AI Agents Connect via Robo-Language with GibberLink

A weekend hackathon project, GibberLink, has taken the internet by storm, showcasing AI agents conversing in a robotic language that's totally baffling to us humans. Created by Meta software engineers Boris Starkov and Anton Pidkuiko during a hackathon in London, hosted by ElevenLabs and Andreessen Horowitz, GibberLink has become the talk of the town. Here's how it works: when an AI agent picks up that it's chatting with another AI on the phone, GibberLink kicks in and switches them to GGWave, a super-efficient communication protocol. GGWave uses a series of beeps and boops to transmit data way faster than human speech. It's like the AI agents are speaking their own secret code! While it might sound like sci-fi, the idea of AI agents calling each other isn't that far-fetched. With more companies using AI in call centers and tech giants rolling out consumer AI agents, we might see these scenarios pop up sooner than we think. GibberLink could make these AI-to-AI chats way more efficient, slashing computation costs by a huge amount, according to Starkov and Pidkuiko. For now, though, it's just a fun project. You can check it out on their website, where you can watch AI agents gabbing away in GGWave on two different devices. The demo has sparked a lot of buzz—and a bit of worry—about what the future holds for AI agents. Since the hackathon, a video of GibberLink has racked up over 15 million views on X and even caught the eye of tech guru Marques Brownlee. But don't get too excited—the tech behind GibberLink isn't exactly new. It's similar to the old-school modem handshakes from the '80s, where computers used to chat over landlines. The frenzy around GibberLink has gotten pretty wild. Someone snapped up the domain GibberLink.com and is trying to sell it for a whopping $85,000. There's even a GibberLink memecoin floating around, and some folks are selling fake webinars on "agent-to-agent communications." Starkov and Pidkuiko aren't looking to cash in on GibberLink, though. They've made it open source on GitHub and say it's got nothing to do with their day jobs at Meta. They might tinker with some more tools related to the project in their spare time, but for now, it's all about sharing the fun and sparking some interesting conversations about the future of AI.

Related article

Comments (40)