Institute of Thinking№ 001 — 9 MAR 2026
Full-duplex, and what comes after audio
Most AI today works like a walkie-talkie. You speak, it waits, it answers, you wait. Humans don’t converse that way: we listen while we speak, interrupt, overlap, and correct ourselves mid-sentence. Full-duplex models do the same. They hear and talk at the same time, and that single change makes a machine feel present instead of merely responsive.
We see the full potential of full-duplex, and our mission is to carry it beyond audio. The next step is real-time video: AI that doesn’t only hear what you say, but sees what is happening as it happens, and answers while it is still useful. The value of an assistant often sits in the instant, a correction before the mistake, a warning before the fall, and a delayed answer is frequently no answer at all. Instant feedback is what lets AI become useful in real life, not only in a chat window.
Devotion, our multilingual full-duplex voice API, is the first step. Voice comes first because it is the hardest interface to get right, and every language deserves it.
Where instant feedback matters
These are examples of the commercial applications we have in mind. They are directions, not products we have shipped.
- Customer service and sales. Voice agents that can be interrupted naturally, in the caller’s own language, and that can see a product or a document the customer is showing.
- Training and coaching. Real-time feedback on a sales call, a language lesson or a physical skill, delivered while the learner is still performing it.
- Field work and maintenance. A technician points a camera at a machine and an assistant guides the repair step by step, hands-free.
- Healthcare support. Hands-free assistance for clinicians and care staff, and accessible companions for people who cannot easily use a screen.
- Accessibility. Live description of the surroundings for blind and low-vision users, and live conversation support for people who find typing difficult.
- Retail, hospitality and logistics. Multilingual front-of-house assistants, and warehouse or kitchen helpers that react to what is happening on the floor.
- Robotics and devices. Machines that can be talked to, and corrected, in the middle of a task.
We will write more here as we build.