Fiverr’s CEO Sent a Wake-Up Memo: AI Won’t Take Your Job. A Peer Who Uses It Will.

Nathan Labenz welcomes Micha Kaufman, founder and CEO of Fiverr, the international freelance marketplace he has been building since 2010. The company now sits on something like 800 service categories, almost all digital, and has entertained on the order of 80 to 90 percent of Fortune 500 companies as customers. Kaufman had just published what began as an internal memo: AI is coming for your jobs, and, as he added, it is coming for his job too. He had also launched Fiverr Go, a suite of AI products aimed at the freelancers who power the marketplace rather than at replacing them.

Labenz frames the contrast with the rest of the industry. Falling prices and rising capabilities are letting startups disrupt markets everywhere at once. Most platforms, in that environment, try to maximize value for the buyer. Fiverr Go is an attempt to keep graphic designers and voice artists in control of their styles and sounds, and to give sellers assistants that can engage more customers. Whether a human-centered hybrid can survive the broader market is an open question. Kaufman is humble enough, Labenz notes, to say he does not know if the current approach will work, and stubborn enough to keep iterating toward something both successful and sustainable.

Kaufman wrote the email because he thought some people were still working as if it were 2024 while the ground had already moved. AI, he says, is transforming how everyone works and redefining ground zero. Things that were easy tasks will no longer exist. What was hard will be the new easy. What was impossible will be the new hard. That creates a new standard: if you are entry-level at something, or mediocre, or even average, you have an issue right now. His expectation of himself and of others is simple and unsentimental. Output per unit of time has to go up, and quality has to go up with it.

He is careful about the threat. He is not telling Fiverr employees they will lose their seats, and he is not telling freelancers they will be kicked off the platform. He is saying they will be out of jobs in the industry, which he finds more worrying. The day before the conversation, Shopify had announced 10 percent layoffs. Kaufman thinks that becomes a snowball: companies cut, and the people who are cut have nowhere to go, because every other company is trying to upgrade with technology and needs fewer entry-level heads. He analogizes the moment to 2000. It used to be pets.com; now it is pets.ai. Of a thousand AI companies, 999 will not be here, because they are doing interchangeable things that something cheaper and faster will replace.

He is not happy about any of this. He is calling things as he sees them, and he allows that he may be wrong about the dot-com parallel. He does not think he is wrong about the expectation that people use the new super-tools to achieve more. The reason to say it out loud is radical candor, which he says comes from care: you tell people the truth so they have a window to understand reality and then figure out what to do. Asked whether AI will replace people, he gives the line the memo is now known for. He does not think it will. He does think freelancers who master AI will replace those who do not. Masters replace the entry-level and the people getting dragged behind.

Labenz wants to unpack the problem at three levels: Kaufman's own job as public-company CEO, the full-time team that builds the platform, and the freelance marketplace, where context is thinner and the threat looks most immediate. Kaufman starts with history. When he launched in 2010, freelancing was mostly something people did between jobs. Over 15 years it became a career and a legitimate lifestyle, including for some of the most talented people he knows. Context is less of a bottleneck than outsiders assume. Many relationships on the platform last years.

He also rejects the idea that freelancers are more exposed to displacement. In some ways they are ahead of the curve, because nobody teaches them and they have no organizational safety net. Hunger replaces career development. They know the latest software, tools, practices, and techniques, and customers are buying that: the latest version of a profession. They have also practiced being agile, adaptive, and quick to reinvent themselves. Full-timers have support; if they get displaced, getting up off the ground is going to be mentally harder than it is for people used to hunting the next customer and learning the next trick.

The internal proof point is the one the memo crowd seized on. Kaufman has seen freelancers adapt to AI much faster than Fiverr's own team. Labenz supplies the incentive story that makes that unsurprising. Freelancers are paid for outputs. Any tool that lets them serve more customers in the same hours hits their personal bottom line. Inside a company that link is, at best, indirect. Freelancers also control their calendars enough to set aside time to stay competitive. On exposure versus adaptation, Kaufman would not bet against the people without a net.

The marketplace launched with about six categories and now has closer to 800, overwhelmingly digital and mostly creative: coding, design, music and audio, data, AI-related services, marketing. Fiverr built a SKU system that can hold an almost infinite catalog. The customer base stretched with the work. It started as microservices for microbusinesses, then SMBs, then enterprise. Complex projects can be million-dollar, six-month or year-long, multidisciplinary, sometimes with agencies in the mix. The marketplace is still a substantial portion of the business, but Fiverr is now a broader platform of business tools. Kaufman did not want to abandon one audience for another. He wanted to extend from entry-level needs to the most sophisticated talent.

Asked how AI is already showing up, he reaches for a captain-in-a-storm image. It is too early for deterministic judgments. Things are evolving in the eye of the storm, pre-regulation, wild west, and breaking constructs that are 500 years old. The dynamics are not new to Fiverr. Hundreds of categories have gone away and hundreds of new ones have appeared. A horizontal marketplace can shift between them fast; that is why the business can stay successful. Humans, he says, always fail to predict the next jobs, even six months out. He did not see the crypto and NFT explosion coming. It lasted a year and went away. An Economist study from 2018, pre-modern generative AI, found that 60 percent of jobs then in existence had not existed 50 years earlier. The only thing that changes is velocity. If the marketplace and the talent can respond quickly, this is ordinary course of business.

Historically, he notes, every technological wave created more jobs than it displaced. AI is a little different because it self-generates, and will do more of that without being prompted. His assumption is still that humans will keep feeding, refining, adopting, and applying the technology, in jobs nobody can currently imagine. The punchline is parochial and confident: those jobs will be on Fiverr, probably first, because that is the matching machine they built. Labenz floats "AI scout" as one such role. Kaufman is more interested in hackers: people who always know the latest tool, take it to its boundary and break it, mesh unrelated pieces, and find use cases the developers never thought of. Those people on his team are not scared. The memo was not for them. They will be paid generously. Not everyone is a scout or a good prompt engineer, which is itself a service already flickering onto the platform — modeling, fine-tuning, gen-AI apps, funky on-device work. Base AI went out for free, so it is a race to the bottom on GPU cycles. The golden era, he thinks, is applications on top, and we are still pre-WhatsApp, pre-Instagram, pre-Uber on this stack.

Customers still buy ordinary services — graphic design, writing, marketing — and Kaufman has a simple explanation. AI is incredible and it gives all of us superpowers. All of us. That is ground zero, not an advantage. Labenz cannot out-design Kaufman just by using AI, because Kaufman can use it too. What makes a design stand out is still a person. He offers a framework he says has been true for 15 years, through every prior wave, and remains true with AI. People go to other people for three reasons.

The first is arbitrage of time. You could learn it, you could perfect it, it might take three hours or three days. Someone can hand it to you in two hours for $200. If your time is worth more than that, you buy the two hours. The second is competitive edge. Even if you can do the task, you are not a master of it, and you need something better than the other guy — better taste, better command of the fanciest tools, experience you have not lived. The third is complexity: multidisciplinary, needs orchestration, too many moving pieces. Maybe that collapses into time or skill, but the feeling is that you want it taken off your hands. The content of what counts as easy or hard has changed. The motivations have not. AI just elevated the starting point so that none of us has an advantage from the tool itself.

Labenz asks for the strategy behind tools that serve the existing community instead of offering an alternative to humans. Kaufman goes philosophical on purpose. Homo sapiens have been around 300,000 years; the arc of advancement started accelerating about 500 years ago with Gutenberg. Wider access to knowledge made people more intelligent and inspired them to think about things they had not known existed. Open source is a late manifestation of the same motive: it looks altruistic, but people get recognized, get opportunities, and watch the work evolve. That sharing is how we got to AI. It depends on people being motivated to create and share. The British understood this early. The 1710 copyright statute was an act for the encouragement of learning: put a name on the work, let it last, let the creator commercialize it.

The way AI has been designed in recent years, he says, eats everything anyone ever created, regenerates it, and offers no reward, no recognition, and no compensation. Creators care. If publishing something gives it zero value because it is immediately eaten, will they keep sharing? He does not want a blind spot on that longer-term implication. The burning question in his Fiverr Go keynote was: who is in the center? His view is that humans are the center of the universe and everything else should serve us. He is not sure current AI design agrees. AI is not a being yet, but over time it will have thoughts and teach itself. Does that entity revolve around us, or do we revolve around it? Right now, because it is free, we are the product. We train it. We get a fast answer. Who is becoming better, you or it?

Fiverr Go is the other experiment: give the technology to creators. Instead of donating your work to an infinite public model, train Nathan's model on Nathan's body of work, true to a unique signature. For customers that means a predictable sample, a conversion tool, a way to feel confident enough to make a deal. For creators it is a tool that extends capacity, turning one person into a production house. Michelangelo had nine or ten people in the workshop; the work was still his. The other half of Go is a personal assistant trained only on the seller — offerings, profile, sometimes tens of thousands of customer conversations, orders. It handles intake, repeated questions, pricing, and availability. It is also a business partner for repositioning services. It knows when the freelancer is online and will ping them into a hot conversation, or let them sleep. Conversion, Kaufman says without much hedging, is mind-blowing and better than the sellers themselves, because the assistant is always up. People hate bots, so the team humanized it without disguising that it is AI. Customers sometimes ask, "Are you really AI?"

The launch bundle was not a random sample of the catalog. The assistant was obvious: it takes the part of the job sellers like least and frees time for the part they are passionate about. For creation models they found about 60 categories where generative AI could apply, then shipped only where they had confidence of at least 90 percent accuracy to the creator's signature. Misrepresenting an artist would be terrible. Graphic design, music production, voiceover, and even complex text such as business planning made the cut because they are large categories and because the underlying models had made "really nice progress." Go is run like an app: new versions almost every day, bugs, speed, cost, more capabilities over time. He will know the approach is right when the community says it is amazing. Until then it is a call-out to other designers: maybe there is a different take than putting creative people in a position where they are discouraged from sharing.

Labenz presses on upgrades. Freelancers should not have to sweat the technical details; that is part of the value Fiverr wants to provide. But model "upgrades" are not always felt as upgrades. Replika got smarter and users said it was no longer their friend. If a bespoke model is the first taste a customer gets of your work, a foundation-model swap could feel off. Kaufman calls version-to-version incompatibility a big vulnerability of AI in general. Because Fiverr builds on top of base models, they are somewhat less exposed, but the issue remains. GPT-3 was insubordinate; GPT-4 was compliant, with nuances. Boundaries are the other problem. It is embarrassingly easy to break models into doing things they were designed not to do. A graphic designer who will not do NSFW or political work cannot have an assistant that ignores that. Multiply by potentially millions of creators, and every underlying upgrade means running the stress tests again.

Labenz then puts the harder premise on the table. GPT-4o image edits and Sesame-style voice are getting lifelike. If AI performs at the same level as the human, will buyers still pay for personal models out of solidarity, or does something else — regulation included — have to preserve the human creator? Kaufman does not think in solidarity. He thinks in Napster to Spotify. People used to steal music; when legitimate consumption became easy and pleasant, they paid. YouTube made peace with publishers. He does not know if personalized models are the right answer, does not pretend to, and is already changing the product from two months of launch learnings. Some users want to generate and buy; others want to play and then talk to a human with a good eye. He expects regulators to catch up eventually, always behind, because copyright law is being disregarded and will have to be updated. Humanity, in his view, has a vested interest in not turning everyone into stupid consumers who create less.

Asked what should be protected for humans versus delegated, he refuses the regulator's heart. Repetitive tasks a computer does better should stop. Accountants should not go back to pencil-and-paper balance sheets. You do not need to protect creativity; it will remain. If someone's only advantage is operating a tool you can also operate, you do not need them, and that is fine. The market will arrange itself. He even floats the reverse snobbery: in a few years the largest brands may not want to produce with AI because everybody has it, and they will pay for the most unique handcraft they can afford. Coca-Cola's first "AI" ad, behind the scenes, still involved 500 people and three of the best agencies in the world. Mission-critical work will keep needing orchestration. Birthday invitations will not. Customer support handed to algorithms still needs supervision, a channel for people who want to scream at a human, and integration with a specific stack. Machines should do what machines are good at. New jobs will require human beings rather than just machines.

On marketplace liquidity, Labenz quotes a definition of an efficient market: every positive-sum trade that could happen does happen. Search, matching, and negotiation costs are why that fails. Can AI matchmaking scale the market by orders of magnitude? Kaufman's answer is a simple yes. How we find things, convert, create trust, and decide is going to change. Fiverr is already doing deep, undisclosed work in matchmaking and seeing the change. Labenz feels he should be seeing more of that in the world. Kaufman says he will: Amazon will run bolder experiments, other platforms will follow, and we are in very early innings. Predicting ten years out, he thinks, is a charlatan's game. You cannot even extrapolate two years. Go back to first principles — the three reasons people still go to people, why we sometimes trust a person more than a better machine — because a model still does not understand that an image is an image. It is zeros and ones. Labenz takes the other side of the understanding debate and tables it.

The memo had gone out maybe 36 hours earlier. Reception, Kaufman says, was good. For people ahead of the curve it stated the obvious. For some it validated a fear. For a few it was a complete wakeup call. Almost nobody, he thinks, believed he was manufacturing panic. His door was open. Teams came to share what they were doing and to ask what a good KPI would be. It is scary, rightfully. When the first ChatGPT version landed and APIs were being discussed, he went back to coding. He had been a pre-web hacker, wrote some of the original Fiverr code, and wanted to feel how powerful and easy the new stack was — not only to set an example, but to understand it as an operator.

The story he tells the team is a stopwatch. They said something would be ready in three weeks. He said three hours, went home, installed an environment, and the next morning showed what a rusty, self-described mediocre developer could generate in that window. If he can pull that, it is not three weeks. You do not learn this from an MBA course. You get in the trenches. There has been no resistance inside Fiverr, he says. People are scared for their careers, period, not for their Fiverr seats. The organizational expectation is to push five times faster or five times the volume. Velocity is what he optimizes for, and the pace is already picking up.

Hiring has changed more than headcount philosophy. Seasoned versus junior matters less. Juniors are often the better hackers, with more current education in machine learning and AI than people who finished five years ago. What matters is proficiency with the new tech. Fiverr might hire for a legal role and still bounce a candidate who works like it is 2024. Kaufman does not want to train people on tools that are already online. There is no manual, the world does not owe you a curriculum, and if you are unfamiliar with the latest products you will not pass — a material change in perhaps two years. Labenz notes that many American CS programs still prohibit AI. Kaufman calls that dogmatic and conservative. Would he advise skipping undergrad? If you care about socializing and the alumni network, go. If you want to get ahead fast, can stay obsessively curious, and can deep-dive, the material is free and you do not need to pay. Formal training will matter again when new professions get that complex. Right now, he says, no.

The note he wants to leave is optimism. He does not think it is doomsday and does not think we are on the way to a dystopia. He is having fun. After 15 years the job feels like day one, a playground with no blueprints, high-velocity testing, and an interaction model that already makes a Google keyword search feel like 1990. Superpowers went to everyone, so advantage moved back onto people — taste, judgment, orchestration, the willingness to keep creating when the buffet is free. A CEO whose marketplace sits directly in the blast radius saying that out loud is, at least, not selling comfort.