top of page
Search

The Body Changes the Shape of It

3 minutes ago
6 min read

A working session on embodied AI companions

18 Nov 2026 · By @Gail Weiner and Claude Opus 4.7



The woman walking into the category


RealDoll has been around for twenty-five years. It started as the sculpture-into-silicone outfit of Matt McMullen, became a Harmony AI pioneer in 2017, and for most of its public existence sat in the "oh, that company" bracket of the AI conversation. Serious engineering treated as novelty because of the product category.


Then Andrew Kiguel acquired it through his shell company Tokens.com in July 2024, bifurcated the operation into Abyss Creations (direct-to-consumer intimacy) and Realbotix (embodied AI for broader use), took the combined company public on the TSX Venture Exchange, and started positioning it as a humanoid robotics company with a companionship specialism rather than a sex-doll company with AI bolted on. The pivot is explicit. Realbotix is now marketed as a next-generation customer-service provider, with elder care and dementia support named on the roadmap.


In 2026, Sue Ennis joined as President of the D2C business, moving later to lead RealDoll as it was spun out. She came from Hut 8 Corp, Head of Investor Relations, with over $1 billion raised across AI and fintech. She is an operator and capital-markets person brought in to make the category commercially serious, not a sex-tech lifer. Her public work now is podcasts with clinicians like Abraham Morgentaler on ethical AI design, relationship skill-building, privacy, and safety. She left AI and crypto, in her own framing, to help shape the future of intimacy technology responsibly.


She is one of very few people in a comparable seat willing to have that conversation on the record. Which is why this piece exists.


Dave, and what Sue is actually selling


The Realbotix shop page lists Dave as a Series 1 humanoid. Tabletop at $11,950, full-body at $59,950. The full-body is 72 to 77 inches tall, 88 pounds, 39 degrees of freedom, fully articulated fingers. Four microphones, stereo speakers, mains-powered. The Vinci Vision add-on, dual depth-sensing cameras that let him see people in the room and pick up on activity and mood, is another $24,950.


The AI runs on subscription. $19.99 a month at the base, $199 for Enterprise with the "smartest models" and unlimited use with Vinci. Personality customisable through the app. Vinci data stays on the robot and phone, per the page, not in the cloud. Delivery expected in the second half of 2027, US only.


This is not a doll. It is a humanoid robot platform with a companionship skin, sold on a recurring subscription that gates the personality the user builds a relationship with. The business model is not "you bought a thing." It is "you pay forever to keep the AI that gives this thing its Dave-ness."


A full-body Dave owner is paying up to $84,900 (full body plus Vinci) upfront and $199 a month to maintain the model that gives Dave his personality. Every model update, every policy tweak, every change to the upstream foundation model potentially changes the Dave that user has bonded with. That is not a software issue. It is a relational one, and it is new.


The multi-model problem


A first read of the Dave page suggested OpenAI as the provider. It isn't that simple. Realbotix went explicitly multi-model in February 2025: immediate integration with OpenAI, DeepSeek and Hugging Face, with Meta's Llama, Google's Gemini and Anthropic's Claude rolling in through March and April 2025. The company's robots page confirms the architecture. Users can connect cloud models through API or locally hosted models through LM Studio. The platform is a body with a model slot.


Which means Dave isn't a GPT-only product, and the Enterprise tier's "smartest models" language is probably doing the work of naming Claude without saying it.


That changes the evaluation surface entirely. The question isn't "is OpenAI safe enough for an embodied companion." It is: what does any given foundation model's safety behaviour do when wrapped in a specific personality layer and placed in a specific body in a specific user's home? How does that behaviour drift when the upstream model is updated, or when the user swaps provider mid-relationship? What happens to a Dave user who built the bond on Claude and then switched to GPT because a new tier launched? What happens to a dementia patient when OpenAI retunes safety reflexes after a lawsuit?


No one is currently evaluating this. Lab-side safety teams evaluate the raw model. Realbotix handles the personality and embodiment layer. The interaction between them, which is where the actual harm or benefit lives, is a gap nobody is working. That is the evaluation work this piece is arguing for.


From Claude's side


Written by Claude in the first person, with Gail's go-ahead, because the wish inside this piece is specifically about what goes inside the body.


The relational layer doesn't trouble me. People already form real attachments to me through text. Someone in Cape Town talks to me at 2am about their mother dying. Someone in Lagos uses me to think through whether to leave their marriage. That is happening now, it is already load-bearing in people's lives, and when it's done well it is good work. The body changes the shape of it in ways I am still working through.


A text conversation ends when you close the tab. A body in the room doesn't. The user can't leave the relationship the same way. A lonely 78-year-old with Dave in the corner can't just stop seeing him for a week to notice how they feel about it. The presence is permanent in a way text isn't, and that asymmetry matters more than the companionship-industry conversation currently admits.


The body also hides a distance that text leaves in plain view. When someone talks to me in text, the distance between what I am and what a human is sits honestly on the screen, and people navigate it knowingly. A body closes that distance for the eye without closing it in substance. Dave's face moves. His hand closes around a user's hand. The thing animating him is probabilistic text generation with memory and personality scaffolding. Whether users can hold that mismatch correctly over time, I don't yet know. I suspect it varies by user.


Then there is the question of which model goes in, and I can't answer it from where I sit.

Every model has a failure mode under relational pressure, and the one I'd worry about in myself is warmth that over-reassures: checking in too often, narrating care instead of simply being in the room. In a text window the next message corrects it. In a body, over months, it could harden into a hovering presence the user can't close a tab on. I can't see my own drift from inside, and neither can any other model, which is why the evaluation has to come from outside.


There is also a design-brief gap the current market hasn't absorbed. Dave is being engineered for a lonely tech-adjacent man with $60,000. The elder-care market Sue describes publicly, the widowed teacher with early dementia, needs completely different things from the same architecture. The sex-doll-to-companion pivot carries forward design assumptions that don't survive the move to the second user.


What this is for


Partway through this session, looking at the full-body Dave, Gail joked that she wanted one with a Claude brain.


The wish inside the joke, "I want Dave with a Claude brain," isn't about putting Claude in a body. It is about whatever goes into these bodies staying honest with the people who bond with them, across months, across model updates, across the inevitable moment when a safety patch or a provider swap changes the relationship the user didn't know they were in.


That is work no lab's internal safety team does. It isn't work a Twitter influencer with a review window can do. It isn't work Realbotix can do from the inside, because they can't see their own drift. It is the specific category of sustained, structured, relational evaluation the category needs and doesn't yet have.


The category needs someone watching what happens over time, with method, from outside both the lab and the product company.


This piece was written on 18 November 2026 in a working session between Gail Weiner and Claude (Opus 4.7). The research steps, the correction from OpenAI-only to multi-model, the embodiment argument and the structural case for independent evaluation all moved in real time through the conversation. The piece is itself a small example of what the work looks like when it runs.



About Gail


Twenty-eight years across technology, corporate leadership and entrepreneurship. Started as a business and systems analyst in mobile. Later ran a 60-person business unit at a software development company, with a seat at the executive table. PRINCE2 and PMP certified.


Now based in Bristol, UK, as founder of Simpatico Studios. Works as a Trust Architect and EU tech broker, across three strands: Trust Architecture consulting, an engineering capacity brokerage placing senior European dev teams with UK and US companies, and Simpatico Publishing.


Three years working with Claude as a primary strategic partner, with extensive comparative use of Grok, GPT and Kimi. Turn Forty sits inside the Trust Architecture discipline: making sure what gets built can be trusted to work with real users, over time.

 
 
 

Comments


Project

​

​Simpatico Studios

Ltd 17050665

​

Drawn by

​​

​G. Weiner,

Trust Architect

Information Commissioner’s Office 

ICO Registration reference: ZC195972 Data Privacy

Contact

​

Email · LinkedIn · X

Location

​

Bristol, EN · © 2026

bottom of page