This is a story about a digital twin that knows exactly one thing. A writer named Dom, who writes for TechCrunch, got an AI avatar of herself made by Synthesia — and the avatar can only talk about one story. Dom trained the interactive version on her piece about why venture-backed startups commit more fraud than non-VC-backed startups. Ask it anything else, and it redirects to that story.
How Dom Got Her Avatar
Dom was at Synthesia’s new office in New York when the company offered her an avatar. She agreed without hesitation. The process involved a mini film studio inside the office, where Synthesia took photos of her and recorded two minutes of her voice.
She had to give consent for the avatars to be made. The result was Dom’s digital twin, which comes in four versions: a personal avatar that just reads whatever script is given to it, and two interactive ones that can talk back and listen, all with and without glasses.
The Technology Behind the Twin
The avatar is built from several models working together. There’s a voice-to-text model that turns speech into text, an agentic language model that understands the text and takes action, a text-to-voice model that turns a response into audio, and a video model that animates the avatar as it talks.
Synthesia built the video and voice models itself, but the company lets customers choose alternatives from other labs like Cartesia, ElevenLabs, Google, or OpenAI. Enterprises can host their avatars on whatever cloud they want, or pay Synthesia to host them.
What the Avatar Actually Says
Dom tested the personal avatar first, typing a generic script about how fall has arrived in New York, her favorite time of year. The voice matched hers well, and it didn’t pick up the hoarseness she had when she recorded the audio sample.
Then she tried the interactive avatar, which is deterministic — it only answers what it was trained on. Dom asked it where she was before TechCrunch and what part of New York she lived in. Each time it redirected to the venture fraud story.
Her friends found the voice interesting but not much like hers, and the likeness wasn’t as good as the personal avatar. Her mom called it “amazing,” which is pretty high praise from her. When her parents tried asking it questions only they would know, the model kept redirecting to the story.
“I don’t remember giving birth to two of you,” her mom joked after testing the model.
The Future of Journalism
Dom wonders if people would be OK with turning on the news and seeing it presented by an avatar. One investor told her no immediately. Others weren’t so sure.
Could avatars augment or even replace journalists? Would CEOs want to talk to an AI avatar of a journalist rather than a human one?
Dom likes connecting with people, writing stories, and researching new topics. But the biggest part of journalism is trust — and that doesn’t seem like it could ever be outsourced to an AI.
Outside Journalism, Cloning Yourself Is Fine
Dom is confident the idea of cloning yourself could be appealing outside of journalism. No more catching up on work after a holiday or vacation — a version of you could always be around, answering questions.
The avatar’s limits are the point. It’s not a general conversationalist. It’s a specialized tool, trained on one story, that answers questions about that story and nothing else.
The Numbers Behind Synthesia
Synthesia is a hot digital avatar startup alongside D-ID, HeyGen, and Colossyan. It hit a $4 billion valuation earlier this year and said last year it had crossed $100 million in ARR.
The company offers three main products:
- A video-creation and distribution platform with classic avatars, where someone types in a script and the avatar repeats it
- An agentic platform called Sessions where people can interact with avatars in surveys or roleplay
- An API platform where people can take Synthesia video and voice models and combine them with other tech services to build interactive avatars or other types of products
What We Make of It
The avatar is a technical tour de force, but it’s also a very specific toy. It’s trained on one story and one story only. It can’t improvise, it can’t branch, and it can’t answer questions about anything else — not Dom’s family, not her career, not the weather.
That limitation is the joke. It’s also the warning. An AI that can’t deviate from its script is a mirror, not a partner.
The avatar is a curiosity. It’s also a reminder that AI still needs humans to tell it what to do — and that the humans are the ones deciding what counts as journalism, what counts as trust, and what counts as a conversation worth having.
Key Facts Box
- Company: Synthesia
- Valuation: $4 billion
- ARR: Crossed $100 million last year
- Competitors: D-ID, HeyGen, Colossyan
- Models used: Voice-to-text, agentic language, text-to-voice, video (Synthesia’s own)
- Other labs: Cartesia, ElevenLabs, Google, OpenAI
- Training data: One TechCrunch story about VC-backed startup fraud
The avatar is a gimmick. It’s also a glimpse of a future where companies can clone their spokespeople — and where the clones are limited to what they were told to say. That’s the story Dom is telling, and it’s a story worth telling, even if the avatar can’t tell it for her.
Source material: “I created an interactive digital avatar of myself — and you can talk to it,” TechCrunch.
Get the Notebook.
The day's best stories and every fresh verdict, in plain English, in your inbox by seven. One email a day, no more.