I Used the Rabbit R2 for a Week — Here's How AI Agents Actually Work Today

I remember watching the Rabbit R1 launch in early 2024 and thinking, "This is either going to change everything or be a paperweight." The R1, for those who forgot, was a bright orange handheld device that promised to use AI to do tasks for you — book Ubers, order food, send messages — without needing to open apps. It was a flop. The reviews were brutal. The device was slow, the AI didn't work, and it basically became a useless toy within weeks.

So when Rabbit announced the R2 on June 9, 2026, I was skeptical. But I'm also curious. I ordered one ($249) and used it as my primary device for a week. Here's what actually happened.

What Is the Rabbit R2?

It's a small, handheld device — about the size of a deck of cards — with a 3.5-inch touchscreen, a scroll wheel, and a camera on the back. It runs a custom operating system called Rabbit OS, which is powered by a large action model (LAM) — basically an AI that can interact with apps on your behalf. The idea is that instead of opening DoorDash, browsing menus, and placing an order, you just say "Rabbit, order me a pepperoni pizza from Domino's" and it does it.

The R2 is the sequel. It has a better processor (Qualcomm Snapdragon 8 Gen 4), more RAM (8GB), and a new AI model trained on more tasks. The company claims it can now handle over 10,000 different actions across 500 apps. That's a big claim.

Setting It Up

Unboxing was nice. The device is cute, like a tiny Game Boy. You charge it via USB-C. Setup involved downloading the Rabbit app on my iPhone (yes, you still need a phone for initial setup), connecting to Wi-Fi, and logging into my accounts — Uber, DoorDash, Spotify, Google Calendar, Gmail, WhatsApp, and a bunch of others. You give Rabbit permission to act on your behalf. That's a big trust ask. I was nervous about it.

The setup took about 20 minutes. Then I was ready to test.

What Worked Surprising Well

Music and podcasts. I said, "Rabbit, play the latest episode of The Daily on Spotify." It took about 4 seconds, and then the podcast started playing through the speaker. The speaker is decent — not great, but fine for a tiny device. I could control volume with the scroll wheel. It was genuinely faster than pulling out my phone, unlocking it, opening Spotify, and searching. This became my default way to start music and podcasts.

Timers and reminders. "Rabbit, set a timer for 15 minutes." Done. "Remind me to call Mom at 6 PM." Done. The voice recognition is excellent. It understood me even when I mumbled. It also integrates with Google Calendar — I asked it to schedule a meeting for next Tuesday, and it created the event with the correct time and sent invites to the people I named. That's genuinely useful.

Translations. I was at a Korean restaurant last week, and I used the camera to translate the menu. It took a photo, OCR'd the text, and showed me the translation in real-time on the screen. It worked perfectly. Google Lens does this too, but the Rabbit was faster and didn't require me to open an app.

Ordering food. This was the big test. I said, "Rabbit, order a large pepperoni pizza from Domino's." It asked for my address (which it already had from my profile), confirmed the order, and placed it. I got a notification when it was on the way, and the pizza arrived. No app. No typing. Just talking. It felt like magic, even though I know it's just API calls and automation.

What Failed Miserably

Uber. I asked it to book an Uber to the airport. It took 30 seconds, then said it couldn't find a ride. I asked again. Same thing. I eventually just used my phone. Later, I checked the Rabbit app, and it had tried to book an Uber Black (which costs more) instead of UberX. There was no way to specify the type of ride. The AI doesn't understand context well enough yet.

Complex tasks. "Rabbit, find a Thai restaurant near me with good reviews and make a reservation for two at 7 PM." It found a restaurant — but it was 20 miles away. I said "closer," and it found another one — but it had 2.5 stars. I said "good reviews," and it got confused. After five minutes of back and forth, I gave up and used Yelp. The Rabbit is good at simple, single-step tasks. Anything with multiple parameters or subjective judgment is still beyond it.

Messaging. I asked it to send a WhatsApp message to my wife saying "I'll be home in 10 minutes." It did it. But it sent it to the wrong person — my friend Mike who I hadn't talked to in months. I had to apologize to Mike. The contact matching is buggy. Also, it can't read messages back to you unless you explicitly ask. There's no inbox. So if someone replies, you won't know unless you check your phone. That kind of defeats the purpose.

Privacy Concerns

This is the elephant in the room. Rabbit says all data is encrypted and they don't store your credentials — they use a token system. But you're still giving a third-party company direct access to your accounts. If Rabbit gets hacked, someone could order things on your behalf. The company has been transparent about their security practices, but I'm not entirely comfortable. I changed my passwords after the test week.

Who Is This For?

Honestly? It's for tech enthusiasts who love gadgets and have patience for imperfections. It's not ready for mainstream use. The failure rate on complex tasks is too high. The messaging issue is a dealbreaker for me. But for simple things like timers, music, and quick orders, it's actually faster than a phone.

I think the Rabbit R2 represents the direction we're heading — AI agents that do things for us instead of just answering questions. But we're not there yet. The R2 is a beta product sold as a finished one. Should you buy it? Only if you have $249 to burn and you like being an early adopter. Otherwise, wait for the R3. Or just use your phone. Your phone is still faster for most things.

I'm keeping mine, though. It's fun. And when it works, it really feels like the future.