Google began removing Google Assistant on 4 September, a decade after launch. Gemini takes over the “Hey Google” trigger and the power-button long press. The rollout is staged over several weeks and covers Android phones and tablets, Wear OS watches, headphones and earbuds, and vehicles running phone-projected Android Auto. Cars with Google built directly into the dashboard keep Assistant for now.
Most of the write-ups have framed this as a product upgrade with a nostalgia note attached. It is more usefully read as the largest forced migration in consumer software this year, and the mechanism is worth looking at closely, because it is the one every platform reaches for eventually.
Until your particular device is switched, you can move between Assistant and Gemini freely. After it switches, you cannot go back. That is not a bug in the rollout; it is the rollout. A reversible migration produces a slow drift and a long tail of stragglers you support forever. A one-way door produces a step change in your usage numbers on a date you control.
The number this is really about
Every assistant-class product is judged on engaged users, and every AI product in 2026 is judged on that number harder than on revenue. Assistant sat on an enormous installed base built up over ten years of default placement. Gemini has had to be chosen. Retiring Assistant converts a decade of accumulated default placement into a Gemini number, without Gemini having to win a single one of those users on merit.
This is the oldest lesson in platform economics and it keeps needing to be relearned: distribution beats product, and default placement is the purest form of distribution there is. The interesting question is not whether Gemini is better than Assistant. It is that the question stopped being asked the moment the toggle disappeared.
Our take: If your business touches Assistant — voice actions, routines, smart-home integrations, a skill your customers trigger by voice — you have just had a distribution channel retired on a schedule you did not set, and the migration is one-way for your users too. The general rule: any capability you access through somebody else’s default is a lease, not an asset, and the lease has no notice period you can enforce. Price it accordingly. If a channel you don’t control drives more than a fifth of your usage, the mitigation work is due before the deprecation notice, not after.
The cost side nobody mentions
There is a reason platforms run assistants for a decade and then stop. Assistant was, architecturally, a cheap intent-matcher: parse the utterance, route it to a handler, return a canned response. Gemini answers the same query with a large model. The unit cost of “set a timer” goes up by orders of magnitude when the thing answering it is a frontier model rather than a lookup.
That is why the transition has taken this long and why cars with embedded Google are excluded for now — the hardware and connectivity assumptions in a 2019 dashboard were built for the cheap version. Expect the same pattern everywhere: the migration lands first where the compute can be pushed to a recent phone or to the cloud on a fast connection, and last where the installed hardware cannot carry it.
What to watch
- Complaint volume on regressions. Forced migrations get judged on the boring stuff — timers, alarms, smart-home commands, offline behaviour. If Gemini is slower on the ten things people do daily, the goodwill cost is real even though the usage number still goes up.
- The embedded-car carve-out. “At least for now” is a dated promise. Watch for a second wave.
- Whether Google publishes the migrated numbers. Reporting Gemini engagement without separating migrated Assistant users from acquired ones would tell you exactly how much of the growth story this was carrying.
- Third-party voice integrations. Who gets ported, who gets dropped, and how much notice they got.
Ten years of default placement just changed logos. Nobody had to be persuaded, which is the point.