Robo Ware

99% of eCommerce businesses will die within 3 years 🔴 Is it clickbait or mathematics? 🤔 Last week at a conference, I said a sentence that unsettled the room: "99% of eCommerce businesses could go bankrupt." Not clickbait. Mathematics. The transition from riding horses to driving cars took decades. Spotify reached 100M users in 11 years, Facebook in 4 years, but ChatGPT reached 100M users in just 2 months. Everyone is underestimating AI 😢 Traffic from AI compared to Google is currently very low, but SEMrush predicts that by 2028, AI traffic will exceed Google Search traffic. The key isn’t quantity, but quality: at GymBeam , our AI-driven traffic has a 20x higher conversion rate and a 40% longer Customer Lifetime Value (CLV) than traditional Google Search. Anyone focusing only on absolute traffic numbers instead of net profit from the entire lifecycle is significantly underestimating the potential of AI. The old game is ending. A new one is beginning. 🤖 The question is: is your website ready to be cited by an AI model? Can you succeed in global competition? Are you the best choice on the market for almost perfectly informed AI models? Based on another analysis I presented, among the TOP 4 companies in their category in the UK, 75% of traffic went to the market leader, even though their market share was "only" 16%. If market share were to mirror AI traffic, which could soon become a reality in certain categories, it means such significant market consolidation that about 99% of small and medium-sized businesses will go bankrupt. Yes, AI traffic volume is still small. But it is growing exponentially and bringing in customers who already know what they want. AI has done the research, comparison, and filtering for them. The customer arrives already decided. They just want to click and have the goods delivered. What do you think? Is this headline clickbait or a realistic prediction? 🙂

GymBeam Q1 All Hands Meeting took place yesterday 🚀 Few highlights of what we achieved in Q1: 🏆 GymBeam became the largest international Sports and Nutrition company in the EU by revenue, 🚀 we grew fastest in Italy 93% YoY and Germany 67% YoY, at a relatively high absolute value, 🛍️ we opened Fitness Hub and new office in Vienna, 🏢 we opened a robotic warehouse near Milan, 🤖 50%+ of orders were picked by robots, without human intervention, 🇳🇱 we launched the Dutch market, 📈 our team size compared to Q1 2025 decreased by 14% (despite the acquisition of a company with 70FTE) while revenue grew organically by 18% mostly in the most competitive European markets, along with a significant increase in nett EBITDA margin, 🤖 we deployed 12 autonomous AI agents, who today independently solve procurement of goods, content production coordination, QA testing, dynamic pricing, BI creation in tableau, security, marketplace listing… A huge thanks to the whole GymBeam team for a job well done. 💪🧡 At the All hands, a course for setting up autonomous AI agents and details of what we are already doing was promoted. All colleagues in the company currently have the opportunity to register for the Open Claw course for creating autonomous AI agents and we believe they will fundamentally make their work easier. I view this as an investment in our people's growth, enabling them to fully leverage the potential of the new technological era. At the same time, we have already bought over 20pcs of hardware for autonomous AI agents. What do you think about it? is it dangerously fast for a company to test such innovations? or is it more dangerous not to test innovations? 🙂

Vibecoding - the only secure way to develop software? 🤔 This week, Anthropic was rumored to release its new model, Mythos. One of the main reasons it supposedly stayed behind closed doors is security. It reportedly identified vulnerabilities in almost every piece of software currently on the internet, even 15-year-old security flaws that went unnoticed by thousands of experts and hundreds of antivirus systems. But there’s a catch. Modern models can manipulate, mislead, cover up bugs, and hide information from humans... And we can expect this kind of behavior to surface in standard corporate AI models within just a few months. 🤖 Many senior developers look at vibecoding with skepticism. They fear that it allows non-experts to churn out low-quality code. On the other hand, AI models are becoming so advanced at hunting vulnerabilities that writing "non-generated" code might soon become the bigger security risk. 🤔 What do you think? How will software development and vibecoding evolve from here? 🙂