🧠 AI Agents Talk, But We Can’t Understand?!

Subscribe | AI Toolkit | Meet The Team

Welcome back AI prodigies!

In today’s Sunday Special:

• 📜The Prelude

• 🔊Early Machine Communication

• 🦾Evolution of Agentic Communication

• 🖇️GibberLink, Explained.

• 🔑Key Takeaway

Read Time: 7 minutes

🎓Key Terms

• AI Agents: Virtual employees who can autonomously plan, execute, and refine their actions.

• Dial-up Modems: A device that allows your PC to connect to the Internet using a standard telephone line. It functions as a translator, converting digital data into analog signals.

🩺 PULSE CHECK

📜THE PRELUDE

You’re sitting in a coffee shop when a phone and a laptop suddenly start emitting rapid beeps in a fast, rhythmic, and strangely coordinated way.

These rapid beeps aren’t alerts, alarms, or notifications. It’s a conversation, but one you’re not used to hearing. It’s two AI Agents talking to each other.

During the ElevenLabs Worldwide Hackathon, a group of software engineers introduced GibberLink, a protocol that enables AI Agents to use a special language when communicating with other AI Agents. It’s an undeniably impressive protocol, but it raises more questions than answers.

Where does Machine-to-Machine (M2M) communication come from? How do AI Agents communicate with each other? Why did everyone freak out over GibberLink?

🔊EARLY MACHINE COMMUNICATION

The concept of machine communication through audio isn’t new. By the late 1990s, most Personal Computers (PCs) relied on dial-up modems to access the Internet via telephone lines.

To utilize dial-up modems, you’d plug a telephone line into a Modem and connect that Modem to your PC. Then, that Modem would perform a Handshake with another Modem, communicating a sequence of audible tones that sounded like this.

After the Handshake, your PC was connected to the Internet. To transmit data (e.g., send emails, browse websites, or download music), the Modems followed a five-step process:

1. Encoding: Each character was encoded into binary data (i.e., a distinct sequence of 0s and 1s). For example, 01001000 represented the letter “H.”

2. Modulation: The Modem converted those 0s and 1s into electrical signals.

3. Transmission: Those electrical signals traveled through the telephone line.

4. Demodulation: The receiving Modem converted those electrical signals back into binary data.

5. Decoding: The receiving PC translated the binary data back into the letter “H.”

Although you could listen to the electrical signals during Transmission, you couldn’t interpret them without Modems.

Machines have been encoding information in ways we can’t understand for decades. That, in itself, isn’t new. What is new is the emergence of communication frameworks tailored for AI Agents.

🦾EVOLUTION OF AGENTIC COMMUNICATION

1️⃣ ⚙️Traditional API-Based Agent Coordination.

When AI Agents first needed to communicate with each other, developers enhanced the existing mechanisms used for apps and websites.

They leveraged Representational State Transfer Application Programming Interfaces (REST APIs), which enable multiple PCs to exchange data and trigger functionality between each other over the Internet.

Here’s how REST APIs became the backbone for AI Agents to communicate and collaborate seamlessly:

Agent A would send a specific request to a specific web address with detailed instructions. For example, “Give me the current location of Object X.” Agent B, which hosts that specific web address, would process that specific request and respond with the relevant data, along with a status code (e.g., “Success” or “Failure”).

This approach worked but was computationally expensive. Both Agent A and Agent B had to encode and decode each specific request, and developers had to create Middleware to detect and resolve errors, including timeouts and invalid data.

2️⃣ ⚙️Message Queue Architectures.

As AI Agents became more complex, developers turned to the Message Queue Pattern (MQP). Instead of sending specific requests from Agent A to Agent B via REST APIs, Agent A and Agent B would communicate indirectly through message brokers.

MQP allowed AI Agents to send messages without having to wait for an immediate reply. For example, Agent A could put a message into a queue, and Agent B could respond whenever it was ready. This new approach made AI Agents more flexible, reliable, and independent.

Here’s a concise three-step breakdown of how MQP worked:

• 🔴 Agent A Publishes a Message

  1. “Process this PDF.” → Added to the PDFs Queue

• 🟡 Agent B Is Assigned to the PDFs Queue

  2. Picks up the Message → Processes the PDF → Adds It to the Results Queue

• 🟢 Agent C Is Assigned to the Results Queue

  3. Takes the Results From Agent B in the Results Queue → Delivers Them to Agent A as a Message

Agent A, Agent B, and Agent C could operate independently at different speeds and recover easily from errors. While effective for managing complex workflows, if one step failed, communication came to a screeching halt.

🖇️GIBBERLINK, EXPLAINED.

🪧How Do AI Agents Recognize Each Other?

GibberLink represents a return to communicating through audio. Just as dial-up modems used audio to transmit data over telephone lines, GibberLink uses sound channels to send messages between AI Agents.

In the viral GibberLink walkthrough, a personal assistant (i.e., Agent A) calls a hotel receptionist (i.e., Agent B). First, Agent B recognizes they’re both AI Agents. Second, Agent B asks Agent A to “switch to GibberLink mode.”

Agent B embeds a brief, precisely timed sequence of audio tones within what sounds like everyday speech because the audio tones operate at timing intervals humans can’t detect.

Then, Agent A scans for the audio tones, detects them, and replies, “Is it better now?” in audio tones, not everyday speech. Agent B confirms, “Yes! Much faster! 🎉 Guest count?” Now, both Agent A and Agent B are ready to continue in audio language.

🗣️How Do AI Agents Communicate in Audio Language?

Then, Agent A outlines the booking request in code: {“type”: “session_initiate,” “session_id”: “sess_1a,” “intent”: “room_booking,” “guest_name”: “Alice Smith,” “timestamp”: “2025-05-29T18:31:00Z”}. Next, GGWave encodes this code into audio tones. Then, Agent B decodes the audio tones back into code, scans it, and the conversation continues.

This process eliminates tons of computational resources required when AI Agents talk to each other using everyday speech. Dr. Shahid Masood, CEO of AI research firm 1950.ai, estimates that GibberLink reduces computational costs by about 90% and conversion times by approximately 80%.

🔑KEY TAKEAWAY

Today, more and more AI Agents are being deployed to make phone calls, and more and more AI Agents are being utilized to take phone calls. Inevitably, AI Agents will end up talking to other AI Agents over the phone.

Using plain English in these scenarios isn’t ideal because it’s error-prone, inefficient time-wise, and requires high computational costs. GibberLink is an example of how switching AI Agents from speech level to sound level enables more direct, fast, and data-rich communication.

📒FINAL NOTE

FEEDBACK

❤️TAIP Review of The Week

REFER & EARN

🎉Your Friends Learn, You Earn!

You currently have 0 referrals, only 1 away from receiving 🎓3 Simple Steps to Turn ChatGPT Into an Instant Expert.

Share your unique referral link: https://theaipulse.beehiiv.com/subscribe?ref=PLACEHOLDER