One System, Many Brands — Starting at Home

At 2057 Holdings, the operating principle is one system, many brands. Each company keeps its own identity; underneath, they share one platform, one set of standards, and one way of verifying work.

We recently applied the same principle somewhere more personal: a home.

The problem

A typical smart home is a portfolio without a parent. In this case: Cync and Tapo lighting, Leviton and Kasa switches, Samsung televisions, Apple HomePods, Sonos, and Aqara and Tapo locks. Each brand works on its own. None of them work together.

The system

Everything now runs through Home Assistant, on dedicated hardware inside the home:

• Lighting, switches, and locks on one touch-screen panel, regardless of manufacturer.
• Parental controls for the children's televisions: a nightly curfew, adjustable extra time, and a multi-night pause.
• Synchronized audio in three zones across Sonos, HomePods, and television speakers.
• An ambience video library stored on-site and delivered to any television over the local network — no media server and no subscription.

How it was built

The configuration was produced by describing requirements in plain language to Claude, Anthropic's model, which built each change, tested it, and reported what it had verified before it was accepted. The same discipline we apply across the portfolio — never assume, verify — applied here.

What comes next

The next layer is an AI interface: describing an outcome ("I want to watch a movie in the living room") rather than recalling device and routine names. Predefined routines remain for frequent and safety-critical actions, where speed and predictability matter. Anything new can simply be described.

Two boundaries are fixed. Actions that unlock a door or open a garage are refused by the control tool itself, regardless of the request. And the model can act only on the devices explicitly exposed to it; the home's control system itself runs on-site.

The full write-up is here: Describe It, Don't Name It.