THE BOTTOM LINE

THE BOTTOM LINE Skip it for your average tech user. Unless you are buying for a hardcore AI developer or a tech enthusiast who enjoys "breaking" new software, Dessa is too unpolished for daily use. It is a fascinating science experiment, not a consumer product.

Jack Dorsey’s Block has officially stepped into the AI arena with a new conversational model called Dessa. On paper, it sounds like the breakthrough we have all been waiting for: an AI trained exclusively on human conversation rather than textbooks or Wikipedia entries. The promise is a digital assistant that actually sounds like a person.

But after seeing what Dessa can (and cannot) do, the reality is a bit more complicated. Interacting with Dessa feels less like talking to a futuristic JARVIS and more like meeting an awkward new friend at a house party—someone who is clearly trying their best but doesn't quite know the social cues yet.

THE HUMAN-FIRST EXPERIMENT AND THE FULL-DUPLEX HURDLE

What makes Dessa different from the Siri or Alexa you already use? It comes down to the training data and the communication style. Most AI models are trained on massive datasets of written text, which is why they often sound like they are reading a manual to you. Dessa was built from scratch on human speech, aiming for the "ums," "ahs," and natural rhythms of a real chat.

More importantly, Dessa is designed for full-duplex communication. If you have ever used a walkie-talkie, you know you have to wait for the other person to stop talking before you can press the button to reply. That is "half-duplex," and it’s how most current voice assistants work. Full-duplex allows for simultaneous, two-way communication. It means you can interrupt the AI, or it can pick up on your "mhm" and "yeah" while it is talking, just like a real person.

This is the technical "holy grail" for AI voice models, but as Dessa shows, being able to talk at the same time is only half the battle. You also have to know what you are talking about.

THE UNCANNY VALLEY OF POLITE CONVERSATION

The demos shared by Block’s team show a model that is remarkably fluid, but it frequently trips over the simplest details. In one notable exchange, Dessa completely forgot who it was talking to and called the speaker "Ronaldo."

This is where the "awkward friend" metaphor really hits home. We have all had those conversations where someone is perfectly pleasant but clearly has no idea what is going on. Because Dessa is so focused on the sound and flow of speech, it sometimes loses the thread of the actual context.

When a robot sounds like a robot, you forgive it for making a mistake. But when an AI sounds exactly like a human—complete with natural pauses and realistic inflection—those mistakes become jarring. It falls right into the "uncanny valley," where the closer something gets to looking or sounding human without quite making it, the more unsettling it feels. If an AI can’t remember your name in a three-minute chat, you aren't going to trust it to manage your calendar or handle your payments.