In the winter of 2024, the Rabbit R1 looked like the future, a small, vivid orange handheld device that promised to make the smartphone obsolete. It was one of the first serious attempts to build dedicated AI hardware for a world just waking up to generative AI, and it sold out quickly enough to feel like a movement. You could hold it to your ear, speak to it conversationally, and ask it to book a car, order dinner, or settle a question without ever touching a glass screen. The launch at CES was pure buzz, with the design by the celebrated Swedish firm Teenage Engineering giving the gadget an almost cult-like appeal. Then the reviews came in, and they were brutal. WIRED gave it a 3 out of 10. Critics found that the “agentic” features, those clever autonomous tasks the device was supposed to run in the background, simply didn’t work reliably. A gadget that had been treated as a herald of a post-app world became a cautionary tale. But Jesse Lyu, the founder and CEO of Rabbit, insists that the story is more complicated than people think. To him, the R1 was not a flop, and the past year and a half has only made him more confident in the bet the company made. In his telling, the R1 was an early, imperfect messenger for an idea that was always going to arrive; it just seems more obvious now because the entire industry is chasing the exact same thing.
That shift in mood is hard to overstate. In the world of artificial intelligence, “agents” are now the talk of every boardroom, every lab, and every tech conference. These are not simple chatbots that answer queries; they are programs that take action on your behalf, navigating websites, comparing prices, making purchases, and orchestrating complex tasks with minimal supervision. From one angle, this is exactly what Rabbit claimed to be doing long before the term was fashionable. Lyu cannot help but smirk when he thinks about it. When news broke that Meta’s new Muse agent had been banned from Amazon for trying to shop for a user through a virtual machine, he recognized the scene immediately. The Rabbit R1 had used a virtual machine to execute app actions for users, and that decision had drawn criticism from people who thought it was a gimmick or a shortcut. Now, that same approach is mainstream, or at least common enough that the world’s biggest tech companies are running into its practical limits. Lyu says, “We’ve been through all of this a year and a half ago.” The confirmation of a trend does not by itself prove that Rabbit executed it well, but it does complicate the idea that the R1 was a silly experiment. The technological weather has changed, and Rabbit, once an isolated oddity, now looks like a small oracular company standing in the middle of a storm it predicted.
When Lyu talks about the R1, he is careful to separate two kinds of success. The first is financial, and by that measure, he believes the device was a triumph. Rabbit delivered more than 100,000 of those orange-red devices around the world, many of them to people who were not just curious but genuinely excited. The company kept return rates below 5 percent, which is striking for a category where die-hard early adopters are often the quickest to demand refunds. There is a reason for that. Unlike most AI companies, Rabbit never forced its customers into a subscription. Once you owned an R1, you owned the experience, and the company shouldered the cost of the underlying models and infrastructure from its own margins. Lyu estimates that the hardware itself carried a profit margin of roughly 45 to 50 percent, a number that sounds almost implausibly healthy for a startup in a field where hardware is notoriously unforgiving. He also believes, with the confidence of a founder who has been questioned many times, that the engineering bet was right. Internally, he says, “we didn’t make the wrong bet.” The challenge, in his mind, was never the fundamental architecture or the ambition. It was simply the material reality of competition. Rabbit was operating with a fraction of the resources available to the AI giants, which have raised billions of dollars and employ armies of researchers. Lyu says Rabbit has raised around $60 million in total and currently has roughly fifteen employees. That is a deliberately tiny operation, and it means that every missed step, every critical review, and every unexpected delay carries enormous weight.
Around this time last year, things were genuinely dire. The bruising reviews were only the beginning. Rabbit had planned to launch in India, a market that could have given the company a massive infusion of revenue and validation. But regulation got in the way, and the launch was blocked. The company found itself financially strained, so much so that a handful of its employees went on strike, telling the world they were not being paid. That is the kind of crisis that usually ends a startup. Lyu, however, speaks about it with a certain calm, almost as if he had already prepared himself for the harsh cycle of expectation and disappointment. Rather than turning bitter, he says he respected the decision of those employees, understanding that people have families, obligations, and their own limits. The company did manage to secure more funding and has, by his account, stabilized. “There are no issues with the company’s health,” he says, a short declaration that covers a long and difficult period of scrambling. It is worth pausing to imagine how many conversations, sleepless nights, and late night negotiation sessions live behind that single sentence. The story of Rabbit is not just the story of a product review or a viral moment. It is also a story about what it means to keep a company alive when the world has already moved on, when the headlines have faded, and when most people have stopped paying attention. For Lyu, survival was not about making a dramatic comeback or chasing another viral launch. It was about quietly rebuilding the foundation.
The company’s response has been deliberate, and in many ways, it has been the opposite of the flashy hardware strategy that marked its debut. Lyu and his team decided not to rush out a second-generation device. Instead, they poured their energy into software updates, polishing the R1 and making it more stable, more useful, and less frustrating. In the past year and a half, Rabbit has released around fifty updates to the device, a steady rhythm of iterative improvement that does not generate headlines but slowly changes the experience. The company also built a dedicated Discord community of more than 12,000 users who act as an extended design team, sharing feedback and suggesting new features. That closeness has shaped the product in tangible ways. If you own an R1 today, you can integrate OpenClaw or Hermes AI agents into it, giving the little orange device capabilities far beyond what it shipped with. It is no longer locked down to Rabbit’s own proprietary systems; it has become a small, hackable companion for people who enjoy tinkering. Now, Rabbit is taking the next step. Earlier this week, the company announced OS3, a standalone “agentic operating system” designed to live across multiple screens. It still arrives as a software update for the R1, but you no longer need a piece of Rabbit hardware to use it. Instead, you can access OS3 from a desktop browser, from Telegram on your phone, or even from iMessage. This is a striking pivot for a company that once insisted that the future was a dedicated device in your hand, not an app on the phone you already carry. But it also shows a kind of maturity. Rabbit is not abandoning its vision; it is finding a more practical way to let people experience it.
In the end, the question of whether the Rabbit R1 was a flop depends entirely on your definition of the word. If a flop is a product that failed to win over critics or change the world in its first year, then yes, the R1 had a difficult beginning. If a flop means a company that made a costly mistake and has nothing to show for it, then the story is very different. Rabbit survived a near-death experience, kept its hardware afloat, earned sustainable margins, and now has a seat at the table just as everyone else is starting to chase the same dream. The lessons are not simple. The R1 was too early, under-resourced, and burdened by expectations it could not meet at the moment of its birth. But it was also a real attempt to answer a question that still haunts the industry: how can AI move beyond the chat window and actually do things in the world? Lyu’s tone is not cocky. He admits to the constraints, the hard choices, and the humility that comes from watching a promising product get torn apart in public. Yet he seems genuinely comfortable with the journey. He knows that technology moves in strange arcs, that ideas need time to ripen, and that a small team with limited money can sometimes see the future clearly even if it cannot build it perfectly the first time. The R1 may never become the artifact that history remembers as the turning point. But it was a signal, and now the signal has become a chorus. Whether Rabbit can build on its lead, make OS3 work across all those surfaces, and continue to evolve beyond its early promise is still an open question. But Lyu has already achieved something rare in the churning world of AI hardware: he kept his company alive long enough to watch the world come around to his point of view. That is not a failure. It might be the most human kind of triumph there is: to be right, to be early, and to still be standing when the future finally catches up.
