Kaya.gs

Gabriel Benmergui was a passionate Go player—the #1 player of his generation in South America—and briefly studied Go in Korea with dreams of becoming a professional. But he realized he could contribute more to the game through software than competition. In 2011, he noticed that existing Go servers were built on outdated technology and lacked the platform mentality needed to grow the game's community. "There was a huge need for a new space, that would develop new features faster and be more modern than the current options," Gabriel recalls. This gap—combined with his newfound software engineering skills—sparked the idea for Kaya.gs.

With just one year of web engineering experience under his belt, Gabriel and his co-founder built a responsive, real-time HTML Go server in about 2 months using cutting-edge (for 2011) technologies: Node.js for event management, socket.io for WebSockets, Ruby on the backend, and a custom React-like rendering engine on the front-end. The technical challenges were immense—WebSockets weren't reliable across browsers, the JavaScript ecosystem was primitive, and real-time games are inherently hard. But the magic moment came when they placed their first stone on the board and watched it update in real-time across the network.

Gabriel and his partner launched a crowdfunding campaign on Idea.me (Argentina's Kickstarter alternative) but hit a critical problem: credit card companies blocked international donations on day one. With no revenue, they did what scrappy founders do—they built their own payment solution overnight using PayPal, with a playful "Candy" section on the website that led unsuspecting users to their donation tiers. It worked. They raised $20,000 from over 300 backers, some donating $100+ each. One user even donated Bitcoin at $3/coin (they'd later sell when it hit $60, "thinking we hit the jackpot"). The crowdfunding success wasn't just about money; backers became passionate evangelists. Gabriel's active forum presence—responding to every user inquiry while other servers ignored their communities—drove organic growth. By launch (Christmas 2011), they had thousands of registered users, 100 concurrent players, and over 10,000 games played.

What worked brilliantly: the community vision, the Friday night feature releases that users would flock to see, the user-driven product prioritization, and solving the chicken-egg problem of multiplayer platforms with AI bots. "We were already so much ahead in those terms," Gabriel says of their feature velocity compared to competitors releasing one feature every three years.

What failed catastrophically came from three reinforcing problems. **First, product direction**: Gabriel was "selling vision" and building innovative features while neglecting core reliability. Frequent server crashes made the playing experience unreliable—unforgivable for a real-time game. **Second, engineering inexperience**: They used experimental browser features that got disabled, had no DevOps (Gabriel manually restarted crashed servers), and accumulated technical debt that slowed future development. **Third, morale collapse**: After a year of ramen-budget living and stalled growth (monthly revenue stayed around $2,000), founder friction emerged. "When things aren't going well, it turns a mechanism in your brain that wants it to fail fast," Gabriel reflects. The three failures fed each other—technical debt prevented quick iteration, which hurt growth, which killed morale, which destroyed productivity. They ran out of money and energy at the same time.

Gabriel eventually got a full-time job to pay rent and shut down Kaya.gs, open-sourcing the code. He's since become a Senior Product Engineer at CircleMedical, a YC-backed startup. Looking back, he wishes he'd focused on one core metric (number of games played) rather than shipping features that impressed but didn't drive engagement. He also regrets not building a stronger technical team earlier to avoid the execution bottlenecks that plagued him. But the hardcore users understood: Gabriel was there every day, did his best, and couldn't make the traction stick. No angry emails came.