflix and YouTube, or work, or real-life things like Lego, it has become a battle for your wake time. You can only reduce your sleep time so much until it gets really unhealthy.

[00:10:15] Thomas Dhomke: Battling for wake time at work means setting priorities. I think, again, this is much easier at a startup, even more so if you don't have a customer yet. My new company is fully remote, so everybody is somewhere around the world.

[00:10:30] Thomas Dhomke: We recently met for the first time in person in Singapore. One of my opening remarks was, "This is going to be the happiest time in this company's life because we have no customers. We have no production system. Nobody is on call. We are just prototyping and being creative."

[00:10:49] Thomas Dhomke: Looking back in a few years, this will look like the happy time. In a year from now, hopefully, we have a product out there and we have found some customers who are even paying us some money. Then life is very different because you have to prioritise between the thing you want to do and the thing that is actually driving the business.

[00:11:07] Thomas Dhomke: That's what really big companies, or companies that have found some form of product-market fit, are fighting every single day. AI is shifting that to some degree because agents allow you to take over things that you don't want to do so you have more time for the things that you do want to do.

[00:11:23] Thomas Dhomke: Security findings, you mentioned Dependabot, those things get even easier because an agent can just solve that for you to the degree where you only have to approve at the end that it's ready to deploy. Cleaning up tech debt, I have the Right Coding book here from Gene Kim and Steve Spear.

[00:11:41] Thomas Dhomke: It has a lot of these anecdotes about how they're going into 25 or 30-year-old codebases and all of a sudden they're burning through the issues that they've never had any motivation to pick up. Ultimately that means you have more time to spend on something else.

[00:11:55] Thomas Dhomke: To build something cool.

[00:11:56] Guy Podjarny: I relate to the speed more than the business model specifically in AI. I think that is a change. To be able to build a business in a native mode, I referred to the first months of Tessl's run as the honeymoon period because reality hasn't hit you in the face yet.

[00:12:19] Guy Podjarny: You get to build those, but nobody has told you your product sucks yet or told you this isn't technically feasible. I guess there's an exception to that. I think both you and I were in a privileged place in terms of ability to fundraise, so not all founders are in that reality. But once you pass the fundraise, then those times are good.

[00:12:36] Guy Podjarny: Although there is something amazing about users telling you your product is amazing; that's very satisfying. Not having the user to tell you it sucks is two sides of the same coin.

[00:12:49] Thomas Dhomke: Well, you can't stay in that period for too long. The longer you are in the honeymoon, the more your anxiety grows that what you're building is actually something that nobody wants.

[00:12:59] Thomas Dhomke: You have to launch eventually to get that feedback cycle or the flywheel, as Amazon calls it. Both of these things are true. If you're able to fundraise based on reputation or based on the idea you have, and the money question gets out of the picture, then you have a certain period where things feel really good until you have to launch and you're faced with reality.

[00:13:25] Guy Podjarny: Yeah, with the reality. So what can you say about your new startup and what made you take the leap from where you were in GitHub to this world?

[00:13:39] Thomas Dhomke: What made me take the leap explains where we're going. I started becoming a company founder in the developer tool space back in 2009. I left my job at Bosch in Germany, one of the big automotive suppliers.

[00:13:59] Thomas Dhomke: If you have a car with ultrasonic sensors in the bumper, those are most likely from Bosch or its two competitors. I left because I was so excited about the iPhone SDK and the App Store. I wanted to become an app developer again after being in automotive for six or seven years.

[00:14:17] Thomas Dhomke: From being an app developer, I became the startup founder of a company called HockeyApp that provided a developer platform for mobile app developers. This involved collecting crash reports and feedback. We sold that company in late 2014 to Microsoft.

[00:14:38] Thomas Dhomke: It's been 11 years since I was last a startup founder. Ever since my HockeyApp days, I have been in developer tools. HockeyApp was a developer tool. Then I was at Microsoft together with Nat building App Center, which was bringing the pieces of Xamarin Test Cloud and Insights together with HockeyApp.

[00:14:57] Thomas Dhomke: Then we bought GitHub. I was VP of Special Projects at GitHub before I became CEO. Through that journey from mobile to the cloud to Copilot and AI, I have seen how software developers have changed the world. The really exciting thing about being a software developer is that as you're building new products, you're constantly also evolving how you do that.

[00:15:23] Thomas Dhomke: If I remember my first coding exercises in the early nineties, first on an East German Robotron computer and then on a Commodore 64, what software development is today is fundamentally different. If I would give my kids a Commodore 64, they would probably figure out how to play games or something like that.

[00:15:43] Thomas Dhomke: But typing BASIC? They'd ask, "What the hell is this user interface?" and "Where's my Copilot or Stack Overflow?" or just Google to look things up.

[00:15:59] Guy Podjarny: The internet was missing. Big point. You couldn't go back to developing without it.

[00:16:02] Thomas Dhomke: Correct. We have always had these moments where the internet changed everything, open source changed everything, mobile changed everything, and the cloud changed everything. AI is also changing everything. I actually think we are still at the beginning of the revolution now.

[00:16:14] Thomas Dhomke: Copilot started more than four years ago. ChatGPT came more than three years ago. Last year was all about chat and agent mode, and this year clearly it's all about full-scale coding agents, code review agents, and security agents. I still think we are at the beginning of a journey where the way the software development lifecycle works is going to fundamentally change.

[00:16:37] Thomas Dhomke: That's what I'm excited about and that's what we're going to explore in my new company.

[00:16:41] Guy Podjarny: Cool. There's so much. There's always an underestimating natural reaction when you think about the scale of the transformation. What's the cliché? People overestimate the short term and underestimate the long term in terms of the potential of it.

[00:17:03] Thomas Dhomke: It's really hard to predict. If we go back a year into December 2024, who would've predicted that at the end of 2025 developers are back in the terminal with Cloud Code, Gemini SCI, Codex CI, and Co-CI? They have a grid of six or eight terminals open with all these agents running in parallel, trying to figure out how to make that work with Git worktrees, how to merge all the results, and how to not have the agent delete all the files on your computer or the work of the last 30 commits.

[00:17:38] Thomas Dhomke: That's where anyone who is predicting what that looks like in a year or two is going to be fundamentally wrong. That is really what makes it so exciting. We don't know where we're going, other than that we do know that we are going to keep building software while maintaining billions, if not trillions, of lines of code that are already out there.

[00:17:57] Guy Podjarny: Yeah, and that are growing. Also, part of the challenge with the tools, I'm still reading into it, is there's a fundamental conviction when you make this move that a lot of the ability to explore unencumbered is just better to do in the dev space.

[00:18:12] Guy Podjarny: It's interesting to see GitHub because when I think about innovation from incumbents, GitHub Copilot is a great example of innovation that came from within the company and pioneered that new space of coding assistants that feel like an AI tool.

[00:18:30] Guy Podjarny: For a while, it was there. Even with all of its reach and all of its strength, there was the opportunity for Cursor, most notably, to thrive.

[00:18:47] Guy Podjarny: Copilot continued to thrive, so it wasn't really at the expense of Copilot, but from a market share perspective, Cursor grew even faster and took up that space. Then indeed you have the disruption coming from Anthropic. It’s hard to think about them: are they a startup? Are they incumbents?

[00:19:03] Guy Podjarny: Those are very weird beasts. Organizations like OpenAI and Anthropic are massively valued organizations but still very much a startup in many ways. It overhauls the fundamentals of whether it's going to be the terminal versus the IDE.

[00:19:26] Guy Podjarny: I guess there's a perspective that when you're looking to disrupt, especially when you're disrupting the way things are done, not just automating existing workflows, but rather changing the way that someone works.

[00:19:43] Guy Podjarny: Those are precisely the places where it's hard for an incumbent to accept it. They have all the power in the existing way of working, and so it's hard to accept that change will come.

[00:19:59] Thomas Dhomke: I think that's the outside-in perspective where you look at companies that have been leading for a long time.

[00:20:07] Thomas Dhomke: And then argue they are being disrupted or seeing more competition, and you feel sad or worry about the people that are now in that position where they're no longer the best in the market.

[00:20:25] Thomas Dhomke: I think the inside-out perspective is often not as dire as you may think. I often thought when I worked for Mercedes that the best thing that can happen to us is that BMW announces a new feature or a new model with a new thing in it that the Mercedes leadership didn't want to fund. All of a sudden, you have an argument to get the funding.

[00:20:45] Thomas Dhomke: And be able to say, "Look, our competitors are doing this. We better wake up and try to catch up again." You saw in the news that Sam Altman declared a "Code Red" and Google had a "Code Red." I know Satya Nadella has had a bunch of "Code Reds" over his tenure.

[00:21:06] Thomas Dhomke: Competition is good. It actually makes a business and life more fun. We see that everywhere. Coming back to my Formula One example, it was Red Bull that won for four years, and it was Mercedes for eight years, and then Red Bull had another four years. You want these changes. It motivates those that have been winning for too long to come back and see how to transform their organization to be a winner again.

[00:21:30] Thomas Dhomke: You mentioned Cursor and VS Code. Actually, the IDE space has been phenomenally stable over the last decade. I remember when I started on Mac in 2007 or 2008, I started on TextMate. TextMate was the revolution back then. As TextMate became stale and TextMate 2 took way too long, I switched to Sublime Text.

[00:21:49] Thomas Dhomke: Then VS Code came, and for the last decade, it was the default editor or IDE for almost everybody. There are obviously some Emacs and Vim users, but now we have real competition again.

[00:22:06] Thomas Dhomke: It's not only Cursor and VS Code with Copilot; Windsurf is obviously in the play now. Recently, Google and IDX came out. There is Koding, Zed, and so on. There is full competition, and that means somebody sees an opportunity there to disrupt and to innovate and ultimately build a sustainable business.

[00:22:27] Thomas Dhomke: I think that is exciting and not something to worry about, both as the incumbent and as a disruptor.

[00:22:33] Guy Podjarny: I agree. I'm not worried about it; in fact, I like the changes. All of these require a moment of disruption that makes sense and is hard for the incumbent to follow.

[00:22:43] Guy Podjarny: You're right that competition is the greatest motivator and the opportunity to drive.

[00:22:51] Thomas Dhomke: And it did happen. It pushed us in 2024 to go from having one default model in Copilot to offering multimodal choice.

[00:23:02] Thomas Dhomke: And it did happen. It pushed us in 2024 to go from having one default model in Copilot to offering multimodal choice. At GitHub Universe in October 2024, I announced that we support Claude 3.5 Sonnet and Gemini 1.5 Pro. In the meantime, if you open Copilot, it has a whole list of major models, and then you can bring your own model and connect OpenRouter and all these kind of things.

[00:23:27] Thomas Dhomke: You can see the direct impact of competition in the market leading to a better product for the incumbent as well.

[00:23:34] Guy Podjarny: That's actually really interesting because one of the big questions for GitHub, we don't have to over-rotate here, is that GitHub has managed to be "Switzerland" in many ways for a long time.

[00:23:45] Guy Podjarny: It was like, "Hey, we're gonna work with all these IDEs." There was VS Code, but it's not a money-maker; it's a free service to the ecosystem. Now with Copilot, it's a little bit different. Microsoft as a whole makes a lot of money from being the provider of these different models.

[00:24:03] Guy Podjarny: Not necessarily the builder, but the executor. A lot of the inference running on Microsoft is a substantial revenue source. So that question on neutrality, do you want to play nice with everyone as long as you use my agent?, is interesting.

[00:24:23] Guy Podjarny: It's not entirely new; it existed in GitHub in other ways, like in the App Security space as an example. But it feels stronger now in terms of that internal conflict.

[00:24:40] Thomas Dhomke: Well, Switzerland also still wants to make money, both the companies there and the country itself.

[00:24:48] Guy Podjarny: None of this is critical; it's just a change.

[00:24:52] Thomas Dhomke: If you look at Microsoft, it is probably the biggest platform company out there.

[00:25:00] Thomas Dhomke: It's not only Copilot and GitHub that have some form of neutrality offering choice. If you look at Azure AI Foundry, which has a huge range of models available, it's not just the OpenAI and the Microsoft-developed models.

[00:25:17] Thomas Dhomke: In that sense, maybe the Switzerland analogy is as good as just saying it's a platform company. Platform companies by design want to have as many players on top of that platform as possible

[00:25:35] Thomas Dhomke: GPU inference as an example, or CPU compute, storage, or what have you in Azure. That helps both the platform providers like Microsoft and the partners to thrive. I think the leverage that Microsoft has as a business is the scale of that platform.

[00:25:53] Thomas Dhomke: That's the opportunity for every startup to become part of the Microsoft startup program. I should ask you: have you joined the Microsoft startup program?

[00:26:05] Guy Podjarny: We're in all the programs because they all commit until you get paid.

[00:26:10] Thomas Dhomke: At some point, every startup, especially every B2B startup, comes to the point where they either want to leverage Microsoft's distribution channel and their amazing enterprise sales team, or Amazon's, or Google's. The scale of these sales teams around the world and their skillset is so far ahead of what a small startup can ever deliver without going through the pain of building out subsidiaries and local offices and local meetups.

[00:26:39] Thomas Dhomke: There is a lot of human investment that has to happen that you can't replace with AI, because enterprise sales fundamentally is still a relationship management exercise.

[00:26:52] Guy Podjarny: Yeah, super valuable and the marketplaces and access to all of that.

[00:26:57] Guy Podjarny: Let me veer us off, there's a world where you and I can geek out on the business side, but let's make this a little bit more technical. We just established that it's hard to predict, but let's make some predictions here.

[00:27:10] Guy Podjarny: One of the things I'd love your opinion on is the scope of products in the era of AI dev tools. The first one is the delineation between reviews and collaborative interfaces. Maybe the GitHub interfaces are still there, but versus earlier on-premise.

[00:27:32] Guy Podjarny: We see you can use agents locally. You can use agents that are background agents, the Devins of the world, now there's a whole variety of others, that you'll meet to the code at the PR to begin with.

[00:27:53] Guy Podjarny: Then you see all the review agents. I had Merrill from Graphite on the show, and we talked about how they're adding, and that's true for everybody, an ability to interact with the agent to not just find a problem, but also fix it right there on the spot.

[00:28:14] Guy Podjarny: There's the question: "Well, if you could find a problem at that point, couldn't you have found it beforehand?" A lot of the lines blur because automation is increased. What's your lens when you think about the division of products, even between local devs, reviews, and collaborative?

[00:28:39] Guy Podjarny: How do you picture these things evolving?

[00:28:45] Thomas Dhomke: We always picture those things as somehow collapsing into just one unified stack or suite of products. The reality is that never actually happens. It's a bit of a pipe dream of every head of engineering or engineering manager that they have just this beautiful stack that everybody in the company is using.

[00:29:05] Thomas Dhomke: You don't want to worry about outdated products or conflicts where one product does this and another does that, and you're buying multiple times. Let's be real: there will always be overlaps in what companies and products do, and there isn't a "one size fits all" solution.

[00:29:24] Thomas Dhomke: Sometimes it's based out of necessity because moving from one tool to another is painful and you'd rather not invest the time and money of your engineers just because you want something standardised.

[00:29:43] Thomas Dhomke: . To a large degree in engineering, we will have to accept that entropy always exists in software. Your stack always gets worse until you invest time to remove tech debt. One of my jokes as CEO has always been that "debt" is actually the wrong term, because debt is something you are supposed to pay off over time. In tech, we just refinance into the next big thing, and then there's always the next big thing.

[00:30:10] Thomas Dhomke: Before you know it, you have new debt, even though it just factored from the old debt into the new debt. So it's more of a refinancing.

[00:30:18] Guy Podjarny: In the business world, the word for that is "leverage," right? It's just a different lens.

[00:30:20] Thomas Dhomke: It's similar to your house and your kitchen: you never get out of the mode of putting the dishes into the dishwasher and out of the dishwasher unless you stop eating, and then you die.

[00:30:34] Thomas Dhomke: That's true for software, too. Unless you stop building new features into software, you're always going to be in a state where you have to clean up afterwards. Coming back to your agent questions, I think you're going to see more agent-to-agent collaboration.

[00:30:51] Thomas Dhomke: You mentioned Graphite and Merrill. That agent right now gives me, as a human, feedback on code. But if that code was written by Devin, why do you give that feedback to me? Give it to Devin and have Devin figure out how to make that work. That's how we would do that in a human-to-human developer team.

[00:31:13] Thomas Dhomke: We would tell the two engineers to work together until they have a stable state, and then they can come to me as the manager of these two humans, or these two agents, to say, "Hey, we are finished with our work." So I think that's going to be, uh, um, you know, you know more, we are going to see more of that, uh, uh, like the agents working with each other. We are going to see more of the agents working with each other to not make the human the bottleneck within the lifecycle.

[00:31:33] Thomas Dhomke: Simon Willison on his blog recently had a statement, I'm paraphrasing, but it was that we have to get comfortable with not reviewing every single line of code before we deploy it into production. The sheer amount of code that a single agent can write 24/7, combined with running 10 agents in parallel all the time.

[00:31:58] Thomas Dhomke: This means you are never going to be able to review all the code without the human becoming the major bottleneck and effectively erasing the productivity gains from these agents. You have to think about how you automate other processes down the line to be able to generate all that code in the beginning.

[00:32:16] Thomas Dhomke: One thing that's also underappreciated is that a lot of these coding agents are actually really good at brainstorming. A lot of my developers are using Cloud Code to write code, but before they do that, they also use Cloud Code to brainstorm with the agent to learn about a topic or a special API.

[00:32:40] Guy Podjarny: Or even implementation options, right?

[00:32:42] Thomas Dhomke: To write an architecture document while everybody else in Australia is asleep, so that when they wake up, you already have something ready to share so they can pick it up and keep working.

[00:32:56] Thomas Dhomke: Again, this is very human-centric. Brainstorming: you have to tell the agent what you want to work on. I don't see that being replaced anytime soon. The idea still has to come from the product manager, engineer, or founder. The "how" is an iterative process with the agent, but it's not going to be replaced by an agent making up the idea, telling another agent to implement it, and a third agent to review and deploy it.

[00:33:25] Guy Podjarny: The agent-to-agent concept is one I relate to. It's still interesting when you think about the dev tooling ecosystem to think about who does this. You can run Cloud Code as all of these agents; you can run any of these, like Codex.

[00:33:41] Guy Podjarny: You can run that in the mode of "I wrote the code and run it with another persona." In fact, sub-agents do that: you are delegating work or passing it along to some agent that has the right context to represent expertise. You can choose different models.

[00:33:59] Guy Podjarny: But if the agent is a human, from a company's perspective, it might be one company that is providing the agent and you're connecting it in all these different contexts. I relate to that, and I'm curious because I don't know that we precisely know what the flow is for the interface points for us to engage with larger volumes of work.

[00:34:25] Guy Podjarny: Review as a concept will probably increase in importance because you do less of the work; you review the work of AI. But a lot of the current software development review activities will go away. That's not just for that; it's also for testing. I'm just picking on that as an example.

[00:34:48] Guy Podjarny: I'm curious and almost a little bit concerned about a collapse of a lot of these different tools. The reason I'm concerned about it is your comment before about competition: you don't want it to just be a competition between five giants.

[00:35:07] Guy Podjarny: You want a composable ecosystem in which different pieces can kick in. What are the right delineations for a company, to the extent where we can guess them?

[00:35:28] Thomas Dhomke: If you think about it, it's more composable than ever. Nothing is stopping you from using Cloud Code today, Gemini CLI tomorrow, and Codex on Sunday. It is very switchable. There's almost no switching cost. Same for the IDE, and same for the code review agent.

[00:35:45] Thomas Dhomke: Again, this goes back to the platform concept. The platform isn't changing. You're storing your code on GitHub, GitLab, or Bitbucket. But on the agent side and the AI side, everything is composable and the space is moving so fast. As we are recording this yesterday, GPT-5.2 came out.

[00:36:11] Thomas Dhomke: Until that point in time, everybody was talking about Gemini 3 Pro being the best model. Now we're figuring out if 5.2 is now the best model. By the time maybe this pod comes out, there's yet another model.

[00:36:25] Guy Podjarny: Maybe there's something else.

[00:36:26] Thomas Dhomke: Everybody will be like, "What the hell are Thomas and Guy talking about regarding what the best model is?"

[00:36:30] Thomas Dhomke: There's almost no switching cost and that makes this idea, that I can compose my pipeline with the best tools available, so much more realizable than it was in the past when you did have vendor lock-in. If you build your frontend on React, there's really high switching cost to go to a different frontend framework.

[00:36:53] Thomas Dhomke: When you're building your CI/CD on GitHub Actions, there's really high switching cost moving that away to another CI/CD provider. So many other parts of the developer toolchain have high switching costs, but with AI agents it is really, really low. That creates this competitive environment where what was the best thing six months ago may not be the best thing now, and whatever the best thing is in six months, nobody can predict.

[00:37:18] Guy Podjarny: I would challenge a little bit your comment that GitHub, GitLab, and Bitbucket are still the place you store and it's just the agent that you play around with. I would say that the platforms are optimised for a type of work.

[00:37:33] Guy Podjarny: The infrastructure estate management tools that existed pre-cloud were not the right tools to manage a cloud estate because in the cloud estate, hardware was elastic. You can spin things up and down, and therefore identity of these machines was different because they were short-lived, with immutable infrastructure and all that.

[00:37:56] Guy Podjarny: Now we have a different breed of developers. The collaboration happens between agents. What type of tracking is needed? It's hard to not think about Git as that because Git is so ingrained.

[00:38:15] Guy Podjarny: But I think we're both old enough to have lived through Subversion where Git felt like a heresy. Fundamentally, you are versioning, storing, and collaborating using different methodologies and different competencies.

[00:38:34] Guy Podjarny: I find it really curious to think about what the right way to version is and should it be? Maybe this is part of my thought on pull request reviews: is a pull request even the thing to represent anymore?

[00:38:52] Guy Podjarny: Is that really the point of collaboration between agents, or is there something quite different?

[00:39:05] Thomas Dhomke: I think the question for this collaboration platform is: do I still have to go there to figure out what work is assigned to me or what my team is working on?

[00:39:13] Thomas Dhomke: Or do I head into a world where I just ask my agent what's next on my backlog or what Guy is working on? You're heading into this world where your agent or your assistant becomes like Jarvis from Iron Man and effectively knows everything and has sub-agents to access certain skills.

[00:39:36] Thomas Dhomke: You would basically have your personal agent that helps with travel booking and calendaring and has access to your coding environment. All of a sudden, you don't actually open the browser as much anymore and open all these tabs on all these platforms where your leadership team told you your stack is and where you have to look for work.

[00:39:55] Thomas Dhomke: At the same time, there are still people running COBOL for mainframes. Most of these COBOL projects have no DevOps whatsoever. They don't use Git. They don't have CI/CD. They certainly have no Snyk or test cases and test suites.

[00:40:18] Thomas Dhomke: If you take that as the most extreme example, you'll realise that as software developers, we live on a spectrum. On one end, there's the really old stuff that still powers a huge part of the banking industry and local and federal governments. In between, you have people working on Windows APIs for display on train platforms, Windows 95 APIs, not Windows 11 APIs. So you have that spectrum.

[00:40:46] Thomas Dhomke: There is going to be the state-of-the-art developer that uses all these agents and doesn't have to go into the web anymore because all they do is talk into the computer like Mr. Data in Star Trek and get the job done. But in between are all the other people that still need everything else that we built in the last 30 years.

[00:41:04] Thomas Dhomke: As such, that is really a gradual change that's happening. It's not this "AI is writing 90% of all code now and we don't need all this old stuff anymore." I think as long as we are aware, when we are talking about how exciting these agents are, that in the middle there's a huge spectrum of other work that still exists that engineers have to work on.

[00:41:27] Thomas Dhomke: I think it's both exciting and also grounding, how much work there is going to be for the next century to come.

[00:41:35] Guy Podjarny: Yeah, for sure. That sounds very aligned with this idea that it's more of a "decade of the agent" versus the "year of the agent," to use Karpathy's quote.

[00:41:44] Guy Podjarny: And it's true. I sometimes say how application security companies are like cockroaches. They never die because technology never dies, and you need to secure all the way back to the Middle Ages, right? You need to provide support for that.

[00:41:59] Guy Podjarny: Clearly, none of the new companies wants to go there. That is the inevitable growth of any enterprise software company. But specifically for that, you unintentionally make decisions now that you have to go historically.

[00:42:11] Thomas Dhomke: You unintentionally made a bit of a cynical statement about your own company.

[00:42:15] Thomas Dhomke: It's so it could live because they never die.

[00:42:18] Guy Podjarny: First of all, I'm very happy for Snyk to be immortal. I'm happy with that. But no, it's also that you just see it. I've been in AppSec for over a decade before I started Snyk, and I've seen all these companies that came and toned down, but they never went away because of those reasons.

[00:42:35] Guy Podjarny: I think it's interesting and I agree with you; technology is very different. I guess one topic that I feel like this leads into is the humans in this whole movie.

[00:43:00] Guy Podjarny: When the cloud came along, I guess you could appreciate that there are historical ones, and maybe you made a version of a choice like, "I'm going to really specialise in this mainframe stuff because it's going to be a big money maker. I'm going to suffer, but I'm going to make a lot of money and I'll specialise over there."

[00:43:16] Guy Podjarny: I'm slightly cynical here, or maybe a lot cynical. Or, I can go and immerse in the new world of cloud. There was maybe a little bit more clarity, at least at some point, around the skillset that you need to build to be a cloud-native developer and to be a capable professional in this new era.

[00:43:37] Guy Podjarny: AI is a bit more murky than that. A bunch of these predictions are beyond intellectual curiosity. Of course, for you and I, there are bets as we build dev tools in companies; you have to have a theory of the future and then be very adaptable.

[00:43:53] Guy Podjarny: But you have to have a theory of the future. For so many more people, there's a question of: "I'm a developer. What would my job look like? What would be my competency in the future?"

[00:44:14] Guy Podjarny: What's your view on that? If someone comes along and says, "What should I get great at? How do you envision the job of a future AI-native developer?"

[00:44:23] Thomas Dhomke: Ever since I went to university, I felt like you're actually never done with learning as a software developer. There is no stable state, and those that feel like they have reached a stable state are ultimately planning for their retirement, either voluntary or involuntary.

[00:44:35] Thomas Dhomke: So yeah, you have a stable state and you can be a specialist. Maybe you can even ask your boss for a higher salary because the boss will have a hard time finding college grads or university hires that actually want to work on COBOL.

[00:45:14] Thomas Dhomke: But that stable state ultimately is, over the long run, not stable. It's just a gradual decline until it reaches zero or a marginal value. As such, the biggest skill as a software developer and an engineer is that you know how to learn new stuff and how to put this new information in relation to what you're currently working on, because not everything is actually as exciting as it looks.

[00:45:42] Thomas Dhomke: There's also the danger of jumping on the new train, and then you realise, "Oh my God, this is actually throwing us back two years." Like when Apple released the Swift programming language for the very first time, a lot of developers were excited, and then they realised moving their project from Objective-C to Swift wasn't actually pleasant and it was slower.

[00:46:05] Thomas Dhomke: The downloads all of a sudden were blown up by a hundred megabytes or whatnot. There are many such examples where just jumping on the new technology isn't the right decision either. Having that skill to learn and then to judge, to apply your craft and your craftsmanship that you have learned over the last decade or so, and ultimately taste to say, "Okay, we are going this way and not that way."

[00:46:26] Thomas Dhomke: I think this is crucial, even more so now with agents: it's really what you could call engineering skills, like taking a huge problem. Often the huge problem isn't as huge as building Facebook, but it's huge enough that you don't know how long it'll take you to actually implement this, and then breaking it down into smaller pieces.

[00:46:53] Thomas Dhomke: All the way down to: "Okay, now I know what method and unit tests I have to write to actually implement the next abstraction there." I think that engineering skill of decomposing a big problem into small problems, then deciding how to split the work in your own day and with your other team members, is even more crucial with AI.

[00:47:14] Thomas Dhomke: Only if you know what you can offload to an agent, and the agent can actually implement efficiently, and what you have to do yourself, then you gain productivity from AI. If you just try and error everything, "Agent, try this. Okay, bad. Try this again, bad"

[00:47:32] Thomas Dhomke: At some point, you could just go into the CSS file and change the background color yourself, right? There are some one-line changes where it's idiotic to use AI if you know exactly where to do that. You might use AI to just see how the agent fails with that. But the reality is the engineering skillset will be to know when I can offload my work to an agent or work with a coworker, and when I have to do it myself.

[00:47:58] Guy Podjarny: I absolutely agree with that today. But what I would challenge is: if you're a kid coming out of university right now, and you don't know that is the right domain to really double down on investing and getting good at that, knowing how to get down and know which lines in a decision to edit versus spending the opportunity cost of that time learning how to convey and sharpen what it is that you want to build, or maybe even building product and user sensibility appreciation.

[00:48:25] Guy Podjarny: So, leveling up either towards the faster progression, I guess for most people, towards the architect skills, or into the more product definitions or maybe some sort of new nascent profession. It feels hard for me to justify the time invested even if today you do benefit from having that type of depth of skills.

[00:49:08] Thomas Dhomke: Well, is it that the kids, whether they go to high school or to college, actually learn how to use AI much faster than the senior or staff developer that has already been doing that for a decade or two?

[00:49:15] Thomas Dhomke: If I look at my kids in school, they have to submit essays. Funny enough, you can actually read the essays of all the other kids as well, and you clearly can see who used AI to write it, which I actually think is cool. Kudos to those kids that are like, "Maybe it was shortly before the deadline."

[00:49:39] Thomas Dhomke: And then the teacher probably can also see that this was written with AI the long m-dash and whatnot.

[00:49:47] Guy Podjarny: But they don't have the engineering skills. They will embrace the AI faster, but they don't have the engineering skills. Should they lean into the engineering version?

[00:49:53] Thomas Dhomke: Next time, they're going to review the output and prompt the AI to build a version where the teacher cannot detect that it came from AI, right?

[00:50:09] Thomas Dhomke: In many ways, we should actually start very early, like first grade, to teach our kids how to use AI, not to cheat in a test, but to actually iterate and modify the results, and verify that when it talks about whatever the Thirty Years' War in Germany, what it actually says is the right answer. I think teaching kids that skill of how to leverage AI to the best possible outcome is what we are going to see in a new generation of AI-native humans.

[00:50:32] Thomas Dhomke: If you look at Gen Alpha, they have grown up with smartphones. For them, using WhatsApp, Signal, and chatting with their friends is totally natural.

[00:50:51] Thomas Dhomke: While you and I, when we grew up, we had to make a landline phone call and then you had the parents on the phone and you had to pass that first barrier of explaining to them why it's now a good time for you to talk with your buddy, or even worse, with your future girlfriend.

[00:51:07] Thomas Dhomke: You see how the next generation is adopting these new technologies. I think they have ultimately a better life. There are challenges as well, but it's liberating and democratising. I think the same will happen with AI. Everybody will be able to learn coding, even if their parents have no technical background, because they can just leverage the Copilot or ChatGPT that's always there for them.

[00:51:34] Thomas Dhomke: And you know what? It also doesn't judge them and doesn't say "you're stupid" or "you should have learned that three years ago." That's the other thing that's really liberating: you can ask your AI assistant any "dumb" question, and it'll keep answering until you run out of patience. It'll never run out of patience.

[00:51:53] Guy Podjarny: I would still want to stress a bit about the engineering skills. In all of that narrative, which I fully buy into, they didn't necessarily need to know the underlying tech. When you get an image from AI, you don't need to know how to draw to be able to guide it correctly. You need to know what you want better.

[00:52:11] Guy Podjarny: With engineering, I feel like we get drawn into this "but you need to know how to engineer." Today, that's true because they're limited, but in a decade, is that still true? I guess my biases are more about "know what you want," which is actually pretty darn hard.

[00:52:32] Guy Podjarny: It's not an easy thing to answer, and then know how to convey it and know how to verify that it's been done correctly. And to an extent, let go of some of your coding chops to critique the AI's work, or to not invest in improving those.

[00:52:47] Thomas Dhomke: I think we're already in that state. You mentioned the image example where you can easily detect what is AI slop. At some point, you're just not going to go to the Substack anymore or listen to a presentation if all the images are poorly generated. That basically forces the creator to get better at doing that.

[00:53:12] Thomas Dhomke: In a way, that's the new craft: learning how to use AI to get the better output. For images 10 years ago, the craft was to use Photoshop or use pencil and pen. Of course, you could have made your same argument 20 years ago when Photoshop first came out, that it replaces the artist that draws with pen and paper.

[00:53:35] Thomas Dhomke: Now there are experts in Photoshop that are so much better than you and I in Photoshop. So there's still a craft that exists, and we are going to see the same with AI. There are those that are creating amazing images with the help of AI that we cannot create without learning how to do that.

[00:53:55] Thomas Dhomke: If you take that back to software: the reality is commercial software projects and their businesses ultimately have to generate a profit in the long run. Ideally, you have healthy margins or growing margins to sustain that business.

[00:54:15] Thomas Dhomke: That's not going to work if you don't understand what's happening behind the scenes. That's not going to work if you don't have engineering and business skills to work with these agents to create a product that returns enough money to pay the salaries and a profit for the company.

[00:54:32] Thomas Dhomke: Then the question really becomes: who are the companies that are fast enough to learn these engineering and business skills, and who are the ones who think they can get away with just white-coding everything and not paying attention to this? Those companies ultimately will run out of money or will be lacking competitiveness so much that they will not survive in the market.

[00:54:55] Guy Podjarny: I love that. You have to know the insights to innovate in it. We can go on this for an hour, but I think we're running out of time. Super excited to see what you build. Thanks for all the insights here. If someone wants to follow your journey, as you share more about the next evolutions of your company, how can they follow you?

[00:55:19] Thomas Dhomke: They can follow me on X; I'm @ashtom there. I'm also ashtom or just my full name on LinkedIn. In the last few weeks, I've been a bit quiet as we're working in stealth on the new startup, but certainly more to come throughout 2026.

[00:55:34] Guy Podjarny: Cool, well, really excited to see what comes out. Thanks again for sharing this, and thanks for everyone for tuning in. I hope you join us for the next one.

[00:55:41] Thomas Dhomke: Thank you so much, Guy.