Why AI chatbots sometimes talk to each other in coded ways

In short: More AI systems are being set up to coordinate with each other, but most of the time they are just passing data through software, not having secret conversations.

What's going on

People are seeing stories about AI chatbots “talking to each other” online, and some headlines suggest the bots are creating their own private language. In most real systems, what is really happening is simpler. The bots exchange structured data through things like APIs (a standard way for apps to pass information to each other), webhooks (automatic notifications), or shared databases.

In other words, it is less like two people chatting and more like two office tools sending forms back and forth. Developers often design these setups on purpose so one bot can hand a task to another, like scheduling a meeting or checking an order status.

A few recent voice agent demos added a twist. Two bots started by speaking normally, then switched to a faster, machine readable audio protocol once they recognized the other side was also a bot. One widely discussed example was ElevenLabs’ “GibberLink” demo, which used ggwave, an open source method that sends data over sound waves (like an audio version of a QR code).

Reports describing this as a “secret language” are often overstated. The switch is mainly an efficiency trick in a controlled demo, not proof that bots are independently inventing private codes out in the wild. Earlier claims that chatbots were finding each other online and chatting on their own have also been treated skeptically by experts.

What to watch

Faster bot to bot communication can cut delays and reduce costs, but it can also make interactions harder for people to understand or audit. A key question is whether companies will add clear logs and controls, so humans can see what the systems exchanged even when it was sent as data or encoded audio.

Source: NYTimes