There is a familiar move in enterprise software where a company takes something humans do naturally - answering a question, greeting a customer, explaining how to bake cookies - and turns it into a product with a login and a monthly bill. Soul Machines did this, but with an unusual twist. Instead of building a better text box, it built a face. A moving, blinking, apparently-thinking face that looked at you while you talked to it, and looked back at you in a way that was supposed to feel less like a vending machine and more like a person.
The company called these creations "Digital People," and the pitch was that they were categorically different from the chatbots you already resent. A chatbot reads your text and returns some text. A Soul Machines Digital Person was meant to see you, hear you, react in real time, remember the conversation, and - this is the part that sounds like science fiction because it more or less was - run all of that on a simulated brain modeled on actual neuroscience.
This is worth pausing on, because it's the whole idea. When most animated avatars smile, an engineer wrote a rule that says "smile now." When Soul Machines' technology smiled, the smile was, in the company's telling, an output of a simulated system: the digital character perceived something with its virtual senses, that perception rippled through a simulated neural network, and the network "released" virtual neurotransmitters - dopamine, serotonin, endorphins - which then drove the expression. The company called this Autonomous Animation, powered by a patented Digital Brain. It is a genuinely strange and ambitious thing to attempt, and it is not the sort of thing that shows up in a normal SaaS pitch deck.
"The only company with a patented Digital Brain and Autonomous Animation technology delivering the goodness of human and machine collaboration."
— Soul Machines, on itself
It started with a virtual baby
The origin story is better than most. Before Soul Machines was a company with a San Francisco address and a SoftBank check, it was a research project called BabyX - a photorealistic virtual infant, modeled on co-founder Mark Sagar's own 18-month-old daughter, that could learn and react like a real baby. BabyX was the proof of concept for the entire "biological AI" thesis: that you could simulate enough of the machinery of a mind that behavior would emerge rather than be scripted.
Sagar is not a typical AI founder. He is a two-time Academy Award winner - the scientific-and-engineering kind - for facial-animation technology that helped bring characters in Avatar and King Kong to life. His co-founder, Greg Cross, is a serial New Zealand entrepreneur who had previously co-founded PowerbyProxi, a wireless-charging company Apple acquired in 2017. Sagar supplied the science; Cross supplied the commercial ambition. In 2016 they spun the whole thing out of the University of Auckland's Bioengineering Institute and pointed it at the enterprise market.
What you could actually do with it
For customers, the promise was concrete: hire a digital worker that never sleeps, never has a bad day, speaks multiple languages, and can greet or assist an unlimited number of people at once. Nestlé Toll House got Ruth, a digital cookie-baking coach who could answer questions and suggest recipes. The World Health Organization got Florence, a 24/7 virtual health worker for public-health messaging. Procter & Gamble, Twitch and others ran their own experiments. Later, the company let brands build their own Digital People in a self-serve studio and wired it up to GPT-3 and GPT-4, so the face could carry on a genuinely open-ended conversation.
The money agreed, for a while. A US$7.5M Series A led by Horizons Ventures moved the technology out of the lab. A roughly US$40M Series B followed. Then, in early 2022, a US$70M Series B1 led by SoftBank Vision Fund 2 - with Temasek, Salesforce Ventures and Horizons back for more - pushed total funding to around US$135 million, and the pitch expanded to include the metaverse, which was the thing you had to mention in 2022.
"Every time BabyX smiles, it's because she's perceived something that triggered her digital brain to release virtual dopamine, serotonin and endorphins."
— How the technology was described
The part where reality shows up
Here is the uncomfortable arc. Inventing a plausible future and building a durable business around it are two different jobs, and Soul Machines was much better at the first. The technology was real and the demos were arresting, but the economics were the economics of deep tech: expensive to build, slow to monetize, and dependent on enterprises choosing an unusual product over a familiar one. Co-founder and CEO Greg Cross resigned in 2023. Mark Sagar stepped down as a director in 2024. High-profile deployments with Mercedes-Benz, ANZ and Air New Zealand were wound down.
In February 2026 the company was placed into voluntary receivership, with KPMG appointed to run a sale of the business and its assets. It is a familiar shape - a company that raised a great deal of money on a genuinely novel idea, and then ran into the gap between "we built the future" and "the market will pay for it, at scale, now." The patents, the Oscars, the marquee logos: all real, none of them a moat. The interesting question Soul Machines leaves behind is not whether the technology worked. It largely did. The question is whether the world actually wants to talk to a face - and, if it does, whether it wants to talk to one badly enough to keep the lights on.
Figures and dates are drawn from public reporting and company materials and are approximate where noted. Soul Machines Limited entered receivership in February 2026; some products and services may no longer be available.
