Fiverr CTO Overhauls Code Workflows as AI Shifts Product Roles
Ai4
AI-generated code cannot solve software engineering's last mile problem, making deep technical craft and human expertise essential for production-ready applications.
## Background
This video was published on YouTube by the Ai4 channel on August 29, 2025. According to the page description, it is an episode of 'The Ai4 Lens' featuring a fireside chat with Yossi Levin, Chief Technology Officer at Fiverr, spotlighting the ideas, people, and tensions shaping the AI landscape.
## TL;DR
Yossi Levin shares his journey from early solo coding on a ZX Spectrum to leading engineering teams, emphasizing software development as a standardized craft rather than personal art. He outlines Fiverr's internal AI strategy, such as forming an engineering guild led by their Chief Architect, and discusses how 'vibe coding' enables Product Managers to build rapid prototypes instead of writing traditional specifications. Levin explains the 'last mile problem' in software engineering, where human expertise remains essential for architecture, security, and performance. Ultimately, he stresses that deep professionalism and genuine passion will distinguish successful engineers as routine execution becomes automated.
## Atmosphere & style
The conversation carries a thoughtful, reflective, and pragmatic tone, blending humorous personal anecdotes with grounded advice on software engineering leadership. Britney Con serves as an engaging and warm host, guiding Levin through a nostalgic look back at his childhood coding roots into a sharp analysis of modern AI workflows. The dynamic feels like an intimate, lighthearted yet authoritative discussion among industry veterans, pairing classic pop-culture references with serious reflections on the changing nature of tech talent.
## Key takeaways
- [00:04](https://www.youtube.com/watch?v=6AbvE8oPB4U&t=4s) Yossi Levin began coding at age 13 on a ZX Spectrum + connected to a family television, learning through isolated trial and error.
- [00:04](https://www.youtube.com/watch?v=6AbvE8oPB4U&t=4s) Levin views software engineering as a disciplined, standardized craft rather than a personalized art form.
- [00:04](https://www.youtube.com/watch?v=6AbvE8oPB4U&t=4s) Code readability and maintainability should follow strict standards so that an engineer's individual 'handwriting' is undetectable.
- [00:45](https://www.youtube.com/watch?v=6AbvE8oPB4U&t=45s) To instill fast mechanics and keyboard navigation, Levin would take away mouse usage from junior developers.
- [00:45](https://www.youtube.com/watch?v=6AbvE8oPB4U&t=45s) Engineers must thoroughly understand AI-generated code rather than blindly accepting output to avoid severe technical debt.
- [00:45](https://www.youtube.com/watch?v=6AbvE8oPB4U&t=45s) Fiverr established an AI guild directed by its Chief Architect to standardize AI tool adoption across engineering teams.
- [10:12](https://www.youtube.com/watch?v=6AbvE8oPB4U&t=612s) Vibe coding represents a new baseline across the industry rather than a competitive advantage by itself.
- [10:12](https://www.youtube.com/watch?v=6AbvE8oPB4U&t=612s) Product Managers at Fiverr use vibe coding to rapidly build working prototypes instead of relying solely on written specification documents.
- [12:45](https://www.youtube.com/watch?v=6AbvE8oPB4U&t=765s) Human freelancers remain essential to bridge the 'last mile problem,' handling architecture, reliability, security, and performance.
- [12:45](https://www.youtube.com/watch?v=6AbvE8oPB4U&t=765s) Human talent provides distinct advantages in taste, brand identity, and saving time for business owners through delegate setups.
- [16:55](https://www.youtube.com/watch?v=6AbvE8oPB4U&t=1015s) The era of 'diluting the soup on talent' by hiring mediocre developers during boom cycles has come to an end.
- [16:55](https://www.youtube.com/watch?v=6AbvE8oPB4U&t=1015s) Deep technological curiosity and fundamental mastery of software engineering are the key human differentiators in the AI era.
## Detailed walkthrough
Here is a detailed, faithful walkthrough of the transcript:
[00:04](https://www.youtube.com/watch?v=6AbvE8oPB4U&t=4s) Early Beginnings and Coding as a Craft
Britney Con, Senior Content Producer for Ai4, opens the session by welcoming Yossi Levin, Chief Technology Officer at Fiverr. While Levin has spoken at Ai4 before, Con sets the stage for a deeper discussion beyond code—focusing on leadership, craft, and remaining human in a transforming work environment. She lightheartedly apologizes for "aging" him, bringing up his early start in software development as a child, long before YouTube tutorials, forums, or ChatGPT, and asks how those solitary roots shaped his engineering management philosophy.
Levin recalls starting to code at age 13 before personal computers were widespread. Sitting on the living room floor connected to his family's single home television, he programmed on a ZX Spectrum + with 48K memory. He describes it as a deeply isolated experience as the "only weirdo" among his peers who programmed. Without guidance, he was forced to invent his own methodologies through trial and error, moving away from meaningless single-letter variable names like `i`, `x`, `y`, and `z` toward structured, readable code.
As his career matured into engineering leadership, Levin recognized that software development should not be treated as purely personal art, but rather as a disciplined craft. While art allows individual flair, craft emphasizes standardization so that any developer can read and maintain the codebase seamlessly. He insists that he shouldn't see a developer's individual "handwriting" in the code. When late-night fixes or hacks are necessary, he encourages engineers to leave honest, unvarnished emotion and context in the comments—admitting when a section is temporary "crappy code" to be replaced later rather than trying to pass it off as "God code." He cites a classic Visual Studio 6 meme as a guiding motto: write and maintain your code as if the person who has to maintain it after you is a violent psychopath who knows where you live.
- **Hardware Used:** ZX Spectrum + (48K memory).
- **Core Philosophy:** Software development is a standardized craft (like carpentry), not individualized art.
- **Motto:** Code like the maintainer is a psychopath who knows your address.
[00:45](https://www.youtube.com/watch?v=6AbvE8oPB4U&t=45s) Enforcing Professionalism and Modern AI Workflows
Con highlights Levin's view of coding as an all-encompassing craft, asking why he places such significant emphasis on details down to how someone types, uses an editor, or structures their thoughts. Levin connects this to pair programming, recalling his frustration in university when working with beginners whose editor pacing, copy-pasting habits, and indentation styles felt inefficient. As a team leader, he actively sat beside developers to instill proper mechanics—enforcing tabs over manual spaces and eliminating mouse usage for repetitive text editing or stepping through a debugger. In fact, if an engineer used a mouse to copy and paste, Levin would physically unplug their mouse and take it away for the day to force them to master keyboard shortcuts.
Turning to modern fast-paced engineering environments, Con asks how developers can balance craft with speed and deadline pressures. Levin warns that while AI code generators dump hundreds or thousands of lines of code onto developers, engineers must never blindly accept generated output without understanding every flow. Ignoring the inner workings of generated code creates immense technical debt that will inevitably break during a late-night incident.
Regarding management, Levin compares the AI development revolution to the Industrial Revolution, though with a distinct twist. While the original Industrial Revolution adapted human workers to rigid factory conveyor belts, the AI shift invades the intimate space between an engineer and their laptop screen. He argues that line managers cannot simply throw tools like Cursor or Windsurf at engineers with a "good luck" attitude, as this only creates confusion. To address this, Fiverr established an internal AI guild led by their Chief Architect, placing an expert in every team to refine workflows, tweak backend settings, and train developers on exact tool practices.
- **Management Tactic:** Taking away a developer's mouse for a day to teach fast, keyboard-driven navigation and debugger controls.
- **Fiverr Structure:** Formed an engineering guild directed by the Chief Architect, deploying dedicated AI experts into individual teams to standardize AI tool adoption.
[10:12](https://www.youtube.com/watch?v=6AbvE8oPB4U&t=612s) Vibe Coding, Product Managers, and Human Edge
The discussion shifts to the concept of "vibe coding," which Levin describes as a new baseline rather than a competitive differentiator. While describing what you want and having AI generate functional software instantly is impressive, Levin emphasizes that everyone now has access to the same tools. He compares it to the launch of Java—which eliminated manual memory management—or climbing Mount Everest; once the baseline elevates for everyone, speed alone no longer makes a team unique.
At Fiverr, Levin redefines the utility of vibe coding by giving these tools directly to Product Managers (PMs). Rather than writing endless specifications and stories to describe a feature, PMs use vibe coding to rapidly prototype working applications. Channeling a classic *Seinfeld* moment where Kramer suggests simply telling him what movie you want to see, Levin explains that PMs can now build real, working prototypes to demonstrate their exact vision to engineers. This breaks down communication barriers between product management and engineering, providing developers with a functional artifact they can appreciate and refine technically rather than relying on abstract specification documents or Figma designs alone.
```
+------------------+ +--------------------+
| Product Manager | -- Vibe Coding --> | Working Prototype |
+------------------+ +--------------------+
|
v
+------------------+ +--------------------+
| Final Product | <-- Human Craft -- | Engineer Review |
+------------------+ & "Illities" +--------------------+
```
[12:45](https://www.youtube.com/watch?v=6AbvE8oPB4U&t=765s) The Last Mile, Freelance Ecosystem, and Fiverr's Vision
Addressing questions around AI platforms potentially replacing human freelancers, Levin outlines why human talent remains irreplaceable. He explains that vibe coding gets a project close to the finish line, but hits a wall known as the "last mile problem." This is where human expertise steps in to handle system architecture, reliability, performance, security, and the endless list of software "illities" that turn a cool demo into production-ready software.
Beyond software development, Levin highlights three core areas where human freelancers excel:
1. **Iterative Frustration:** Avoiding endless battles with prompt engineering across LLMs like Gemini or ChatGPT when trying to achieve specific outputs (e.g., custom image generation).
2. **Time Arbitrage:** Enabling business owners to delegate technical setups—such as wiring up a payment flow or building an app over a few days or hours—so they can focus on client acquisition, market strategy, and business growth.
3. **Taste and Style:** Providing unique human design taste, brand identity, and customized user flows that stand out from repetitive, templatized AI-generated interfaces.
[16:55](https://www.youtube.com/watch?v=6AbvE8oPB4U&t=1015s) Preserving Deep Professionalism in the AI Era
In closing, Con asks Levin to share a core principle for the next generation of engineers to hold onto as AI becomes more dominant. Levin emphasizes that deep professionalism and genuine curiosity matter now more than ever.
He references a concept from a previous company called "diluting the soup on talent"—the practice of hiring mediocre developers just to get extra working hands during hype cycles and boom periods. Levin asserts that this era of diluting talent is over. Many people entered software development simply because it paid well straight out of college, without a true passion for or deep understanding of the underlying mechanics. As AI takes over routine task execution, the primary differentiator for human engineers will be deep, fundamental knowledge of technology and a passion for mastering the craft.
## Notable quotes
> [00:04](https://www.youtube.com/watch?v=6AbvE8oPB4U&t=4s) write and maintain your code as if the person who has to maintain it after you is a violent psychopath who knows where you live — Yossi Levin
## Video description
Yossi Levin, Chief Technology Officer, Fiverr
Abstract
The Ai4 Lens is our signature editorial track — a curated series of fireside chats and thought-provoking conversations designed and led by the Ai4 content team. These sessions spotlight the ideas, people, and tensions shaping the AI landscape in bold ways.
Society Tracks
AI重塑产品角色，Fiverr CTO重构代码工作流
AI生成的代码无法解决软件工程的最后一公里问题，深厚的技术造诣与人类专业知识对生产级应用依然不可或缺。
Yossi Levin 分享了他从早期在 ZX Spectrum 上独自编程到带领工程团队的历程，强调软件开发应是一门标准化的技艺，而非个人艺术。他概述了 Fiverr 内部的 AI 战略，例如组建由首席架构师领导的工程公会，并探讨了“氛围编码”（vibe coding）如何让产品经理能够快速构建原型，而非撰写传统的规格说明书。Levin 还解释了软件工程中的“最后一公里问题”，指出人类专业知识在架构、安全性和性能方面依然至关重要。最后他强调，随着常规执行任务走向自动化，深厚的专业素养和真正的热情将成为优秀工程师脱颖而出的关键。
Fiverr 首席技术官 Yossi Levin 强调软件工程是一门标准化的工艺而非个性化艺术，代码应具备严格的规范性与可维护性，甚至不应显露出任何开发者的个人“笔迹”。 (0:04)
Levin 警告工程师切勿盲目接纳 AI 生成的代码以免累积严重技术债务；Fiverr 已设立由首席架构师牵头的 AI 行会（AI guild），在各团队中标准化推广 AI 工具并优化流程。 (0:45)
Levin 认为“氛围编程”（Vibe coding）已成为行业普遍基准而非竞争壁垒；在 Fiverr，产品经理正借助氛围编程直接快速制作可运行的原型，以替代传统的书面需求文档。 (10:12)
面对 AI 无法自主解决的系统架构、可靠性与安全性等“最后一公里”挑战，Levin 强调随行业热潮招揽平庸人才的时代已经终结，深厚的技术底层功底才是工程师的核心壁垒。 (12:45)
Reading summary
Fiverr Chief Technology Officer Yossi Levin argues that as generative AI establishes a higher baseline for software creation, engineering must be treated as a standardized, disciplined craft rather than an individualized art. Levin contends that human engineering advantage now centers on system architecture, operational reliability, and solving the complex 'last mile' of production.
Software Engineering as a Standardized Craft (0:04)
Drawing from his early experiences learning to program on a 48K ZX Spectrum +, Levin maintains that software development must function as a standardized discipline akin to carpentry rather than personal art. Codebases should remain uniform without individual 'handwriting' so that any maintainer can seamlessly review and update them. When developers deploy temporary hacks, Levin advocates for honest commenting acknowledging the flawed code rather than disguising it, referencing the adage to write code as though the subsequent maintainer is violent and knows your address.
To build disciplined mechanics, Levin prioritizes keyboard-driven efficiency over mouse reliance. During his time as a team lead, he actively enforced standard indentation and banned mouse usage for text manipulation and debugging, going as far as temporarily confiscating an engineer's mouse to compel mastery of keyboard shortcuts.
Managing AI Adoption and Guarding Against Debt (0:45)
Levin warns that automated code assistants frequently generate hundreds or thousands of lines that developers may accept without fully understanding the underlying logic. Overlooking the inner workings of generated code introduces severe technical debt that risks breaking during late-night production incidents.
Contrasting the AI transition with the Industrial Revolution—which adjusted laborers to external factory conveyor belts—Levin observes that AI alters the intimate space between a developer and their personal screen. To prevent confusion from haphazard rollouts of tools like Cursor or Windsurf, Fiverr established an internal AI guild directed by its Chief Architect. The organization deploys dedicated AI specialists directly into development teams to calibrate configurations and standardize tool adoption.
Repurposing Vibe Coding for Product Teams (10:12)
Levin characterizes 'vibe coding'—generating operational software through natural language prompts—as an elevated industry baseline rather than a lasting competitive edge. Much like the introduction of Java automated memory management, widespread access to generative tools raises baseline capabilities, meaning that raw generation speed alone no longer distinguishes an organization.
Rather than restricting vibe coding to software engineers, Fiverr equips Product Managers with these tools to construct functional prototypes. By replacing lengthy narrative specifications and static Figma mockups with functioning application demos, product managers can explicitly communicate their feature vision to engineers, streamlining collaboration and technical handoffs.
The Last Mile and Human Value (12:45)
While vibe coding accelerates initial prototyping, Levin emphasizes that automated generation inevitably confronts the 'last mile problem.' Transitioning a functional prototype into production requires human engineering to manage system architecture, scalability, security, performance, and general operational reliability.
Beyond technical architecture, Levin identifies three areas where human professionals remain superior to automated outputs: sparing clients the iterative frustration of complex prompt engineering, providing time arbitrage by handling operational setup such as payment workflows, and applying distinctive human taste and bespoke visual identity that generic AI templates lack.
Ending Talent Dilution and Emphasizing Fundamentals (16:55)
Levin asserts that tech industry practices around 'diluting the soup on talent'—hiring mediocre developers simply to expand headcounts during market boom cycles—have come to an end. He observes that past compensation spikes drew individuals into development who lacked intrinsic passion or mechanical comprehension of computer systems.
Because AI now handles routine syntax generation and repetitive task execution, Levin argues that durable engineering careers will require authentic curiosity, rigorous professionalism, and a comprehensive understanding of underlying technical mechanics.
While Levin details how Fiverr embeds AI guild experts and equips product managers with rapid prototyping tools, the discussion leaves open how organizations should formally measure engineering output and accountability when code volume is decoupled from human effort.
[0:04] Yossi Levin: write and maintain your code as if the person who has to maintain it after you is a violent psychopath who knows where you live
[0:04] Yossi Levin: crappy code
[0:04] Yossi Levin: God code
[0:45] Yossi Levin: good luck
[10:12] Yossi Levin: vibe coding
[12:45] Yossi Levin: last mile problem
[16:55] Yossi Levin: diluting the soup on talent
Original transcript
[0:04] Hi everyone. Thanks so much for joining
[0:06] us for the AI4 Lens. I'm Britney Con,
[0:09] senior content producer for AI4. And I'm
[0:12] so excited to be joined today by Yosi
[0:15] Levin, CTO of Fiverr, where he leads
[0:18] teams in product innovation and AI
[0:21] strategy. Now, I know that Yosi's
[0:23] already been featured a couple times
[0:24] here at AI4, and I really hope you've
[0:26] had the opportunity to hear him speak.
[0:29] Um, but in this conversation, I want to
[0:31] go just a level deeper, if that's okay
[0:33] with you. Not just into the code, but
[0:35] into the craft of it all, into what it
[0:38] means to lead, to build, and to stay
[0:40] human as we redefine how we work. So,
[0:43] Yosi, let's start from the very
[0:45] beginning. You started building software
[0:47] as a kid before YouTube tutorials and
[0:50] forums and chatbt, and I'm really sorry
[0:52] to age you right out the gate. Um but do
[0:56] you think that early sense of figuring
[0:58] it out all on your own shaped how you
[1:01] manage engineers today?
[1:03] >> It definitely did. Um I started at the
[1:06] age of 13. Uh were before the age of
[1:10] PCs. So I was using a Spectrum Sinclair
[1:14] 48 ZX Plus with with 48K memory. And now
[1:20] you can guess my age. and sitting on my
[1:24] in the living room on the floor
[1:26] connected to the TV, the only TV at home
[1:29] and fighting for
[1:32] computer time. And
[1:35] it was seriously a lonely experience.
[1:37] Like I have no I had no friends that
[1:39] were programming. I was the only weirdo.
[1:42] And
[1:44] like you had to figure it out. And as
[1:46] you figure it out, you
[1:50] you kind of create your methods. Nobody
[1:53] told me that variables have to have
[1:56] meaning. And so I started with I and X
[1:59] and Y and Z. And then no, no, no. like I
[2:02] don't understand what I'm so you created
[2:04] variable names and you structure your
[2:06] code and and you start to understand
[2:09] things
[2:11] uh and that will make things simple and
[2:15] as I
[2:18] matured into engineering and leading
[2:19] engineering teams,
[2:22] I I found that the need
[2:26] to have the engineers not look at it as
[2:28] art cuz cuz as a young ling I was
[2:31] looking at this is art. I'm creating.
[2:34] But look at it as craft. You're building
[2:37] something.
[2:39] And when it's craft, then
[2:43] you take away in a sense the artistic
[2:46] element and you you bring in a
[2:49] standardization element. So, as I like
[2:52] to say, I don't want to see the
[2:54] handwriting of developers in the code
[2:57] cuz everything has to be the same cuz
[2:59] somebody else needs to read this code
[3:01] and maintain the code and does not need
[3:03] to understand what you were thinking.
[3:06] And if you build your code and you
[3:08] hacked something, cuz we do, and I call
[3:12] it,
[3:14] put your your emotions, not really
[3:18] emotions, but what you went through in
[3:20] the in the comments.
[3:22] Say like, I don't want anyone to think
[3:24] it was God code. Um, like if it was a
[3:27] 2:00 a.m. in the morning, a 2:00 a.m.
[3:30] hack that you had to do because you had
[3:31] to ship tomorrow and you had, you know,
[3:34] it's a crappy piece of code, say, "This
[3:37] is a crappy piece of code." In the
[3:38] comments, replace it one day. I'm not
[3:41] proud of this. This is what it's
[3:42] supposed to do. If you have the design,
[3:44] write the design so somebody coming
[3:46] tomorrow looks at it does not say, "It's
[3:48] amazing code." No, it's amazing code. I
[3:51] cannot read it. No, put it in because
[3:54] you need to hand it over. There was a I
[3:57] never know if it's a meme or it's real.
[4:00] Visual Studio 6 had the tip of the day
[4:02] thing and there is a meme saying that
[4:06] build your code as if the person going
[4:08] to maintain it is a psycho psychopath
[4:09] killer who knows where you live. And
[4:12] that's a very good moto to go by. And
[4:17] so the craft is super important in in
[4:20] this like you you like a carpenter like
[4:23] you do it exactly and you do it right
[4:25] and every chair should look the same
[4:28] like as the next.
[4:30] >> I actually really love this philosophy
[4:32] that you have that coding is a craft and
[4:34] I think it's sort of even bigger than
[4:36] just the output but you really look at
[4:38] it as like how someone types or how they
[4:40] use their editor or even how they think.
[4:42] So why do you think that level of detail
[4:44] is so?
[4:45] >> So
[4:46] I take it to pair programming. Um as
[4:51] when I was in university,
[4:53] I found it very hard to pair program
[4:56] with people who were just beginning cuz
[4:58] the pace was wrong and the way that they
[5:00] edited was wrong to me. So like I
[5:03] couldn't stand it. And and to find
[5:05] someone that's compatible with your
[5:08] programming is something very hard.
[5:11] someone who like was fast enough who
[5:14] copy paste in the right way who like
[5:16] ident the code in the right way and and
[5:19] then as a manager I was as team leader I
[5:22] was sitting with the developers
[5:25] and forcing them like looking at how
[5:27] they code and teaching them the craft no
[5:30] you don't do that no that's the wrong
[5:32] way to ident. No, you don't go in and do
[5:35] like 17 spaces. You use tabs. And yes,
[5:38] tabs should be translated to spaces like
[5:40] all the little things. And to the point
[5:43] that if someone was copying like using
[5:47] the mouse as they were programming and
[5:49] like to copy paste which is the slowest
[5:52] possible way to copy paste like if
[5:54] somebody does that I used to take away
[5:56] the mouse like I was come and it was not
[6:00] pieces it was pieces so it was on the
[6:02] floor I I take the mouse away no mouse
[6:05] for you for the day come back tomorrow
[6:08] and and it was amazing because people
[6:11] got faster
[6:12] cuz they learn how to figure it out.
[6:14] Like if you need with a debugger to
[6:15] click on next step, next step, next step
[6:18] with the mouse, you are doing something
[6:20] wrong. You are not a professional. And
[6:23] like I you as a team leader, you have to
[6:26] teach people to be professionals.
[6:29] >> That's a very interesting management
[6:30] style to steal somebody's mouth from
[6:32] them.
[6:32] >> Until today, um people complain
[6:36] >> they could take it up with HR. But in
[6:38] today's engineering culture, everything
[6:40] is so fast-paced. People want stuff done
[6:42] immediately. So, what is sort of your
[6:45] like top tips for engineers to sort of
[6:48] balance the craft with getting things
[6:50] done under speed and pressure?
[6:53] >> So,
[6:57] first of all, know what you're working
[6:59] on
[7:02] completely understand the flow. And
[7:03] today we use AI and and AI generates
[7:06] it's a code generating machine. It's
[7:08] like dumping tens, hundreds, thousands
[7:11] of of lines of codes on you. And
[7:16] do not not understand it. Do not say,
[7:20] "Yeah, it's okay. It's actually okay. I
[7:21] hope it's fine." And and carry on.
[7:24] Really know what's going on. Cuz one
[7:26] day, 2:00 a.m. in the morning, something
[7:27] is going to break. You're going to be
[7:29] waken up. and well I need to read this
[7:32] code for the first time in my life and I
[7:33] don't know what's going on. It's not you
[7:36] think it's going to make you slow but
[7:37] it's not going to make you slow. So know
[7:40] what's what's going on and and for the
[7:42] managers
[7:44] and this is something super interesting
[7:46] in what's going on.
[7:48] AI for development is a it's like the
[7:52] industrial revolution
[7:54] but whereas the the first industrial
[7:56] revolution we changed
[7:59] we changed the way that things work
[8:01] humans had to adapt. So suddenly you
[8:04] have a conveyor belt. You stand in this
[8:06] place and you have to do this
[8:08] and you do this the entire day. So if
[8:11] you don't do this something breaks down
[8:12] the line you're fired. We bring someone
[8:14] who can do this but you we we complete
[8:17] you made and with a sense of we'll one
[8:19] day replace you with a robot when things
[8:21] will evolve and this
[8:25] um industrial revolution goes into a
[8:28] very private place. It goes into the
[8:30] space between the engineer and the
[8:33] laptop.
[8:35] And it's very hard to get into this
[8:36] space cuz that's like that's how I do
[8:39] things.
[8:40] And the managers need to find their way
[8:44] to get into this space. And it's like
[8:46] line managers. It's the team leaders
[8:49] like dumping here is cursor. Good luck.
[8:51] Here is another tool. Good luck. Wind
[8:53] surf. Yay. Go. No. because you're just
[8:57] throwing more and more and more tools at
[8:59] people that will not know what to do
[9:00] with these tools. You will not make them
[9:02] stronger or faster or better. You're
[9:04] just going to throw tools at them and
[9:07] you go into this and let's see together
[9:09] and let's see how you use the tool. And
[9:11] we in Fiverr we set up a guild and we
[9:13] have an expert in every team that is
[9:17] part of the guild. Our chief architect
[9:18] is leading this and we are working with
[9:22] people on the practices on so it's not
[9:25] understanding this. Let's change let's
[9:26] train it with using cursor at this at
[9:28] this point. Let's like tweak it at the
[9:32] back end to see that it is actually
[9:33] doing exactly what we need for this
[9:35] team, exactly what we need for this
[9:37] person cuz just throwing the tools and
[9:40] it's like super easy. Okay, here is some
[9:44] money. Get tools. It's the worst thing
[9:46] that you can do because you literally
[9:48] confuse the entire organization.
[9:51] And
[9:54] that's I think the the the future is
[9:57] going back to the line manager
[9:59] to retrain the people with the new
[10:02] tools. There's like a whole training of
[10:04] the industry how to reuse the tools, how
[10:08] to
[10:09] we're reinventing the craft basically,
[10:12] >> right? And I think that that gets us
[10:13] into sort of like the buzziest place of
[10:15] engineering right now and that's vibe
[10:17] coding. And yes, vibe coding. And you've
[10:20] said that vibe coding is really the new
[10:21] baseline for engineering teams. It's not
[10:23] really a differentiator anymore. So if
[10:25] that's the case, what does give a team
[10:28] an edge today? And why do engineers
[10:30] still matter more than ever?
[10:32] >> So
[10:34] vibe coding is awesome. Like uh describe
[10:37] what you wanta poof and appears
[10:40] [Music]
[10:42] doing mostly what you intended to it to
[10:44] do. uh with like three lines under
[10:47] mostly um
[10:50] the but then people say we use vibe
[10:53] coding we are awesome we are so unique
[10:55] and we can run faster no you can run
[10:57] exactly at the same pace that everybody
[10:59] can run cuz everybody got this level up
[11:02] you're not special in anything if you
[11:04] remember when Java came out suddenly you
[11:06] don't have to manage memory we are all
[11:08] so awesome nobody needs to do it we can
[11:10] write code faster of course everybody
[11:12] can write code faster imagine you climb
[11:14] a mountain, you climb Mount Everest, you
[11:16] go to Mount Everest,
[11:18] everybody's there cuz you're not special
[11:20] anymore. So, vibe coding is another tool
[11:23] in the arsenal. But I what we do in
[11:26] Fiverr with with vibe coding, I think
[11:30] it's important to use this tool is we
[11:33] give product managers vibe coding
[11:36] instead of trying to describe to the
[11:39] developers what you want and write
[11:40] endless stories and and hope that
[11:43] somebody understand and spec yourself to
[11:45] death. Um just
[11:49] was in Seinfeld just tell why don't you
[11:51] tell me what movie you want to see like
[11:53] one of the episode Kramer said that and
[11:56] so just build it come to me show it how
[12:00] it show me how it works maybe I'll even
[12:02] take some of the code but show me how it
[12:04] works and and show me a real application
[12:07] now you can you can argue that Figma can
[12:09] do the same but it cannot really do the
[12:12] same and now they're doing actually cool
[12:14] stuff in Figma
[12:16] The the idea is to to kind of break the
[12:20] barrier between engineers and and
[12:23] product managers.
[12:25] That's how I like to use it. Like do we
[12:27] break the barriers? You feel a little
[12:29] bit more what you need to explain and
[12:32] the engineer gets something that he can
[12:35] technologically appreciate. Not
[12:37] necessarily just copy paste, but
[12:39] appreciate and understand.
[12:41] So there is still real value even using
[12:43] tools like vibe coding that engineers be
[12:45] on the team. And so in that vein I think
[12:48] there's also been a lot of talk about
[12:49] platforms sort of replacing freelancers
[12:52] with AI and naturally Fiverr's name does
[12:56] come up. This is where we get into the
[12:57] hard-hitting journalism. So what do you
[13:00] want people to understand about Fiverr's
[13:02] vision for human talent in an AI powered
[13:04] world? So I'll I'll I'll go back to vibe
[13:09] coding and then then go to this. So viro
[13:12] coding is awesome. It will get you
[13:14] almost to the finish line and it will
[13:16] get and then you have the last mile
[13:18] problem and that's where expertise
[13:20] comes. That's where the human comes and
[13:23] takes care of all the illities. Actually
[13:25] looked it up in in in
[13:28] Wikipedia which is
[13:30] I don't trust it anymore but whatever.
[13:33] um
[13:34] like a list like you scroll through an
[13:36] endless list of illities that you didn't
[13:39] even know that you need to care about
[13:41] after all these years. And the thing
[13:44] that will make your application take
[13:45] your your software from a cool idea and
[13:49] a nice demo that you can show your
[13:52] mother or your boss
[13:54] to something that is actually production
[13:56] worthy. Something that you can put out
[13:59] there. something that will not break
[14:01] under the slightest of pressure or
[14:04] something that will not
[14:07] somebody said that CV like vibe coding
[14:09] application or CV generators
[14:12] um like something that will actually
[14:15] will not shame your organization in the
[14:17] end that's where human talent comes and
[14:20] AI
[14:22] that so so that's one thing and it's on
[14:24] software but it's only other things and
[14:27] image generation
[14:30] How long are you going to fight with AI
[14:32] to get it to do exactly what you want?
[14:34] Like, okay, let's spend an hour. I spent
[14:36] like one Saturday
[14:38] arguing with with Gemini and then I
[14:41] moved to JGPT and then like and and
[14:45] I I like I really wanted to give it a
[14:47] step just do what I want and so that's
[14:50] where human talent comes. And the
[14:54] maybe the last angle on this is well two
[14:57] angles.
[14:59] One is time arbitrage.
[15:01] Like you sit now you start a business.
[15:04] You decide to build your application.
[15:06] Okay. Is this the best use of your time?
[15:09] Now you start a business and then you
[15:11] sit and you build an application. When
[15:13] do you decide that you want to hand it
[15:14] over to someone else so you can deal
[15:16] with getting clients, understanding the
[15:18] business, understanding the market,
[15:19] doing something that is more than
[15:21] coding? I know it's cool and fun, but
[15:24] can I get someone to do it? I'm not in
[15:27] the position to get someone full-time
[15:30] and start nurturing an engineer, but
[15:35] I am in a position to hire someone for 5
[15:37] hours to get a job done. Hire someone
[15:39] for 3 days, get me this app, wire it up,
[15:43] get this payment flow going. Thank you
[15:45] so much. Bye-bye. I'll be in touch with
[15:46] you if I need you again. this short-term
[15:50] engagements that can actually get
[15:53] get you to use your time right is where
[15:55] Fiverr comes in. That's one thing, the
[15:58] time arbitrage. The other thing and
[16:04] is
[16:06] style.
[16:08] You go to AI, it generates images from
[16:11] other images. It's very hard to be
[16:13] unique.
[16:15] You need you have your site. You go on
[16:18] a vibe coding application to create a
[16:20] site. At the end of the day, it's all
[16:23] pretty much templatized. It's very hard
[16:24] to create a different experience, a
[16:27] different flow. You'll get something
[16:29] that's one of a set of templates. Having
[16:33] some someone with style that understands
[16:35] what you want, that understands the
[16:37] design. It's like interested in the
[16:39] color palette and and like really is
[16:42] this the right font for us? Do you think
[16:44] this font will be better? Goes better
[16:45] with the this entire thing is something
[16:49] that is um where humans come. That's
[16:53] like what the taste stays.
[16:55] >> So we have just a couple minutes left,
[16:57] but I want to leave our audience with
[16:59] some practical takeaways. So you've had
[17:01] a front row seat to how developers and
[17:03] the way that they work has evolved. So
[17:05] what's one belief or principle you hope
[17:07] the next generation of engineers doesn't
[17:09] lose sight of as AI becomes more
[17:11] dominant?
[17:13] professionalism matters
[17:16] and I think that we are losing it. the
[17:22] in one of the companies that I work they
[17:24] called it the rate of diluting the soup
[17:27] on on talent. How much talent can you
[17:31] bring that is not amazing but still you
[17:33] need working hands and I think that is
[17:35] this era is finally over cuz everything
[17:39] is overhyped there's so much money get
[17:40] people on get people on and you get
[17:42] people that don't like it like they like
[17:45] I do it because I love it I'm so crazy
[17:48] lucky to be doing it for be to to be in
[17:52] a thing that I really love to do people
[17:54] do it today because it's a good job like
[17:57] a lot of the engineers go because it's a
[17:58] good
[17:59] make money out of college. I'm making
[18:02] tons of money. I'm but they don't like
[18:04] it and in the end they don't really
[18:07] understand it all the way. We need our
[18:10] engineers to seriously and deeply
[18:12] understand the technology
[18:14] cuz that this will be the edge that
[18:16] they'll have as AI takes over and more
[18:19] and more jobs will be done by AI and
[18:21] more and more jobs will be done by I
[18:23] more and more work will be done by AI.
[18:26] The differentiator for humans is to
[18:30] actually know it in depth. Know the
[18:33] technology in depth. Wonderful. Yosi,
[18:36] thank you so much for your time today.
[18:38] Thank you all. Please join me in giving
[18:39] Yosi a round of applause.
