Fiverr CTO Says AI Vibe Coding Fails at Enterprise Last Mile
Ai4
Non-technical users hit a wall on production security and system maintenance, preserving demand for human engineers who possess architectural judgment.
## Background
Published by the Ai4 channel on YouTube on August 27, 2025, this video features a presentation by Yossi Levin, Chief Technology Officer at Fiverr. According to the page description, Levin explores the unresolved challenges of vibe coding tools and addresses the "Last Mile" problem—the crucial final steps in software development that still require human expertise.
## TL;DR
Yossi Levin argues that while AI-driven vibe coding excels at rapid prototyping and early ideation, it cannot fully replace software engineers. Comparing vibe coding to assembling IKEA furniture, he
## Key takeaways
- [00:04](https://www.youtube.com/watch?v=wSKu-kzpnh8&t=4s) Software engineers have not engineered themselves out of a job; human system builders remain essential for production-grade software.
- [00:04](https://www.youtube.com/watch?v=wSKu-kzpnh8&t=4s) Historical technical shifts (like Java managed memory and AWS cloud infrastructure) provided temporary advantages, but eventually leveled the playing field for all developers.
- [00:32](https://www.youtube.com/watch?v=wSKu-kzpnh8&t=32s) Generative AI models often struggle with precise, conversational adjustments, making it difficult for non-technical users to debug or fix code when things go wrong.
- [00:32](https://www.youtube.com/watch?v=wSKu-kzpnh8&t=32s) When non-technical users vibe code, breaking changes lead to 'write-only code' that acts as an unmaintainable black box.
- [06:22](https://www.youtube.com/watch?v=wSKu-kzpnh8&t=382s) Vibe coding is like IKEA furniture: simple projects offer easy satisfaction, but complex builds quickly become overwhelming and frustrating without expert assembly.
- [09:16](https://www.youtube.com/watch?v=wSKu-kzpnh8&t=556s) Software built purely with standard LLM prompts suffers from homogeneity, lacking distinct visual identity and custom user flows.
- [09:16](https://www.youtube.com/watch?v=wSKu-kzpnh8&t=556s) Non-technical vibe coders lack domain vocabulary, preventing them from prompting for critical architectural needs like Single Sign-On or AES-256 encryption.
- [09:16](https://www.youtube.com/watch?v=wSKu-kzpnh8&t=556s) The 'Last Mile Problem' represents the barrier where a 90%-complete prototype must cross into security, compliance, scalability, and production deployment.
- [09:16](https://www.youtube.com/watch?v=wSKu-kzpnh8&t=556s) Entrepreneurs often waste critical strategic time trying to fix broken database schemas and production code instead of focusing on sales and growth.
- [15:05](https://www.youtube.com/watch?v=wSKu-kzpnh8&t=905s) The 'Next Mile Problem' involves long-term maintainability—such as zero-downtime deployments, database schema migrations, and backward compatibility.
- [15:05](https://www.youtube.com/watch?v=wSKu-kzpnh8&t=905s) At Fiverr, product managers are encouraged to vibe-code interactive prototypes to enhance spec discussions with software engineers.
- [15:05](https://www.youtube.com/watch?v=wSKu-kzpnh8&t=905s) Routine 'keyboard monkey' coding roles will be replaced, but engineers with deep architectural judgment and domain expertise will thrive.
## Detailed walkthrough
### [00:04](https://www.youtube.com/watch?v=wSKu-kzpnh8&t=4s) Introduction and the History of Technological "Edges"
Yossi Levin, Chief Technology Officer at Fiverr, opens his talk by addressing a pressing question in the modern tech ecosystem: have software engineers finally built the AI tool that will completely replace them? He gives an immediate spoiler—the answer for now is no. After polling the audience to confirm that most people are familiar with Fiverr (a digital freelance marketplace offering services across more than 700 categories ranging from software development to registering an LLC), Levin shifts to his personal history in programming to contextualize the current AI wave.
Reflecting on his start as a 13-year-old in his parents' living room, sitting on the carpet and plugging a ZX Spectrum Plus (a 48K RAM computer) into the television, Levin emphasizes how solitary early coding was. In that era, there was no internet, noStack Overflow, and no tech-savvy friends to consult; it was just the child, a printed manual, imagination, and sheer trial and error. Over his decades-long career, he has observed numerous technical shifts—from ancient assembly and basic routines to managed-memory languages like Java that eliminated pointer arithmetic and manual memory freeing, to the cloud revolution led by AWS, which replaced $50,000 upfront capex hardware investments with flexible, pay-as-you-go infrastructure.
Levin argues that each of these shifts, despite feeling like a massive competitive advantage, provided "zero edge" long-term because every developer gained access to them at the exact same time. The true final frontier in software creation has always remained human developers—the individuals who take business ideas in plain English and translate them into machine-executable instructions.
- **Fiverr Scope**: Digital services only across 700+ categories; freelancers handle digital builds, not physical construction.
- **Sinclair ZX Spectrum Plus**: 48KB memory, hooked directly to home TVs with no internet community support.
- **AWS Infrastructure Edge**: Eliminates $50,000 upfront hardware costs (capex) in favor of small, disposable cloud experiments ($100 or less).
---
[00:32](https://www.youtube.com/watch?v=wSKu-kzpnh8&t=32s) The "Delicate" Developer and the Pitfalls of Generative AI
Levin humorously addresses the stereotype of software engineers, labeling them "annoying, delicate souls" and "super snowflakes" who constantly demand training on new tech stacks or complain when their machines aren't fast enough. To illustrate how AI attempts—and often fails—to handle these human nuances, Levin shares a real-life prompt engineering experiment he conducted while making a laser-tag birthday party invitation for his daughter.
Levin wrote a highly descriptive, explicit prompt requesting an image of two girls and two boys (modeling his daughter, two male classmates, and another friend). The generative image model repeatedly failed to generate the correct balance of genders. When Levin attempted to reason with the model by conversing with it, the interaction deteriorated into an endless, frustrating argument. Levin felt as though he was arguing directly with a pedantic developer. Ultimately, instead of trying to be "correct," Levin hacked his way through the session by altering his inputs until he got a usable result.
He connects this image-generation anecdote directly to "vibe coding." When a non-developer attempts to generate complex software using an LLM, it works brilliantly until a bug occurs. Once the machine output breaks, a non-technical user cannot read the raw code, lacks the vocabulary to query the LLM effectively, and is left completely stranded.
[06:22](https://www.youtube.com/watch?v=wSKu-kzpnh8&t=382s) The IKEA Effect and Building a Simon Game in a Zoom Meeting
Levin uses IKEA flat-pack furniture as a direct metaphor for vibe coding. Building a simple 13-step or 23-step IKEA item yields immense satisfaction ("Look, mommy, see what daddy built!"). However, attempting a 100-step wardrobe setup over an entire weekend with a father-in-law creates absolute misery, leading to misplaced doors, disassembly loops, and frustration. Had he known the effort required beforehand, Levin asserts he would have gladly hired an assembler—a reality that explains why IKEA acquired TaskRabbit.
To test vibe coding firsthand, Levin attempted to build a web application during a boring executive Zoom meeting. He asked an LLM to generate a complete, working clone of the classic memory game *Simon* contained within a single HTML file so he could run it locally with a double click.
While the AI delivered an operational single-file HTML application, it initially relied on playing external `.wav` audio files. When Levin instructed it to eliminate all external asset dependencies, the AI generated synthesized sound frequencies on the fly using native JavaScript audio APIs—a capability Levin was surprised to learn existed. However, as he requested further features (such as persistent high scores), layout bugs emerged. Text began jumping around the UI, and error messages shifted screen elements out of alignment. Lacking formal JavaScript expertise at the time, Levin couldn't convince the LLM to fix the layout glitches, ultimately declaring it "good enough for the kids" and giving up—just like a partially assembled IKEA wardrobe.
---
[09:16](https://www.youtube.com/watch?v=wSKu-kzpnh8&t=556s) The Commodity of Vibe Coding: Zero Edge and the "Last Mile" Problem
Levin synthesizes his core criticism: "vibe coding is IKEA for code." While functional for basic tasks, software built entirely through vibe coding suffers from three structural flaws when compared to commercial applications:
1. **Homogeneity**: Much like staying in European Airbnbs where every kitchen has identical IKEA dishes, software created via standard LLM prompts lacks distinctiveness. Businesses need custom user flows and unique visual brand identities to stand out.
2. **Lack of Domain Vocabulary**: Non-technical users do not know the underlying architectural vocabulary (such as requesting Single Sign-On [SSO] or AES-256 encryption), preventing them from requesting necessary security and compliance standards.
3. **Zero Competitive Advantage**: Because every creator has access to the exact same high-level AI tools, capabilities are leveled universally. If 10x tools allow everyone to build 10x faster, a product must be 100x more unique in its underlying domain expertise to capture attention.
Levin highlights actual support requests posted to Fiverr by overwhelmed entrepreneurs who attempted to vibe-code their own startups. He cites posts from non-developers pleading for help after deleting production databases, breaking Laravel setups, or getting stuck trying to integrate complex third-party APIs and backend database logic. He asks business founders whether struggling with database schemas is truly the best use of their executive time compared to sales, marketing, and product strategy. This threshold—where a working prototype must cross into security, compliance, and stable architecture—is what Levin labels the **"Last Mile Problem."**
- **"Write-Only Code"**: Code generated by LLMs that functions on the surface but is completely unreadable, unmaintainable, and acts as a complete black box to the creator.
- **Fiverr Trend**: Spike in client requests from users who built an initial AI prototype but need professional engineers to fix broken workflows or secure production environments.
---
[15:05](https://www.youtube.com/watch?v=wSKu-kzpnh8&t=905s) The "Next Mile" Problem and the Future of Engineering
Getting an initial codebase deployed to production is only the beginning of a software system's lifecycle. Levin defines the **"Next Mile Problem"** (and the miles beyond) as the ongoing maintenance and evolution of software over time: executing database schema changes without dropping live tables, rolling out zero-downtime feature deployments, maintaining backward compatibility, and managing complex system dependencies.
While professional engineers naturally navigate these maintainability challenges, non-technical vibe coders quickly hit a wall. When an uninitiated user wants to update a production application, their prompting vocabulary often tops out at "I need a new field," causing the underlying unreadable codebase to break.
In summary, Levin validates vibe coding as an exceptional prototyping tool that accelerates early-stage ideation. At Fiverr, he actively encourages product managers to vibe-code interactive prototypes so developers can discuss concrete, functional models rather than abstract feature specifications. However, bridging the gap from prototype to production-grade infrastructure remains extremely difficult without human systems engineering.
Concluding with his outlook on job security, Levin assures the audience that software engineering jobs will not be completely eliminated by computers just yet. However, the nature of the role has permanently changed. Engineers who act merely as "keyboard monkeys" writing routine code will be replaced; developers who possess deep architectural judgment, domain knowledge, and true engineering expertise will thrive.
- **Booth 261**: Fiverr's booth at the Ai4 conference, encouraging attendees to connect with human engineering talent to complete their vibe-coded projects.
## Notable quotes
> [00:00](https://www.youtube.com/watch?v=wSKu-kzpnh8&t=0s) Creating new systems has never been easier or faster, thanks to vibe coding. But have software engineers finally engineered themselves out of a job? Will developers become obsolete? — Yossi Levin
## Video description
Yossi Levin, Chief Technology Officer, Fiverr
Abstracts
Creating new systems has never been easier or faster, thanks to vibe coding. But have software engineers finally engineered themselves out of a job? Will developers become obsolete?
In this session, Yossi Levin, CTO at Fiverr, explores the unresolved challenges that vibe coding tools leave behind and address the “Last Mile” problem—those crucial final steps that still require human expertise. Whether you’re a skeptic, an innovator, or just vibing, this talk charts where human system builders remain essential in turning ideas into successful, production-ready solutions.
Yossi Levin, Chief Technology Officer at Fiverr, argues that "vibe coding" with generative AI will not replace human software engineers anytime soon. While large language models excel at rapid prototyping, universal access to these tools erases any competitive advantage and leaves non-technical creators stranded before reaching production-grade software.
Universal Tools Yield Zero Competitive Edge (0:04)
Drawing on decades of programming experience—from coding on a 48KB RAM ZX Spectrum Plus without internet access, to adopting managed-memory languages like Java, to seeing AWS replace $50,000 upfront hardware investments with sub-$100 cloud experiments—Levin argues that major technological shifts ultimately provide "zero edge." Because every developer gains access to new abstractions simultaneously, the tools themselves do not create a lasting advantage.
Levin applies this same principle to vibe coding. When everyone uses the same AI tools to build 10x faster, the resulting applications become commodities, much like European Airbnbs stocked with identical IKEA dishes. Without custom user flows and distinct brand identities, AI-generated software suffers from homogeneity, meaning businesses now need 100x more uniqueness in their underlying domain expertise to stand out.
The IKEA Effect and Prompting Dead Ends (6:22)
Calling vibe coding "IKEA for code," Levin compares AI software generation to assembling flat-pack furniture: building a 13-step item is satisfying, whereas a 100-step wardrobe causes enough misery to explain why IKEA acquired TaskRabbit. Non-technical users experience similar friction when generative AI misinterprets instructions or introduces bugs. For example, when Levin prompted an image model to generate a birthday invitation featuring two girls and two boys, conversational corrections devolved into a frustrating argument until he bypassed the issue by hacking his inputs.
To test code generation directly, Levin prompted an LLM during a Zoom meeting to build a clone of the memory game Simon inside a single HTML file. The model impressed him by replacing external .wav files with native JavaScript audio APIs to synthesize sound frequencies on the fly. However, when he asked to add persistent high scores, UI text and error messages shifted out of alignment. Lacking deep JavaScript expertise, Levin could not prompt the AI to fix the layout bugs and abandoned the effort—illustrating how non-developers get stranded when machine output breaks.
The Last and Next Miles of Software (9:16)
According to Levin, LLMs frequently generate "write-only code"—software that functions superficially but remains an unreadable, unmaintainable black box to its creator. Crossing from a 90%-complete prototype to a commercial system is what he terms the "Last Mile Problem." Non-technical founders lack the architectural vocabulary to prompt for essentials like Single Sign-On (SSO), AES-256 encryption, regulatory compliance, or scalable database schemas. On Fiverr's marketplace, which spans over 700 digital service categories, Levin notes a spike in overwhelmed entrepreneurs hiring professionals after deleting production databases, breaking Laravel setups, or stalling on third-party API integrations.
Beyond initial deployment lies the "Next Mile Problem": the ongoing maintenance and evolution of live systems. Whereas professional engineers manage backward compatibility, zero-downtime feature rollouts, and database schema updates without dropping live tables, non-technical users lack the vocabulary to safely modify production code. A basic prompt such as requesting "a new field" can easily break an unreadable codebase, raising the question of whether founders should spend executive time debugging schemas rather than focusing on sales and strategy.
Despite these structural limitations, Levin endorses vibe coding as a powerful ideation tool, noting that Fiverr encourages its own product managers to vibe-code interactive prototypes so engineers can evaluate functional models rather than abstract specifications. Ultimately, he forecasts that while AI will replace "keyboard monkeys" who only write routine code, software engineering roles will endure and evolve to favor professionals with deep architectural judgment and domain expertise.
[0:00] Yossi Levin: Creating new systems has never been easier or faster, thanks to vibe coding. But have software engineers finally engineered themselves out of a job? Will developers become obsolete?
[0:32] Yossi Levin: annoying, delicate souls
[0:32] Yossi Levin: super snowflakes
[6:22] Yossi Levin: Look, mommy, see what daddy built!
[6:22] Yossi Levin: good enough for the kids
[9:16] Yossi Levin: vibe coding is IKEA for code
[15:05] Yossi Levin: I need a new field
Original transcript
[0:04] Hi, I'm Yos Levin. I'm CTO at Fiverr.
[0:07] Um, and I'm going to talk about this.
[0:12] Uh, have we finally managed to build the
[0:15] thing that will replace us for real? Uh,
[0:18] the answer is at the end. Spoiler, no.
[0:22] Um,
[0:24] about Fiverr. Who here knows Fiverr?
[0:29] Okay. So, we're good. So, for those who
[0:31] don't know Fiverr, Fiverr is a
[0:33] marketplace for freelancers. We have 700
[0:36] more than 700 categories.
[0:39] Software development will register your
[0:41] LLC. Everything digital,
[0:44] not something physical. We will not come
[0:46] and build something for you. That's like
[0:48] our freelancers don't do that, but
[0:50] everything digital.
[0:54] About me a little bit. This is how it
[0:57] all started. I don't know if if you can
[1:00] recognize this. Who can recognize this
[1:02] computer? Who is old enough to recognize
[1:05] this computer? Everybody like hands
[1:07] down. Uh this is a Spectrum Sinclair for
[1:11] ZX Plus Plus with 48K memory. It was an
[1:17] amazing machine at the time. That's how
[1:19] I started programming at the age of 13
[1:22] on my parents' living room sitting on
[1:24] the floor connecting this thing to the
[1:26] TV. And and
[1:30] the big difference then and today is
[1:33] that I had no one to talk to. There was
[1:37] no internet to go on and find a
[1:39] solution. I had no friends that knew
[1:43] what I was doing with this weird
[1:44] machine. You were alone. You and the
[1:48] book and your imagination and your
[1:50] creativity. This is it.
[1:56] And I was lucky since I started so long
[2:00] ago and it is long ago to see tons of
[2:03] evolutions going on in the industry
[2:07] like software like this was very
[2:09] advanced. It had basic on it but before
[2:11] that there was like ancient languages
[2:14] that then evolved suddenly you have Java
[2:18] and languages that actually manage
[2:20] memory for you. You don't have to take
[2:21] care of pointers and freeing memory and
[2:23] leaking memory. It was amazing.
[2:27] But everybody got it. It was not an
[2:29] edge. Everybody got it at the same time.
[2:32] Your decision if you adopt it or not,
[2:34] but everybody got it at the same time.
[2:37] And then came AWS.
[2:42] An amazing edge. You don't need capex
[2:45] because before that you had to buy, you
[2:47] have an idea, you have to buy this
[2:49] hardware in order to try the idea out.
[2:52] Put $50,000 just to get started. M AWS
[2:55] poof can do whatever you want. Can try
[2:58] something today, scrap it tomorrow. No,
[3:00] I'm done. There goes $100 maybe.
[3:04] But everybody got this edge at the same
[3:07] time. It was accessible to everybody. So
[3:10] it was not really an edge. We think it's
[3:12] an edge. The fact that you early adapter
[3:14] and so on, but these technologies did
[3:17] provided no edge to anyone.
[3:21] The last frontier
[3:24] is developers.
[3:27] The people that will take your ideas and
[3:30] translate them from English
[3:32] to computerish, whatever language.
[3:35] And developers are
[3:41] annoying. I'll be like super like I
[3:45] picked my words carefully here.
[3:46] Annoying.
[3:48] Uh the delicate souls. you you super
[3:52] snowflakes. They you have to take care
[3:53] of their needs. You have to train them
[3:55] up. A new technology train me up. I'm
[3:57] insulted. I'm like my computer is not
[3:59] fast enough. Yes.
[4:02] I started I said I won't get started and
[4:04] I started. Um
[4:07] so genai
[4:10] and please take a look at this prompt
[4:13] and I'll start from this and we'll get
[4:15] to software development and tell me
[4:17] what's wrong. Can you detect what's
[4:20] wrong?
[4:29] Yeah, exactly. Thank you. I asked for
[4:32] two girls and two boys. I was very
[4:35] descriptive. I was super accurate. I was
[4:38] actually proud of this prompt because I
[4:40] said exactly what I want. It was for my
[4:43] uh girl, my my daughter birthday invite.
[4:48] It was a laser tagged theme and it was
[4:50] doing it with two boys from her class
[4:53] and another friend from her class and I
[4:55] described everybody accurately but I
[4:57] didn't get what I want. So I said,
[4:59] "Okay, I know what to do. I'll just
[5:02] explain what I want."
[5:07] Yeah. I started running this discussion
[5:10] and I hoped
[5:13] Gai will go with me. But I started
[5:16] running an argument and I did not like
[5:18] it at all.
[5:21] At all. It felt like I'm working with
[5:24] the developer.
[5:29] What do you do?
[5:32] You got it right. Like
[5:38] I gave up. I didn't try to be correct. I
[5:41] did exactly what we do when we program.
[5:44] So, do you behave a certain way?
[5:46] Whatever. I'm not going to fight this
[5:48] way. I'm just going to get the job done.
[5:52] And that's how I hacked it. Now, imagine
[5:55] this was a software, a piece of software
[5:57] that was created by a machine, a a Vibe
[6:03] coding machine, and you're not a
[6:05] developer.
[6:07] What you going to do? You're going to
[6:09] look at the code, and you have no idea
[6:10] even what to ask. You have no idea how
[6:12] to talk to it. You have no idea what
[6:14] you're reading. It's super cool when it
[6:16] works. Awesome when it doesn't. Uh awful
[6:19] when it doesn't.
[6:22] IKEA. Everybody has this at home. Like
[6:24] at least one.
[6:27] Good. So, this takes 13 steps to build.
[6:31] Super gratifying. Amazing success. This
[6:34] is 23 steps to build.
[6:38] Daddy is amazing. Look, mommy. See what
[6:40] daddy built. Awesome.
[6:43] This is horrible.
[6:46] This is a hundred steps to build. I
[6:49] stepped a whole I spent the whole
[6:50] weekend with my father-in-law who is a
[6:54] lovely person.
[6:57] Really,
[7:00] we spent the whole weekend, like two
[7:02] days fighting this. We set the doors up
[7:05] and then we didn't set up set them up
[7:07] correctly. We had to take them apart,
[7:09] put them again, and so on and so on
[7:11] until we figured it out. Had I known at
[7:14] the at the beginning what I knew at the
[7:17] end,
[7:19] I'll get someone to do this. I don't
[7:22] need this type of gratification from
[7:24] building something that is not my job.
[7:28] And
[7:30] there's no that's the reason IKEA got
[7:33] task rabbit because everybody else
[7:35] suffering from the same thing.
[7:38] Back to software. This is my first
[7:41] application
[7:43] with Genai
[7:46] management meeting. Super boring on
[7:48] Zoom. I decided to try something out.
[7:51] Lucky I didn't have to explain what
[7:52] Simon is. Everybody know what Simon the
[7:54] game is, right? Perfect. And I just I I
[7:59] want a Simon game. And a Simon game. I
[8:04] said I want it in one HTML file. I got
[8:06] it. One HTML file. Double click.
[8:09] Amazing.
[8:11] Few problem.
[8:13] That sounds He was playing a W file.
[8:17] Why would you play a W file? So I don't
[8:20] want attachments. Figure it out for me.
[8:22] Cuz I knew what the problem is. I knew
[8:24] what I was looking for. So it created
[8:26] sounds with
[8:28] JavaScript which I didn't know even is
[8:31] possible. And I was pretty amazed. And
[8:33] then I started adding features, high
[8:36] score and so on. And it started being
[8:39] complex. I don't know JavaScript to save
[8:42] my life. Well, I didn't know before
[8:44] that. Then I knew cuz I had to save my
[8:47] life.
[8:50] And
[8:54] it was actually it was fun. It worked.
[8:56] Mostly worked. In some cases, the text
[9:00] jumps, the error message causes things
[9:02] to move on the screen. I couldn't
[9:04] convince it. I said it's for the kids.
[9:06] They will appreciate it still like the
[9:08] IKEA stuff. They'll still think I'm
[9:10] amazing. I'm done.
[9:14] So
[9:16] conclusion,
[9:18] vibe coding is IKEA for code, but
[9:24] things that work for IKEA does not work
[9:27] for um
[9:30] for software that you build. One,
[9:36] you don't want to have the same software
[9:39] as everybody.
[9:41] Like with IKEA, I go on Airbnb in
[9:43] Europe. Everything feels like home. All
[9:46] the dishes in the kitchen are the same
[9:48] as I have at home. All the places are
[9:51] exactly the same. Feels like home. I
[9:55] don't want my software to feel exactly
[9:56] the same of as any other software. I
[9:59] want my flows I want my flows to be
[10:01] special. I want my look and feel to be
[10:03] special in order to get noticed. Second
[10:07] thing,
[10:09] I don't know know if I even speak the
[10:11] language.
[10:13] I don't know to ask for SSO or to ask
[10:16] for AES 256. I like if you don't know
[10:20] this, then you have no idea what you're
[10:23] asking for.
[10:24] And
[10:26] with all these tools, and I talked about
[10:28] it earlier, there's zero edge.
[10:32] Zero edge. We are all the same. We all
[10:37] got a level up. We all have the cheat
[10:39] sheet. Good luck winning now in the new
[10:42] game because it's a new game. It's zero
[10:44] edge to everybody. And
[10:49] if now you can create 10x
[10:53] 10x people can create I read it
[10:55] somewhere. 10x people can create
[10:57] software 10x
[11:00] faster. Then now you have to be 100x
[11:03] more special in order to stand out in
[11:06] the crowd. And that's the big situation
[11:09] now. And it's not about
[11:13] I'll make my application pink and green
[11:16] and and people will pay attention. It's
[11:18] about the expertise in everything, the
[11:21] human expertise that has to be put into
[11:23] this in order to convince it to do
[11:26] whatever you want it to do. And
[11:32] let's
[11:34] uh uh enjoy uh this amazing stories. Um
[11:39] looks super sad. I think most of us have
[11:43] deleted the database in production one
[11:46] day or the other.
[11:48] Um hoping that the backup is good enough
[11:52] and and
[11:56] nobody wants to be there.
[11:59] and the the CV generators and and and
[12:02] whatever that that's like how you look
[12:05] at this
[12:06] and
[12:08] we in Fiverr we are about human talent
[12:13] and we have people and we see the the an
[12:16] amazing spike starting the beginning of
[12:18] the year like crazy and before that it
[12:20] was gradually growing of people
[12:24] realizing that it was a cool
[12:27] vibe Bing thing, but when reality comes,
[12:32] can I get an expert, please?
[12:35] And I love the the little quotes.
[12:39] Hey, can you please check this workflow
[12:42] for me? It was working perfectly, but it
[12:45] accidentally stopped working.
[12:49] And the brutal honesty of the guy
[12:52] saying, "Uh, help. I started the project
[12:55] on Lavel. I don't know what I'm doing.
[12:59] Um,
[13:01] and the other very long cry for help
[13:07] where I think about it, there's a
[13:08] business owner trying to set his to set
[13:11] up his own site, his own site, his own
[13:13] application, his own business,
[13:16] and now he's dealing with wiring third
[13:19] parties and databases and APIs.
[13:23] Just get someone to do this for you.
[13:25] Is this the most important thing you can
[13:27] do with your business? What about
[13:30] selling marketing?
[13:33] Something different than just Yeah, I
[13:35] hacked it.
[13:37] So, I'm not saying people cannot do it.
[13:40] You can do it. But is this the best use
[13:43] of your time? Do you really want to
[13:45] spend the weekend with your
[13:47] father-in-law building a a closet?
[13:50] And that's what we see. This is for a
[13:53] specific vibe coding uh um application.
[13:56] I'm not going to name names.
[13:59] Uh but that's what that's the type of
[14:01] help people are asking for and we see it
[14:04] on all of them.
[14:07] People need help and that's what we call
[14:10] the last mile problem. You are almost
[14:14] there. You are almost there. And then
[14:17] you have to deal with the illities,
[14:20] the all 10,000 of them. And you know,
[14:24] little things, security, compliance. Oh,
[14:26] I'm saving your your ID. Oh, I forgot to
[14:29] encrypt it. Why do I save your ID? All
[14:32] the things you had no idea you had to
[14:33] care about,
[14:35] but you have to care about them.
[14:38] And write only code was something that
[14:42] uh in my first development team my boss
[14:46] was always like uh um throwing this code
[14:49] away. This is write only code. Go fix
[14:51] it. Like you can never read it. You have
[14:53] no idea what's in it. It works but but
[14:56] who cares? So that's mostly what we see.
[14:58] Code that nobody can
[15:01] what do we do with this black box now?
[15:05] But getting the application out the door
[15:08] is just the beginning. The fund starts
[15:12] now
[15:13] and that's the next mile problem and the
[15:16] mile after that and the mile after that.
[15:19] How do you roll out changes without
[15:22] breaking the universe?
[15:24] How do you roll out changes gradually?
[15:27] How do you make a change to the database
[15:28] a ch and add a column without god forbid
[15:32] deleting the database as you go? How do
[15:35] you change a flow? We're keeping your
[15:36] backward compatibility, your behaviors
[15:38] or anything. Things that humans do
[15:42] easily, professionals
[15:45] do easily, even with Vibe coding tools.
[15:48] But you know what to ask. You know the
[15:51] language.
[15:54] Put my father-in-law in front of this. I
[15:56] need a new field. That's like the
[15:58] maximum that that they'll get to. And
[16:00] again, he's an amazing person. I'm not
[16:02] saying anything bad about him. is
[16:03] seriously amazing.
[16:06] So all this next mile problem is where
[16:09] most of the energy goes in a regular
[16:12] software as well.
[16:15] So to summarize,
[16:19] it's an awesome capability. Like
[16:21] seriously, I'm not like I'm any way
[16:23] saying anything bad about VIP coding.
[16:26] It's awesome. I played with it. It's
[16:28] enjoyable.
[16:30] It's brings tons of value. I like to
[16:33] have my product managers build what they
[16:35] want instead of trying to describe what
[16:37] they want, struggle with their own
[16:39] flows, figure it out because they're
[16:40] technical enough. I like them to do this
[16:43] so developers and them can talk about
[16:44] something that works. And maybe the
[16:47] developers take even take the code and
[16:48] move forward with it if it is logical.
[16:53] All the players got to level up. So we
[16:56] have to deal with the other things
[16:58] marketing presence how what makes you
[17:02] special how will you stand in the crowd
[17:05] and well vibing for prototype I think I
[17:08] made this point clear vibing for
[17:10] prototype is uh tons of fun vibing to
[17:15] production grade is super super hard you
[17:18] need a professional
[17:21] so
[17:23] we do have a
[17:26] job is not going to be taken away by
[17:28] computers
[17:30] yet.
[17:31] I don't know what happens like next year
[17:33] because things are moving so crazy. But
[17:35] for now, we're safe.
[17:38] But we have to understand that our job
[17:40] is now a bit different.
[17:42] We need to look into the existing code.
[17:44] We have to bring the the we are if
[17:47] you're pure coder, a keyboard monkey,
[17:51] then you will not survive. You will be
[17:53] replaced. If you are a professional that
[17:56] actually seriously knows what's going on
[17:59] and can add proper engineering value,
[18:02] you'll thrive because you'll be needed.
[18:06] And this is it. We're at the booth 261
[18:09] um
[18:11] to get some uh good talent uh to get
[18:14] your vibe coding things going. Uh, thank
[18:16] you so
