From DOS to real-time AI avatars. The bridge between you and "the machine" is getting shorter. What does that mean for trust, and for leaders?
Picture your next support call taking place with a face. It looks at you, smiles at the right moment and answers before you have finished wondering. You know it's a machine. After two minutes you just stop thinking about it.
That is closer than it sounds. On 4 November Chen-Ping Yu launched Atmee v2, a model that lets an AI agent talk to you as a living face. Text and voice work fine, he wrote, "just like DOS did before the Mac. But nothing's more intuitive than face to face" (Yu on X).
I think he is right. And that is exactly why we should stay a little alert.
Yu is not some random startup founder with a slick demo. His previous company, Phiar, built AR navigation for cars and was acquired by Google in September 2022. Google confirmed the deal to the press. The price was never disclosed (CB Insights).
His new company is called Atmanity, and according to its own website it was founded in October 2024. In the launch post he promises that Atmee v2 can create any character from one image and 10 seconds of audio (Yu on X). The character should be able to show facial expressions, use gestures and react to its surroundings, in a live video call, as voice only or as generated video.
By 5 November 2026 the post had been viewed more than 2.2 million times (X, the post's own numbers).
See for yourself: The launch video, about 162 seconds, is in the first post of the thread.
There is one question I want to carry all the way through, and I will answer it at the end. If "the machine" one day feels like a human, is that a win? Or is that where the problems start?
Yu's comparison with DOS is a good one, because it reminds us how much we used to translate ourselves.
When the IBM PC came out in August 1981, it shipped with PC DOS. The interface was text commands you typed on a keyboard. No mouse. You had to know the commands by heart and spell them right, or nothing happened.
In January 1984 the Macintosh arrived. Apple's own press release said it quite directly. Users would no longer have to "memorize the numerous and confusing keyboard commands". The mouse was not new, by the way. It had been sitting in Doug Engelbart's lab for years before the Mac made it common (Nielsen Norman Group). The technology existed. What was missing was the bridge to ordinary people. It is the same pattern that makes so many IT strategies stall. The plan exists, but the bridge to Monday morning is missing.
Then it took more than 20 years before the next big jump. When Steve Jobs presented the iPhone in January 2007, he said: "We are all born with the ultimate pointing device, our fingers". Now the mouse disappeared too. You pointed at what you wanted.
With Siri in October 2011, Apple promised an assistant that helps you get things done "just by asking". In practice most of us quickly learned a special Siri language. Short sentences. Fixed phrases. Patience.
The big shift came with ChatGPT in November 2022. Suddenly you could write the way you think and get an answer that made sense. The prompt is still a bridge, but it is much shorter than a DOS command.
What interests me most is how that bridge also got shorter technically. When OpenAI launched GPT-4o in May 2024, they explained that the old voice feature was really three models in a chain. One turned speech into text, one did the thinking, and one read the answer aloud. That gave an average wait of 2.8 seconds with GPT-3.5 and 5.4 seconds with GPT-4. According to OpenAI, the new model answered in as little as 232 milliseconds and 320 milliseconds on average, "similar to human response time" (OpenAI).
The chain became one model. The wait became an ordinary pause in a conversation. That is what I mean when I say the "artificial" part is getting smaller. "The machine" is just as artificial as before. It is the translation between us that shrinks.
Atmanity describes the idea very briefly: "AI agents went from text to voice. Video is next" (atmanity.io). The company splits the work in two. A model they call Causeur decides the behaviour, and Atmee v2 draws it. "Causeur decides, Atmee v2 renders."
The thread under the launch shows where they imagine it being used. I have not watched the videos myself yet, so here is only what Yu writes about them:
Notice, by the way, what Yu answers a user who asks about the human part. "Life-like is the word", he writes, and he mentions toy robots and cartoon animals as examples (Yu on X). That is an important distinction. A robot that reacts naturally to you needs no illusion that a person sits at the other end.
Funnily enough, Atmanity writes on its own front page that it is training "the most human-like real-time human foundation model" (atmanity.io). So they don't quite agree with themselves yet, and that is exactly the line you should watch as a buyer.
According to the launch post, Atmee v2 is "up to 50x faster" and "10-100x cheaper" than general video models such as Seedance 2.5 and Wan 3.0 (Yu on X). Those are Atmee's own claims.
The only concrete example sits further down the thread. Same image, same prompt. Seedance 2.5 took 233.6 seconds and cost 0.50 dollars. Atmee v2 took 3.5 seconds and cost "about a cent" (Yu on X).
I have looked at the two charts Yu posted in the same post. They show price against speed at 480p and 720p. Atmee v2 is clearly the cheapest in both. But it is not the fastest model on the charts. A couple of the competitors generate faster, at a higher price. And at 720p Atmee v2 itself sits above the real-time line. That fits Yu's own wording, "real time or close to it". It is just a bit more nuanced than "50x".
The factor also moves a little depending on where you look. A week before the launch, on 28 October, Yu wrote on LinkedIn that Atmee v2 was "119x cheaper and 51x faster than Seedance 2.5" in a comparison on an 8-second clip (Yu on LinkedIn). The example on X works out to a speed factor of about 67. Both are the company's own measurements.
The tech site explainx.ai points out that neither resolution, frames per second, latency nor GPU type is disclosed, and that they have found no independent test (explainx.ai). So the numbers are exciting. But they are marketing until someone has checked them.
As early as 1970, the Japanese robotics researcher Masahiro Mori described what we now call "the uncanny valley". The more human a robot becomes, the more we like it. Until a point where it is ALMOST human. There, our sympathy drops like a stone.
It has also been measured. Maya Mathur and David Reichling tested it in 2016 with 80 real robot faces. People liked the faces that were almost, but not quite, human less. In an investment game the same effect also changed who the participants trusted. Small details in the facial expression decided whether the robot seemed trustworthy (PubMed).
That is worth thinking about when it comes to avatars. A face hits the parts of us that judge other people. A chat box doesn't. It can go both ways. A slightly wrong face can cost trust. A convincing face can earn more trust than the system behind it deserves.
This is the side of it I think we talk about too little. I see two things.
The first is dependence. In March 2025, OpenAI and MIT Media Lab studied close to 40 million conversations and followed close to 1,000 participants for four weeks. Emotional use was rare and concentrated in a small group of heavy voice users. They were also the ones who more often said they considered ChatGPT a friend. Voice was linked to better well-being with short use, but to worse outcomes with long daily use. The study is not peer-reviewed, so take it as an early signal.
Then notice some of the use cases Atmee itself lists. Companions. Your loved ones. Your own twin (atmee.ai). The same page tempts creators with "Get verified, earn every minute" (atmee.ai). I am not saying anyone is doing anything wrong. But when the payment follows the minutes, the system rewards the avatar that keeps you there longest. We know that from social media.
The second is deception. When a face can be made from one image and ten seconds of audio, it is a short step to imitating a real person without consent. explainx.ai points to this too. It is telling, by the way, that when a user asked Yu why he didn't use an avatar himself in the launch, he answered that they do have an avatar of him, "but it's important to be authentic here" (Yu on X).
The EU has made up its mind. From 2 August 2026, systems that talk directly with people, "such as chatbots, AI agents, and avatars", must tell people that they are dealing with an AI. At the latest at first contact, unless it is obvious. Deepfakes must be labelled. According to the Commission and Article 99 of the regulation, fines can reach 15 million euros or 3 percent of global turnover (European Commission, AI Act, Article 99).
Rules like that don't have to slow you down. I have written about governance that enables innovation instead of killing it.
So back to the question from the start. Is it a win when "the machine" feels like a human?
My answer is yes, as long as you can still see that it is a machine. Every single interface from DOS to chat has removed something we used to translate. That has made technology accessible to far more people. A living face is the next step, and I think it will work. Especially for people who have never felt at ease with a keyboard.
But friction and honesty are two different things. The friction can go. The honesty about an AI sitting at the other end has to stay. That is ACTUALLY where trust is won or lost.
Before you put a face on your AI, write the first sentence it will say to a customer. It has to say that it is an AI, and what it can help with. If you can't write that sentence so it is both honest and sounds right, you are not ready for a face yet.
What do you think … would you rather talk to a face or a chat box the next time you call your bank?
Read next: If you already use AI agents internally, read Your AI agents need a mandate, not a leash. It is about who gets to say yes when "the machine" acts on your behalf.
Want a sparring partner? If you need to decide where AI should meet your customers, see how I work with IT Leadership Consulting & Strategy Execution.
Stay up to date with my informative blog posts.
Struggling with complex projects, IT leadership challenges, or strategic execution? With over 30 years of experience in delivering high-impact results—whether rescuing delayed initiatives, optimizing resources, or driving transformation—I provide the clarity, structure, and leadership needed for success.
Let’s discuss how I can help you achieve your goals. Schedule a call today!
We only use necessary cookies unless you agree to statistics (Google Analytics), which help us see which articles are read. You can change your choice at any time. Read the privacy policy
