Rabbit, the company behind one of the most talked-about AI gadgets of the last few years, is cutting its own hardware out of the loop. According to The Verge, Rabbit's new AI agent does not need the R1 to run.
That matters because Rabbit originally argued the exact opposite case. Its pitch was a dedicated little device that performed tasks inside apps on your behalf, driven by a large action model rather than by you tapping through screens yourself. The R1 shipped in 2024 to loud attention and equally loud criticism, and the "hardware first, capability later" pattern has a poor track record in this category.
Software agents have a structural advantage. They do not need to be manufactured, shipped and charged, and they can live where people already work — the phone, the browser, the desktop. Their reach grows with each update instead of with each supply chain. What they lose is the neat packaging of a device: an agent that operates apps on your behalf needs durable access to accounts, permissions and sessions, and that is exactly where the question of useful versus risky gets decided.
That is also the test Rabbit now faces. An agent that does not require its own hardware still has to finish tasks reliably — bookings, forms, purchases — and still has to explain what it did. Adoption will hinge less on model quality than on trust, error handling, and who is accountable when a step goes wrong.
What to watch: which platforms OS3 actually ships on, whether existing R1 owners benefit or get left behind, and whether security researchers take apart the agent's permission model. Details of platform support, pricing and feature scope were not included in the reporting available at the time of writing.
