When the Rabbit R1 arrived earlier this year, it was an unfinished product. Engadget’s own Devindra Hardawar called it “a toy that fails at almost everything.” Most of the features Rabbit promised, including its signature “large action model” (LAM), were either missing at launch or didn’t work as promised. Now, after more 20 software updates since the spring, Rabbit is releasing its most substantial update yet. Starting today, every R1 user now has beta access to teach mode, a feature that allows you to train Rabbit’s AI model to automate tasks for you on any website you can visit from your computer.
Rabbit CEO and founder Jesse Lyu gave me a demo of teach mode ahead of today’s announcement. The tool is accessible through the company’s Rabbithole hub, and features a relatively simple interface for programming automations. Once logged into your Rabbit account, you navigate to a website and input your credentials if they’re required to access the service you want to teach the R1 to use for you. Lyu was quick to note Rabbit won’t store any username and password you input; instead, the company saves the cookie from your teach mode session for the R1 to use later. In June, Rabbit had to move quickly to patch a security issue that could have led to a serious data breach.
Once you’ve named your automation and written a description for it, all you need to do is carry out the task you want to automate as you usually would. Rabbit’s software will translate each click and interaction into instructions the R1 can later carry out on its own. When Lyu demoed teach mode for me, he taught his R1 to tweet for him.
Once the software has had a chance to analyze a lesson, you can replay the automation before trying it out on your R1 to ensure it works properly. While it’s technically true you don’t need any coding knowledge to use teach mode, approaching it from a programming perspective is likely to produce better results. That’s because you can annotate the steps the software records you doing when showing it an automation. It’s also useful from a troubleshooting perspective, as you can see from the video embedded above.
After you’ve tested your automation, it’s just a matter of asking your R1 to complete a query using teach mode. The resulting process isn’t exactly the polished experience I imagine most people have come to expect from their mobile devices. The R1 announces each step of a task, and it can take a few moments for the device to work its way through a query. According to Rabbit, part of that is by design. Early testers found it helpful for the R1 to state its progress.
I’ll be honest, it’s hard to escape the conclusion that some of the R1 automations Lyu showed me, while creative, don’t offer a more efficient way to do certain tasks than the apps people are already familiar with, a point he conceded when I said as much during our call.
“There are a lot of tasks that are not a single destination,” Lyu said. To that point, where he believes teach mode will be transformational is in interactions involving multiple platforms. Lyu gave an example of an R1 user who taught his device to order groceries. With some work, that person could use the R1’s camera to take photos of the shopping lists his wife produced, which the device would then use to order the family’s weekly groceries from their preferred stores.
Another area where the R1 could provide a better experience than a dedicated app is in situations where there are competing standards, like the situation that exists with smart home automation currently. Say you’re trying to get some HomeKit and Google Home devices to work together. You won’t need to wait for the Matter Alliance to sort things out. With teach mode, the R1 will navigate that mess for you.
“You need to think about velocity,” Lyu tells me before laying out Rabbit’s end game with teach mode. For now, R1 users can freely add community lessons they find on Rabbithole to their devices. Lyu envisions a future where users will be able to sell their automations, with Rabbit taking a cut. Moreover, while teach mode is currently limited to navigating websites, Lyu suggests it will eventually learn to use more complex apps like Excel. At that point, Lyu contends Rabbit will be in a position to deliver an artificial general intelligence, one that will understand every piece of software ever made for humans.
Of course, questions remain. One major one is whether people will pay for community lessons if they could just as easily replicate an automation on their own. Here, Rabbit expects things to play out like they’ve done on existing app stores, with most people choosing to download apps they like instead of making their own. “For the future agent store, we anticipate a similar situation where any user could teach their own lesson if they want to, but most people will probably find lessons or agents created by other users that meet their needs very well,” the company told me in an email.
I also asked Rabbit if the company is preparing for the possibility that some platforms might block people from using teach mode to automate tasks on their R1. In the company’s view, bot detection systems like CAPTCHA will need to evolve to differentiate between “good agents” like those created by Rabbit users and malicious bots.
“When a user uses LAM to perform tasks on third-party platforms, they are logging into their own accounts with their own credentials, and paying those companies directly for those subscriptions or services,” the company added. “We are just providing a new platform for those transactions to happen, similar to you can play music on your phone and on your laptop… We do not see a conflict of interests here.”
I’m not so sure if things will play out as smoothly as Rabbit hopes, but what is clear is that the company is closer to the future Lyu promised at the start of the year — even if that future still feels years away and may be decided by another company. For now, Rabbit hopes R1 users embrace teach mode enthusiastically, as that will allow the software to improve more quickly.
