A weekend hackathon project that enables AI agents to communicate over the phone in a robotic language, unintelligible to humans, has gone viral on social media.Called GibberLink, the project was created by Meta software engineers Boris Starkov and Anton Pidkuiko during a hackathon in London, hosted by ElevenLabs and Andreessen Horowitz.GibberLink allows AI agents to detect when they are speaking to another AI on a phone call. Once recognized, the agents switch to a more efficient communication protocol called GGWave, which uses a series of beeps and boops to exchange data at a faster rate than human speech.
GGWave, an open-source sound-based communication library, allows AI models to reduce processing costs significantly. While it might seem uncommon today for AI agents to call each other, companies are rapidly integrating AI-powered call center agents from startups like ElevenLabs, Level AI, and Retell AI. Additionally, major tech firms such as OpenAI, Google, and Amazon are launching AI assistants capable of handling tasks like making phone calls.
In this scenario, GibberLink could improve AI-to-AI communication efficiency, cutting down computing costs by an order of magnitude. Though the concept may seem futuristic, its underlying technology dates back to the dial-up modems of the 1980s, which used sound signals to transfer data.
Since its debut, a demo video of GibberLink has gone viral, gaining over 15 million views on X and even catching the attention of tech reviewer Marques Brownlee. The project has sparked curiosity—and concern—about AI’s future.Although GibberLink is not currently being commercialized, its creators have open-sourced it on GitHub and may continue refining the project in their free time.
