Your House Just Got More Than One Brain.

That Is the Point.

Google Home is opening smart devices to ChatGPT, Claude, and other AI agents. The product news is convenience. The design news is optionality.

Most homes, like most companies, were built as single-vendor systems. One assistant. One cloud. One app that has to be up for the lights, the thermostat, and the lock to feel “smart.” That is efficient on a calm Tuesday. It is fragile the first time the account locks, the model is rate-limited, or the region goes dark.

Letting more than one agent talk to the same house is the consumer version of the stack we should already be building at work. Do not marry one model. Give the environment a standard door, then choose which intelligence walks through it.

That is the same three-tier logic in miniature. (Note for old friends - I always seem to end up at three tiers)

Cloud frontier models for the hard request: plan the week, diagnose the odd bill, draft the messy email.

A local or private floor for the work you cannot park on a public status page: records, agents that run all day, data you actually own.

The device and the house for continuity. Lights, drafts, summaries, voice, screen context. The layer that still works when the interesting model does not.

The mistake is to treat “open to many agents” as a feature list. It is a stress test. If one vendor can turn off your operating environment, you do not have an assistant. You have a single point of failure with a friendly voice. Anti-fragile systems do not bet that the preferred model stays cheap, available, and well-behaved. They keep a second path wired in before they need it. The companies that copy that pattern—multiple models, a private compute floor, and devices that can still think—will keep moving when the first logo goes gray.

The ones that standardize on a single brain will discover they automated the house and forgot to keep a key.