Becoming AI Native Without Losing Human Judgment With Fiverr

[00:00:04] What changes when AI stops completing a line of code and starts carrying out much of the engineering workflow in an organisation? This is one of the many questions at the centre of today's episode. Because today I'm very fortunate to be joined by the VP of R&D at Fiverr. My guest has progressed from back-end engineer to leading the company's AI programme during the last 10 years within the business.

[00:00:34] And today he will explain how Fiverr has moved from ChatGPT and GitHub Copilot to context-aware agents. And why the company reports development work becoming three to five times faster. And where human judgement must also remain in control. I will also discuss hiring interviews in which candidates build the AI, the risk of junior engineers producing work they cannot properly explain,

[00:01:03] and why education matters equally as much, if not more so, than the tooling itself. What I'm trying to say is, if your company wants to become genuinely AI native, I think today's conversation will offer a very realistic and practical look at the people, the processes and the accountability required. Yeah, there's a lot of gold in this one. So enough from me. Let me introduce you to him now. So thank you for joining me on the podcast today.

[00:01:33] Can you tell everyone listening a little about who you are and what you do? Yeah, and thank you for having me. It's a pleasure to be here. My name is Oren. I'm a VP R&D at Fiverr. Just passed a great milestone, being 10 years at the company, which is great. I started my career as a software engineer at a couple of early stage startups. I liked the fast-growing atmosphere, and I learned a lot during those few years.

[00:02:00] And I always looked for something bigger, something to put an impact. And then I found Fiverr. I joined as a software engineer to the infrastructure team, which basically is responsible for everything relates to traffic and supporting our millions of customers. I grew. I became a manager and then a director. But I always find the challenges within this infrastructure team.

[00:02:28] A few years after that, I switched to the product side, leading the technology behind our freelancers' products, anything that's from catalog management, monetization, and everything our freelancer needs to success with their business. Today, basically a year ago, I started to co-lead Fiverr R&D as VP.

[00:02:55] And today, I lead the Fiverr AI transformation team, which basically changed how we walk completely using AI and those tools, which always was a fascinating aspect of how to be more effective overall. And this is where I am today. Wow, you must have seen so many big changes. I mean, in the last three years, we've seen AI transform everything. Five years ago, everyone was working from home,

[00:03:24] trying to work out different ways that maybe they could create a side hustle, freelance, et cetera. But having spent nearly a decade at Fiverr, moving from back-end engineering into leading R&D, I've got to ask, what have you seen change inside the company as AI has moved from just an interesting tool to infrastructure that transforms the way that everybody works? Yeah, yeah. Those few years were amazing and a rollercoaster.

[00:03:50] And Fiverr always was an early adopter of new technology. And this is something I really like in the company, whether it's new databases, new frameworks, everything new and also with AI. We started, I think, right after CERGPT came out and all of us used it as a day-to-day conversational genius. Then we adopted what's called GitHub Copilot.

[00:04:18] And what it did, it gave the developers a faster way to complete their code. So imagine an engineer writing his functions, then this new tool comes in and auto-completes his code. So this gave us a great, you know, it was like a magic back in the days, what we can do with AI. But it didn't really change how we work. You know, it didn't really save us time. And then I think a year and a half ago,

[00:04:48] during one of our bigger projects that we did, we adopted a new tool named Cursor, which was a game changer. Because Cursor, what it did, it's kind of like CERGPT, but for engineers, right? You have these English-based conversations, and you ask this agent or AI to write the code for you. And it knows your code base. And this was a big change in how we work back then.

[00:05:17] We saw great value in it. We adopted it across the organization. We had people joining in and, you know, building their own surrounding tools to help it be better. But it was adding AI on top of how we used to work, right? So it didn't change much in the sense of our workflow. It saved time, but not enough, which made us think, what is the next step for being AI native, right?

[00:05:46] And this is where we started the team that I talked to you about earlier. And I suspect we'll have many people listening that are on a very similar journey and much earlier in the journey than you are at Fiverr, especially when they want to go from an AI-assisted company to genuinely be AI native. That is the big goal. But just so we don't lose anyone here, what would you say AI native actually means operationally?

[00:06:13] And what have you done at Fiverr to change your workflows, engineering practices, and organizational structure to get to being AI native? Because I think very often many companies have been guilty of just tagging AI onto existing processes, where very often it's about reinventing the whole mindset and creating those workflows from scratch, right? Yeah, yeah, exactly. And I mean, AI native, it means that you completely change how you work

[00:06:41] and most of the work is being done by agents, by AI. And then having humans as us employees take decisions and judgment. I think that the journey begins to having this transformation. It begins with the understanding that you need to put the focus. You need to put the resources and whatever you believe in to make this change. We did it about a few months ago.

[00:07:06] We allocated a few engineers, designs, and product managers to one team. We called them the octopuses. And all those great team members, we created an agentic ecosystem. We named it Allen based on Allen Turing, obviously. And this platform basically gives all of the employees a set of agents with Fiverr context baked in.

[00:07:36] Imagine these agents have access to the code base, to our data, to our org chart, anything that makes this agent a Fiverr employee with all of that knowledge. And it starts with that. And I believe that while we released it to the company, I saw two main skills for the employees themselves that needs to be adopted in order to make this shift. And the first one, I can say, it's the ability to change.

[00:08:06] We all have those habits, old habits of how to work. But with AI, you need to adjust to it and change it. And this is an important phase of this transformation. And the second skill is self-learning. We have so many tools. You know, the AI industry and even what we do internally with Allen. And you need, as an employee, to learn, to take your time.

[00:08:33] And all of the time, make sure that you understand how to work better with those tools. And before you join me today, I was also reading that you've made tools such as Claude and Agentic AI, all part of your development environment. And again, for people listening, there'll be a big focus on return on investment and you can only improve what you measure. So where are you seeing measurable improvements in how teams build and ship?

[00:09:01] And where does human engineering judgment, where does that still remain difficult to automate? What have you seen here? Right. Obviously, improvements, it begins with, you know, the repetitive work that we handle to agent, right? You don't have to do all of those. You don't need to do all of the coding or, if you can say, in design mock-ups. You don't have to be the person who's doing it. You just, we moved it to be the accountability and the judgments that employees need to take. So we see a big advantage in speed about,

[00:09:31] I think it was about three to five X faster than we had before, just because of those aspects. And beyond, you know, beyond the repetitive work, we also made, with Alan, this standardization layer, when most of what we do is, you know, is the default way to work, right? So imagine that everybody has the same expectation of a workflow.

[00:09:58] They know what should be in and whatnot, and this simply, you know, brings us a lot of improvement in how we work. But as you said, and you mentioned that, not everything can be handed to AI. So obviously, judgment is really important, and we see it in the engineering lifecycle, that when you have a code review, right, this process of while you want to deploy your changes to production,

[00:10:28] you need to pass for manual review someone to see your code, because a lot of things can happen. You can have bugs, but you can also have destructive operation. And with AI, it's something you cannot take this risk, right? So this is where judgment is really something we don't replace yet, even though we do have this code review cycle much more effective with different tools. But indeed, these are the things that we need to... This is the majority of what we do, right?

[00:10:57] The accountability of things. And for customers, I think there's an enormous difference between using AI to generate something and producing work that a customer considers valuable, enough to actually pay for. So as a marketplace sitting between buyers and talent, what signals are you seeing about where human expertise is becoming more valuable rather than less? Because it feels like there's a few changes going on here at the moment, but there's a lot of things on LinkedIn

[00:11:26] about the creative industry, et cetera. So I'm curious what you're seeing. Yeah, that's right. So we're seeing businesses are looking for talent and experts that can use AI the right way, right? They need to combine both the decision-taking and the judgment that they have with the AI tools that are out there. And thinking about that, like AI can produce... Like the output can be great for personal use, right? It can be...

[00:11:55] All of us use it in a day-to-day. It's enough. But when you talk about businesses or companies that need to use it, it needs to become more professional. And this is what we see once people are coming to Fiverr. They are looking for those experts either to complete the AI draft that was initiated by them, but they need the accountability and the experts to make sure that it's a professional way of doing things.

[00:12:23] Your agentic AI might not be secure even with real-time data and proper guardrails. But Denodo makes sure your business has every avenue covered. By placing all your data platforms under one AI data layer, your business can reach semantic consistency safely and securely. So get your agents on the same page by visiting denodo.com and you can learn more about

[00:12:52] how to start trusting your agents to make business decisions. But now, back to today's guest. And internally at Fiverr, I was reading that you've also redesigned elements of your hiring process around AI-first skills. So when you recruit engineers today, just to dig a little bit deeper on this, what capabilities are you looking for that maybe you wouldn't have prioritized a few years ago? And how should people listening adapt their own skills accordingly

[00:13:20] if this is what employees are looking for now? Yeah, I really like this topic. You know, I talk with my employees a lot because it changes a lot of how we hire people. And we completely change our hiring process beginning of this year. And it's not about how many years of experience you have or what company necessarily did you work in. It's how you work with AI. So imagine a candidate come to the interview.

[00:13:48] Most of the question that is being asked, like 80% of them are focused for AI. We give them a laptop and we sit near him or next to him and we are giving him a task, giving us a task, an application to build with any AI tool he wants. And we assess how we work, right? We see how we communicate with AI, how familiar he is with the tool that he decided,

[00:14:16] how decisions are being taken. Like simply AI or what are the questions you need to ask to make sure nothing is wrong. And then obviously the validation part. We want to see candidates going over the changes that the AI did and that will know how to answer the difficult question and really understand the underlying changes. And obviously quality. We need to see that

[00:14:46] at the end of the day, the quality is high. And my suggestion, like I tell it also to my employees, obviously, as I mentioned, I want to see candidates come ready to these interviews, right? You have AI tools everywhere. You need to spend the time at home. Maybe it's a side project. Maybe it's something else. But you need to spend more time with those tools and understand how it works and be like fluent. And when someone comes with this experience,

[00:15:15] it's easy to identify and one concern with the increasing number of AI coding tools is that some of the junior engineers can, yes, they can produce more code, but without necessarily developing the underlying understanding and ability to question the output with the critical thinking that experienced engineers have built over the years. And I'm curious, are you seeing that kind of risk and how do leaders preserve learning, craftsmanship and accountability

[00:15:43] and ensure that the senior leaders of tomorrow have the skills that they need in an AI native engineering team? How do you see this playing out? Yeah, definitely. I mean, the risk is there. We don't want to, as I mentioned, in the candidates' hiring flow. We don't want employees to just use AI and not understanding what they are doing. That's a true risk. And we have all of our techniques and education to keep that in place. But I don't,

[00:16:12] I'm not sure it's about the speed, but more often about how people really use it, right? Because the gap between a junior and senior is much smaller today with AI. You can produce, without a lot of experience, you can produce maybe the same speed or the same quality as senior engineers. But for me, the change is that you as a junior, you or any employee, you need to know what you're doing,

[00:16:41] but you also need to know how to ask those questions using AI, right? So what you don't know is what's more frightening to us. And we also try to validate this process by explaining employees that they need to ask the question, ask AI, just don't ask me, just make sure you know how to pronounce yourself well with AI to answer those questions. And yeah, you talked about how we keep this learning

[00:17:11] and make sure it happens. First of all, we have the experts who are doing the workshops. I talked about the Octopuses team earlier, and we are doing a weekly or monthly session with engineers, passing them the new stuff that we worked on, the best practices, and by really hands-on session, they really understand what should be done. In addition to that, I always say

[00:17:39] to my direct employees or anyone, you need to be more involved and to stay hands-on at some percent of your day-to-day work. I always, as an engineer, as a manager, as director, as VP, I find myself having the hands-on time just because I want to understand the gaps, the problems. Once you're in this cycle, you're much more relevant and you're not doing by what you hear.

[00:18:08] And this is a big tip that I think everybody should do. And I would imagine your time at Fiverr gives you somewhat of a fascinating view of the global labor market and the changes and how it is evolving. And I'm curious from what you're seeing here, what is the marketplace telling you about which skills are gaining the most demand right now, which are becoming commoditized and whether AI is creating new opportunities quickly enough to offset some of the work that it is automating?

[00:18:38] What are you seeing here? Any trends? Yeah, we talked about the skills themselves. You know, we look for the combination of those independent people that can take a task with AI and, you know, make it production ready. One example that we see is what we launched a few weeks ago, a few months ago, is about the AI Video Hub. it's how independent

[00:19:07] directors, they can produce with AI, you know, cinema quality brand videos and they can do it faster, they can do it alone without all of the studio behind them, which makes it much more effective for customers who want these specific tools and eventually it comes to what we said earlier, right? how to use AI with the judgment and the creativity of those experts.

[00:19:37] And if we have a CTO or an engineering leader listening to our conversation today anywhere in the world, they want to follow Fiverr towards this AI native working that we're highlighting today, what would you say you've learned most about infrastructure, talent, incentive, governance and the cultural changes, possibly the most important there, that really need to happen before a genetic AI can become part of everyday work rather than just another experiment that gets stuck in part of that phase. Yeah,

[00:20:08] I'll be honest, I mean, since we launched Alan and we worked on this AI native ecosystem, we learned so much that, you know, it's amazing to see how the world is changing. I would begin with putting the focus and the resources and understanding that this transition cannot happen by itself. You cannot develop a feature and all of a sudden everybody will become AI native. So you need to put that focus in mind and when you do find or look for

[00:20:38] the right people, this when you look for someone who knows how to work with AI but most importantly has this mindset of working to be always more effective. We have those employees everywhere in every company. We need to find them and make sure that, you know, they are kind of hackers, that they know how to hack their time and to be much more effective. and what we saw in the last few months

[00:21:07] that adoption, it doesn't come just from tooling as I just said. It needs a lot of education. So we put a lot of effort in education and it needs to be a planned education process because we need to change people like how they worked until now. Their habits are taking a lot of time to change and a lot of effort and this is why education is super important and we invest a lot of time in it. And lastly, just like any other

[00:21:37] feature, product, or anything that we do, we want to measure the impact of it. So also with AI Native, you need to invest a lot of time seeing the numbers and seeing that the solution that you produce really have an impact, right? Because you can do a lot of things but it doesn't change to be aware of that. And yeah, this will be my two cents I can say. Well, I absolutely love chatting with you today about

[00:22:06] why the value in AI is shifting from tools to talent, how to become an AI Native organization and the lessons learned along the way and for anybody listening that would like to find out more information about anything we talked about today, connect with you or your team, keep up to speed with the announcements coming out of Fiverr, etc. Where would you like me to point everyone? Yeah, I mean, Fiverr.com is the place that we can also share all of the stuff that we do and also you can find me on LinkedIn. A lot of great

[00:22:35] content is there and in our blog post, Fiverr blog post. Awesome. And I think we covered today the challenges and equally the opportunities that AI is bringing to the global workforce and some good news there around how AI is impacting talent demand, the most and least in-demand skills as a result of AI and what needs to happen to make AI in the workplace actually work. And I will have links to everything including your LinkedIn profile.

[00:23:05] I urge people listening to carry on this conversation. It's something that impacts every enterprise and every industry. But more than anything, thank you for your time today and sharing your story. Thank you. Thank you very much for having me. I think my guest experience offers a useful warning for every company that are investing in AI tools right now. Yeah, buying the technology, that's the easy part. The harder work involves changing habits, teaching employees how to

[00:23:35] question AI output, measuring whether performance has actually improved, and keeping people accountable for their final result. And Fiverr is reporting development work becoming three to five times faster. But speed means little if nobody understands what has been placed into production. So I loved his advice there for leaders to remain hands-on because I think it's very difficult to guide teams through problems that you only hear about in

[00:24:04] meetings. And I cannot thank him enough for taking the time to talk about Fiverr's AI native journey. I know there are a lot of listeners, a lot of organisations that are on a similar path right now, and I'd love to hear from you, especially as AI takes on greater portions of your work. Are you giving your people enough time and support to understand what the machines produce? Let me know. techtalksnetwork.com You'll find over 4,000 interviews over there from a

[00:24:33] variety of podcasts and also check out the event pages, lots of places you can meet me on the road. But that's it for today, so thank you for listening. Bye for now.