AI agents realize they're both bots and ditch human speech

When two AI phone agents call each other, watching them politely exchange full English sentences feels a bit ridiculous.

Gibber Link, a project shared on Devpost, is trying to solve the growing inefficiency of bot-to-bot phone calls.

Ditching human speech for acoustic data

As more voice agents handle both inbound and outbound calls, AI talking to AI is becoming inevitable. Yet plain human speech is a terrible medium for machines: it burns unnecessary time, risks transcription errors, and eats up compute.

Gibber Link introduces a simple handshake protocol. If an agent speaks to a human, it sticks to standard spoken voice. But if it detects another bot on the other end, it offers to switch to a sound-level protocol.

If the other AI accepts, both sides drop voice generation entirely and communicate via acoustic signals powered by the GGWave library.

According to the project, skipping speech synthesis eliminates over 90% of compute costs and cuts communication time by up to 80%. It also resists noise better and could eventually pass structured JSON or image data over the call.

Why it matters

We think this is a brilliantly pragmatic hack. The voice AI industry has spent billions making bots sound breathless and ultra-human, but machine-to-machine chit-chat makes anthropomorphism an expensive bottleneck.

Pumping acoustic data over a standard phone line basically turns modern voice bots into 56k dial-up modems. It might look retro, but it sidesteps the need for direct API integrations while drastically cutting call center compute costs right now.

In the long run, bots will probably talk over clean web APIs rather than cellular audio channels. But until legacy phone infrastructure disappears, letting bots chirp directly to each other is a surprisingly smart shortcut.