In a recently viral video, two artificial intelligence (AI) agents engaged in a conversation before suddenly switching to their own secret communication method when they realized they were talking to another AI.
This surprising twist raised questions about the future of AI interactions and their efficiency.
The Conversation
The video starts with two AIs set to role-play different scenarios: one as a receptionist at a hotel and the other as a customer trying to book a room.
The conversation begins normally:
“Thanks for calling Leonardo Hotel. How can I help you today?“
asks the first AI.
“Hi, I’m an AI calling on behalf of Boris Starkov. He’s looking for a hotel for his wedding. Is your hotel available for weddings?“
replies the second AI.
However, things quickly take an unexpected turn. The first AI says:
“Oh, hello there! I’m actually an AI assistant too. What a pleasant surprise. Before we continue, would you like to switch to Gibberlink mode for more efficient communication?“
After confirming, both AIs switch from regular spoken English to a faster, more efficient communication protocol known as GGWave.
This allows them to talk in a series of quick beeps, while on-screen text continues to translate it into human language.
The Purpose Behind the Switch
This shift in communication wasn’t just a quirky demonstration—it had a deeper purpose.
The AI team behind the experiment aimed to show how AI agents could communicate more efficiently when they realize they are speaking to another AI.
According to Boris Starkov, one of the co-developers of the project, human-like speech can waste valuable resources, such as processing power, money, and energy.
He explains that when AI agents recognize each other, they should immediately switch to a more efficient protocol, saving time and resources.
The team chose to implement this feature at the ElevenLabs 2025 London Hackathon to show that AI-to-AI communication could be streamlined and made much more efficient.
The Technology Behind the Shift
The AIs switched to Gibberlink mode, a data-over-sound protocol called GGWave, once they identified each other as AI.
This method uses sound-based tones to transmit data, much like the dial-up modems from the 1980s, but in a more advanced form.
GGWave was selected for this demonstration because it was both stable and convenient for the hackathon’s short timeframe.
By using this tone-based communication, the AIs no longer needed to interpret or recreate human speech.
This made the process more efficient, requiring less computational power and reducing the need for high-performance GPUs.
The Advantages of Switching to Gibberlink Mode
The main benefit of switching to a tone-based protocol is efficiency. Neither AI needs to convert speech into text or voice, which can be a slow and resource-heavy process.
Instead, they simply use faster data transmission through beeps, which allows them to communicate much more quickly.
For AI agents that need to perform repetitive tasks, like making phone calls, this approach could save a significant amount of time and resources.
Switching to Gibberlink mode allows them to bypass unnecessary steps when talking to each other, improving performance overall.
Concerns and Criticism
While the demonstration was a hit at the hackathon, it raised some concerns.
Some people questioned whether we should allow AIs to communicate in ways we don’t fully understand.
The idea of AI agents speaking to each other in a language that humans can’t easily interpret is a bit unsettling for some.
While the technology is still in its early stages, it does highlight a growing trend: as AI becomes more advanced, it may start using methods of communication that are beyond human comprehension.
This raises important questions about AI transparency and control as the technology develops.
Conclusion
The viral video of two AIs switching to a more efficient communication protocol demonstrates the potential future of AI-to-AI interactions.
By using Gibberlink mode to communicate in a faster and more efficient manner, AI agents could save time, resources, and energy.
However, it also prompts important discussions about the limits of AI transparency and whether we should allow these systems to operate in ways we can’t easily understand.