The most interesting thing about the home system Benjamin Code built for his mother is not that AI appears somewhere in the stack. It is how much effort went into removing decisions from her side of the interface.
His mother has early-onset Alzheimer’s. In a long walkthrough published in late June, the French developer shows a collection of tools assembled around her apartment: an always-available display, a simplified phone, video calls, sensors, automations and dashboards for the people supporting her.
This is not a medical device, and it is not a care protocol to copy. It is a personal system built around one person and one family. That specificity is exactly what makes the design process worth studying.
Put the complexity on the builder’s side
A recurring rule is that his mother should not have to learn how the system works. The system has to fit the actions that already remain familiar.
A large television can act as a passive information surface before returning to ordinary TV use. Her iPhone is reduced to a small set of useful actions through Assistive Access, Apple’s simplified iOS experience for people with cognitive disabilities. Apple lets a supporter choose available apps and strip their presentation down substantially.
Video calling follows the same direction. Benjamin uses Jitsi, which supports self-hosting, to create a call path that requires as little interaction as possible from the person receiving it.
None of these components is extraordinary on its own. The design work is in how they are combined.
Home Assistant becomes infrastructure, not the interface
Home Assistant sits underneath much of the setup. Door and environmental sensors feed into the same system. Different people around his mother do not need the same controls, so Benjamin built separate views for family members and caregivers.
That is a useful inversion of the usual smart-home demo. The goal is not to automate an object because automation is possible. Each piece starts from a friction that was actually observed.
Home Assistant’s own documentation supports this kind of structure: multiple users, multiple dashboards and different views for different devices or roles. It is a general-purpose home automation platform being shaped around a very specific human situation.
Failure states are part of the product
The strongest part of the project may be what happens when something breaks.
Benjamin describes routines that push devices back toward familiar states. The television returns to a predictable channel and volume. Remote access lets him fix parts of the system without being in the apartment. A mobile connection acts as a fallback when the main internet connection fails.
That is less glamorous than adding another sensor, but it is what turns a prototype into something another person can depend on. A real system needs a comprehensible failure mode, not just a successful demo path.
The lesson travels well outside home automation. Software used by someone who is tired, stressed, elderly, cognitively impaired or simply uninterested in its internal machinery benefits from fewer visible states, obvious exits and a known way back.
AI helps the builder, not the person being supported
Claude appears in the workflow, but not as a medical agent making decisions about Benjamin’s mother. He says he used it to help write Home Assistant dashboards and automations from natural-language requests.
That boundary is important. AI expands what one technically capable person can build. It does not need to become another interface imposed on the eventual user.
For small studios and independent developers, that may be the more consequential use of generative models: making tiny, highly specific pieces of software economical to create. Some of them may serve a family, a workshop or even a single person rather than a market large enough to justify a conventional product.
Personal systems also need personal limits
Sensors, alerts and remote information inside someone’s home inevitably raise questions about consent, privacy and security. The right balance depends on the individual, their family and the professionals involved. Benjamin’s video documents one family’s choices, not a universal template.
That is also why copying his shopping list would miss the point.
The transferable process is smaller: observe a real difficulty, remove an action, hide complexity, design the fallback, then watch what happens and iterate. In this project, the best technology is the technology his mother has to think about the least.