I’ve never been as charmed and frustrated by one gadget as I have with the Poetry Camera.

It’s a delightful object. White and cherry red with a color-matched woven strap, it looks playful and adorably lo-fi. If I saw it on a store shelf, I’d absolutely pick it up. 

But aside from obviously appealing, I’m not exactly sure what it is. I mean, I know what it is. It’s a camera that makes AI poems instead of photos. You take a picture, and instead of printing a photo, you get an AI-generated poem inspired by the scene, printed on thermal receipt paper. But after printing dozens of poems, I can only report feeling frustrated instead of inspired.

There’s no screen on the camera itself, just a shutter button and a dial that lets you pick a different poem style. It only works when connected to a Wi-Fi network, relaying your image and a prompt tied to the camera setting you’ve chosen to the cloud. About 30 seconds later, the printer spits out a poem. Tear it off like you would a grocery store receipt, read it to your friends/spouse/cat, rinse and repeat. The poems themselves all sound a bit like this one, inspired by a picture I took in my kitchen:

Fingers curve the mug-
white cabinets hold their
secret:
another April

The Poetry Camera is the product of a collaboration between Kelin Carolyn Zhang, an ex-Twitter designer, and Ryan Mather, an ex-Googler. They brought the concept to life through painstaking iteration, taking it from a wacky idea to a cardboard prototype to a functional product. They gave a thoughtful presentation at Figma’s annual conference last year about the highs and lows of their collaborative relationship; later in 2025, they parted ways. Zhang oversaw production of Poetry Camera’s Batch 2, assembled at a factory in Shenzhen as part of a residency with MIT rather than manually with the help of friends in New York. The second round of cameras went on sale for half its original price: $349 rather than $699. That batch is sold out; a third batch is promised for May.

The mechanics of the Poetry Camera are nifty. How do you get a gadget without a screen or a mobile app connected to Wi-Fi? You use Poetry Camera’s simple web app to generate a QR code. Point the camera at the code and it’ll link up automatically. Clever. There’s an LED around the shutter that communicates connection status or problems, and the printer also spits out a message to let you know when it’s online. There’s something about a gadget communicating with its user with a physical, printed message that’s sort of cute.

You can also access a portal for your particular camera where you can customize the prompts for each poem setting. That got me really interested. Poetry is great and all, but the sonnets and haikus about the line of shoes in my entry got old pretty fast. 

Rewriting the prompts sounded fun. I learned that you have to actively instruct it not to write a poem, even with an entirely new prompt that doesn’t mention poetry. But once I’d done that, I successfully created a mode that prints an appropriate quote from Jurassic Park based on what it identifies in a scene. Another mode describes the current weather conditions when I take a picture out the window and gives me a forecast for the day. But not all of my prompts worked, and the trial-and-error process of figuring out why became tedious. 

The camera puts itself to sleep after a couple of minutes, and when it does, you need to start it up again and wait for it to reconnect to the network. When it fails, the camera prints out one of a handful of error messages, styled as a poem. This was cute the first time it happened, but it wore thin after a half-dozen attempts. It also means you don’t know exactly what the problem was — did my prompt hit some guardrails? Was I standing too far from the Wi-Fi router? Related: I couldn’t get the camera to connect to my iPhone’s hotspot no matter what I tried, so my experimentation was bound to the house.

I have no doubt that Poetry Camera is the product of talented, dedicated minds. But it feels to me like an artifact of AI as we knew it years ago when we were all first delighted by ChatGPT — when an LLM writing something that looks like a poem was a novelty and we were all a little less weary of chatbots.  

Call me old-fashioned, but I think the value in an art form like poetry is directly tied to the humanity of its creator. I tried to set this aside and accept the Poetry Camera free of bias, but maybe I was just never going to have a good time with it. The Poetry Camera puts together words that kind of sound deep and meaningful on the surface, but also feel soulless and read like empty calories. AI can be a powerful tool in making software, but writing meaningful poetry requires, at a bare minimum, a soul. A computer doesn’t have one of those, no matter what venture capitalists may say otherwise. 

I’m still not sure what Poetry Camera is, but  I do know one thing: it’s not for me.

Photography by Allison Johnson / The Verge

Upgrading your phone with a camera grip attachment is one thing. But doll-sized telephoto lenses that you mount on top of the rear camera? C’mon.

I wrote the Vivo X300 off as a gimmick, a funny concept designed to generate attention rather than actual sales. But then I spent a weekend carrying the phone and its elaborate kit of goofy little lenses around — and I had way too much fun.

The Vivo X300 Ultra is an update to one of the very best phone cameras out there. It’s only available in China at the moment, with a global launch that will almost certainly exclude the US. Its rear cameras are no joke: a 200-megapixel main, 200-megapixel 3.7x telephoto, and a 50-megapixel ultrawide. Vivo sells a pro camera grip separately to add a physical shutter button. And if you want to really go for it there are those two telephoto extender lenses, each of which mounts to a special plate putting it in front of the telephoto camera. This year’s edition features a more compact version of last year’s 2.35x adapter, adding up to a 200mm equivalent, as well as a new 400mm option. 

If you’re going to carry a phone like the X300 Ultra, you might as well go all in. I stopped short of using the dedicated bag that Vivo sent alongside my review unit — partially because it smells weird, but also because I needed something bigger for a full day out at the spring fair. But I attached the special camera kit case to the phone, along with the pro camera grip, the telephoto lens adapter plate, and a crossbody strap ripped straight out of Apple’s fall 2025 catalog. I stepped out of the family SUV with 150mg of non-drowsy Allegra coursing through my veins, camera slung across my chest and a suitably ergonomic backpack on my shoulders. Peak parent mode.

I took a few shots around the midway without the attachment lenses, but once we headed for the kiddie roller coaster, I knew it was time. I popped the 200mm extender on and didn’t take it off for at least another three hours. Did I look like a weirdo? Yeah. But I was having too much fun to care. 

It took a minute to wrap my head around using a 200mm lens on a smartphone camera. I remembered that I had to care about shutter speed. But once I got the hang of it I couldn’t believe what I was getting away with. Depth! Layers! An optical telephoto lens lets you play with composition in a way that never quite translates with digital zoom. I was taking shots I’d never even attempt with another phone camera, and it was mostly working out. 

I figured I’d put the extender lens on the phone, take a few photos, and then remove it until I needed it again, but that’s enough of a hassle that I just left it alone. Anyway, once I started realizing all of the potential for telephoto photos, that’s all I wanted to shoot.

It did mean that the lens kept banging against the carnival ride seats every time I bent over to buckle my kid into a sparkling rocket ship or whatever, but this didn’t seem to do any damage to the camera equipment. Guess I have that sturdy Zeiss construction to thank.

A kid on a roller coaster coming straight toward the camera was still tricky for the autofocus system, and I missed plenty of shots. But that would be a challenging scenario for a camera of any size, and once I shifted to a more forgiving composition it started to click. And it took me a minute to realize that half-pressing the camera grip’s shutter button was only locking focus and exposure, not acquiring focus — because it is a smartphone, the camera is just doing that all the time on its own. But once I got comfortable mashing the shutter button all the way and trusting that auto-focus was working, I started really having some fun. This was also around the time that the demolition derby started.

On our way to our seats we passed a sign prohibiting professional photo and video capture, including cameras with “removable lenses.” I had the X300 Ultra over my shoulder with its 200mm extender attached, but nobody batted an eye. I’m not usually one to use manual exposure controls on a phone, but I definitely needed them as the cars started crashing and the early evening light faded. 

I switched from the 200mm to the 400mm, and back again when the light became too dim. From my spot about halfway up the stands, 200mm wasn’t quite long enough and 400mm was a little too close. There’s the option to use digital zoom on top of the teleconverter’s optical magnification, but I opted to stick with native focal lengths as a matter of principle. 

I managed to get some shots I really like with each of the extenders — plenty of detail on the cars that we couldn’t see from the stands, the focus on the drivers’ faces as they chased down another car. Side note: Cars crashing into each other rules. I kept shooting until the light was too dim and detail really started to suffer. The antihistamines had worn off and the preschoolers were restless, so we called it a night.

Whenever I get my hands on a telephoto lens I remember how much fun it is. Having that experience while using a smartphone was totally unexpected. There’s nothing stopping me from renting a telephoto lens for my Sony a7c to bring to the fair, and I probably would have gotten more of those roller coaster photos in focus.

But it would likely have caught security’s attention at the demo derby, and definitely would have put more of a strain on my aching back and shoulders with the added weight. The 200mm lens attachment is small and light enough I would barely notice its presence in my purse, and I could definitely see taking it along on another family outing.

I’m frankly surprised at how fast I went from dismissing the whole extender lens concept as a silly idea to wanting the big phone brands that sell in the US to adopt something similar, like, yesterday. Where is Samsung on this? Why can’t I buy a funny little lens for the Galaxy S26 Ultra? Consider me an extender lens convert.

Photography by Allison Johnson / The Verge

I saw a lot of weird phones at Mobile World Congress last month: robot phones, cameras disguised as phones, phones for dogs. But the one that caught me most off guard was the one my friend (and Verge alum) Sam Byford brought to dinner: an iPhone Air.

“Ha!” I said. “You actually use that thing?” 

“Yeah,” he said. “It’s great.” That’s when he pulled out a second phone — the Xiaomi Leica Leitzphone, which he was using alongside it. 

Got it, I said. The trick to enjoying the iPhone Air is to simply have another, much better phone on your person at all times. I found this very funny and recounted the anecdote to anyone who would listen. But friends, I come to you with a confession. I think Sam is right: The iPhone Air might actually be good.

In fairness, I never really thought it was bad bad. I gave it a 7 when I reviewed it last year. The benefits of its slim profile and light weight are easy to understand, and it really does leave an impression when you hold it for yourself. But a phone with worse battery life, a single rear camera, and one measly speaker? For the same price as a regular phone? I kind of wrote it off as a weird thing that happened on the way to the Apple folding phone.

I picked up my iPhone Air review unit again after MWC when I needed to compare its camera to the 17E’s. After the 17E review, I swapped my eSIM to the Air to see if using it for a longer period of time told me anything new. I figured I’d get tired of it after a week or so, but I didn’t. I kinda got hooked.

It helps that I’m not doing any traveling right now, and I’m very rarely away from a charger for an extended period of time. My soft-pants, remote-work lifestyle is very forgiving to a phone with weak battery life. Even so, it lasted through a lengthy excursion out of the house last week. I definitely plugged it in for a recharge when I got back home, since battery percentage was hovering just above 20 percent and I’m allergic to low power mode. 

It’s no accident at all that when Apple loaned me the Air to review it also sent the super-slim MagSafe battery pack to go with it. I’ve been putting it to use while I revisit the Air, and not only as an insurance policy when I go out; sometimes it’s nice to have so I can top off the battery while I’m using it around the house without having to tether myself to a wall charger. That got me thinking: Maybe MagSafe is the thing here.

One of my own personal hang-ups with the Air is its screen size. It’s a slim phone, not a small phone, and I still find it a pain to use one-handed. But what if I just pop on a magnetic ring grip when I’m settling in to do some real scrolling? I started leaving one on a side table next to the couch to have it handy for such occasions. When I’m done it comes off the phone; the Air stays just as light and pocketable as ever.

I also started using a MagSafe wallet with the Air. I like the idea of a wallet on the back of my phone, but I usually find them too bulky, even without a case on whatever phone I’m using. But since the Air is so slim to start with, I don’t mind the added bulk of the wallet. It’s a pretty handy way to carry a couple of essential cards when I’m running out of the house for a quick errand, and when I’m back home I can just plop it onto the key tray. 

You could even use MagSafe as a solution for one of the Air’s other weak points: the crummy speaker. I mean, AirPods exist, for starters, which is usually good enough for me. But if you really wanted to go in on the modularity aspect, you could pick up a MagSafe Bluetooth speaker to thwack onto the back. Suddenly, this isn’t just the weird skinny iPhone with the bad battery anymore. It’s a modular phone.

I think the toughest thing to overcome on the iPhone Air is its camera. Unless Apple plans to release a Vivo-style camera lens system that can transform the Air’s rear camera into an actual telephoto when you need it, you’re stuck with a basically fine wide-angle lens and a little 2x crop zoom in a pinch. But more people these days are fine carrying a point-and-shoot camera in addition to their phone anyway. So if you were never very fond of the iPhone camera look, maybe the Air can be your excuse to bring a dedicated camera more often. Personally, I find the built-in camera options good enough that I just put up with it. But I’d be seriously tempted to try adding an ultrawide adapter if I was going to extend this experiment any longer.

I still think that the iPhone Air is impractical for most people. You know, people who don’t carry two phones, or get excited when they hear the words “Moto Mods.” The battery is only going to get worse over time, and maybe the modularity theory would start to get old after a year of managing a small stable of accessories. I can see the novelty wearing off. As for me, my time with the Air is over for now and I’m back on the 17 Pro. Is it nice having three actual rear cameras to choose from? And not worrying about the battery lasting a full day? It is. But I miss the iPhone Air more than I thought I would. And if it does end up being a weird side quest on the way to a folding iPhone, then at least it was an entertaining one.

The Selfix phone case is a classic example of a great idea in theory that falls apart in practice.

It’s a case for the iPhone 17 Pro with a circular screen on the back. It mirrors the middle of your main phone screen, making it easier to frame up selfies using your phone’s higher-quality rear cameras rather than the selfie camera. A nice concept! I wish it worked better.

Selfie cameras have come a long way, and the updated sensor in the iPhone 17 series’ front-facing camera is more than capable in good lighting. But there are still ways that the bigger sensors and lenses in your phone’s rear cameras outperform them: They can do slow-motion video, better low-light images, and more convincing portrait mode photos. The Selfix case makes all of that possible — if you can put up with it for long enough.

Dockcase Selfix

The Selfix is a case for the iPhone 17 Pro and Pro Max that includes a built-in screen for rear-camera selfies. It’s a nice idea that doesn’t quite work in practice.

Where to Buy:

• $79 at Kickstarter

The Selfix case is on preorder now through Kickstarter, with a full retail price of $129 and pricing for backers listed at $79. But I have some real hesitations recommending it even at the discounted early bird rate. For starters, it’s bulky. I’m working very hard here to overcome my biases: I don’t usually use cases, and I’m fresh from a couple weeks of using the iPhone Air. This case on the iPhone 17 Pro was always going to feel unwieldy to me. Still: I think it is unreasonably big. You do get a built-in microSD card slot, which is nice. It meets the UHS-I, Class 10, U3, and V30 standards, meaning that you should be able to record 4K video directly to the card without dropping frames provided that you have a fast enough card.

My biggest complaint with the Selfix is that it just doesn’t do the job very well. The screen is a circle with a 1.6-inch diameter, so it naturally cuts off a big chunk of the rectangular preview image on your main phone screen. It also has some real chunky bezels, so the usable screen area is a little smaller than the inside of a paper towel tube. All of that means there’s still some guesswork involved; if you frame your face in the center of the rear screen you’ll get a selfie with a lot of blank space above your head.

The Selfix’s screen is touch-sensitive, but I couldn’t get tap-to-focus to work consistently. The unit I tested is labeled as “Beta,” which may be the culprit here. You need to use the volume keys on your iPhone to take a photo or start video recording anyway, because the part of the screen with the virtual shutter button is cut off. It’s also kind of unsettling shooting video and not seeing any indication that recording has started. For something that’s designed to reduce the guesswork of using the rear cameras, there’s an awful lot of guesswork still involved.

Then there are the side effects. The case allows for passthrough wired charging, but no data connections, meaning no wired CarPlay or transferring files to your laptop over USB. There’s a magnetic ring in the back of the case so you can keep using MagSafe accessories, but no wireless charging. You need to take the case off to do any of these things. 

On top of all that, it’s pretty easy to forget the rear screen is turned on when you’re done taking your selfies. As a reminder: It mirrors whatever is on your phone screen, so you could wind up giving the rest of the world a clear view of your banking app or whatever you’re looking at. There’s a dedicated button on the side of the case to turn the screen on and off, and when it turns on you’ll see a little USB plug icon in the Dynamic Island to let you know an accessory has been plugged in. But there’s no persistent onscreen reminder that the case is mirroring the display, and I can see that going wrong in some unfortunate ways.

Look, I love a weird phone accessory. And I appreciate the Selfix case trying to do more than one thing — help you use the rear cameras for selfies, expand your phone storage, and protect your device. I just wish it did its main job better.

Photography by Allison Johnson / The Verge

You may be familiar with Gemini as the thing that’s in every Google service you use — whether you want it or not.

While it’s been a constant, sometimes unwelcome presence in Gmail for at least the past year, it’s a relatively new addition to Maps. And you know what? It’s kind of great.

To put it to the test, I had Gemini plan a day-long itinerary for me around the city. After an hour or so of having Gemini find stuff for me — playgrounds near the new light rail extension, kid-friendly restaurants with vehicle themes, you get the gist — I was impressed. Some of the suggestions were obvious, but I also bookmarked a handful of spots not on my radar.

Gemini had big shoes to fill: my own. I’m a Google Maps fiend. I use it for normal stuff like getting around, but also, sometimes I just like scrolling around and seeing if anything new catches my eye. You can find some real gems this way. I’ve found bike routes, playgrounds, hidden parks, and new coffee shops to try. I’d spend all day every day wandering around Seattle on public transit going to bookstores and fancy stationery shops if I could. It’s how I usually spend a day off, but I also tend to get overwhelmed by the endless possibilities and just wind up visiting one of a couple of neighborhoods I know well. So I had Gemini chart me a path into less familiar territory.

Gemini pops up as “Ask Maps” and presents you with a text box when you tap on it. It answers questions based on data in Google Maps, including user reviews, but can pull in information from other sources, too. If you ask whether to bring an umbrella on your trip across town, it can check the weather for you — that kind of thing.

I gave it my parameters: I would be traveling by public transit and I wanted a stop for lunch, a nice walk somewhere, and a coffee shop where I could work on my laptop, in that order. I wanted to visit two different neighborhoods and needed to be home by 4:30. Its first suggestions were very me-coded — a cafe next to a bookstore and a reliable coffee shop downtown — but I’d been to both recently. After a little back-and-forth it was settled: tacos, plants, and a Scandinavian-inspired coffee shop.

Tacos Chukis is familiar to me but I’d never been. I almost walked by the place, since it’s tucked into the back of a building with a half-dozen other retail shops and no sign on the sidewalk. But Gemini steered me to the right spot, and right on time: It had only opened up 15 minutes before I walked in. My AI itinerary hinted that the house specialty with grilled pineapple was a popular choice, and I found out why. Three excellent tacos later, it was time to head to my next stop.

Except I was ahead of schedule, so asked Gemini to find a unique shop nearby I could check out before walking north to the park. It confidently recommended Elliott Bay Books — a great spot, but definitely not “one block east” as it claimed. This was the only major hallucination I encountered in this experiment, but it could have been a real pain if I’d followed its instructions. Did I mention it was pouring rain outside?

After I politely informed Gemini that it was suggesting I walk 10 minutes in the wrong direction, it corrected course and sent me to Kobo: a beautiful little store with Japanese goods. I’d been to their other location a few times but didn’t realize there was one nearby. 

By the time I set foot in Volunteer Park, the front of my jacket was soaked. My umbrella was doing the heavy lifting, but I needed it to cover my backpack with laptop inside. Hence, the sog. Gemini had suggested a scenic loop around the park or a trip into the conservatory — basically a giant greenhouse — if I wanted to dry off. No contest.

Plants fucking rule, man. Did you know there’s a tree that hollows itself out to attract ants to live in it? And the ants fight off would-be invaders to protect the tree? That’s wild. That’s also a tree that exists at the Volunteer Park Conservatory, a building I’d seen but never been inside. Admission was $6 and I was a little grumpy with Gemini for failing to mention that, but it proved to be a small price to pay for time inside a warm, tranquil oasis on a rainy day. 

Someone who works at the conservatory saw me marveling at one of the towering palms and snapping photos with my phone. She took me over to the cacti room and insisted on taking my photo with the giant cacti there. “It’s so beautiful in here I could cry!” she said as she left me among the cacti. I had to agree. There’s something ghostly about a cactus, and they come in so many different shapes and sizes, down to the classic saguaro that I know mostly from cartoons. There’s the ones covered in fluff, the big round ones that look like the world’s worst footstools, and another one that looks like it’s covered in peeling waxpaper. They’re sort of haunting and mystical up close, like seeing an owl in the wild. Not the kind of scenery I expected on a very wet day in the Pacific Northwest.

I’ve been thinking lately about the way tech companies all seem to want us to use AI to buy more things — even more so after a recent conversation with my friend Will Sattelberg at 9to5Google. The way that every AI demo ends in booking a flight or buying a new pair of sneakers is getting pretty old. But it’s not just a tech company thing — I’ve also been examining my own tendency to seek out a transaction whenever I leave the house.

I gravitate toward places I can leave with a new book, or a coffee, or a little treat, I think partially as a way to alleviate the anxiety of just… existing in the world. How do you pick a place on the map when there are thousands to choose from? What if I choose wrong and have a bad time? Buying a little trinket proves to myself that I went somewhere worthy, my thinking goes. But the feeling never seems to last, so before long I’m back on Google Maps planning another excursion to find the one perfectly curated home goods store that will fix me. 

In any case, I did leave the conservatory with some mementos — a soggy admission ticket and a couple of kid-sized gardening tools from the gift shop. My kid loves digging up the yard while I pull weeds, and it’s something I want to lean into more. Sometimes the perfect excursion out of the house can just be playing in the dirt that’s right outside, you know? Anyway, a warm, dry route 10 bus was waiting for me to head toward my final stop of the day: coffee.

I hadn’t heard of Day Made Kaffe, which struck me as odd, because I’m in the neighborhood where it’s located pretty frequently. Based on Gemini’s description of a minimalist but warm and laptop-friendly coffee shop, it also sounded like my kind of place. I had been there before, I realized when I walked in, before it was a coffee shop. The place used to be a — you guessed it! — fancy home goods store where I bought a couple of Christmas gifts in 2024. Time is a flat circle, etc.

Gemini didn’t miss; Day Made is extremely my shit. The coffee was great and the vibes were immaculate. The cardamom bun that Gemini said I should get wasn’t available, so I got a pastry with guava jelly to compensate for the dismal weather. I watched the Artemis II launch on mute, left the shop at 3:40 as Gemini had instructed, and got on my last bus of the day. The time when I walked in the door at home? 4:26. Nailed it.

If my big day out in the city was a success — and I think it was! — then it was made possible by people, not Gemini. People wrote the reviews and recommendations that led me to Tacos Chukis. Gemini is just the middleman. But when you’re dealing with a vast and often overwhelming dataset like the one in Google Maps, a tool like Gemini strikes me as a very useful one.

I often rely on user reviews to suss out how kid-friendly a place really is, and Gemini lets me search a lot of those reviews across a wide area all at once to find a place that both serves dinosaur-shaped chicken nuggets and craft cocktails. But crucially, when it’s time to get from point A to point B, the LLM doesn’t just freestyle and talk you through transit directions — you just open up the transit directions in Google Maps, which includes accurate real-time information. 

Gemini also does a decent job of showing its work along the way as it makes suggestions so you can see where it’s coming up with claims. It’s not immune from hallucinations, and that’s a particularly big concern when you’re relying on it to steer you through the real world. But knowing that’s the case, I still think this is an impressive tool — whether you’re trying to find a restaurant nearby with high chairs and right now because everyone is hungry and crabby, or whether you’re on a more leisurely adventure of discovery.

Photography by Allison Johnson / The Verge

I’ve been testing out Gemini’s new task automation on the Pixel 10 Pro and the Galaxy S26 Ultra, which for the first time lets Gemini take the wheel and use apps for you. It’s limited to a small subset right now — a handful of food delivery and rideshare services — and it’s still in beta. It’s slow, it’s clunky at times, and it doesn’t solve any serious problem you had using your phone. But it’s impressive as hell, and I don’t think it’s hyperbole to say this is a glimpse of the future. We’re still a long way off, but this is the first time I’ve seen a true AI assistant actually working on a phone — not in a keynote presentation or a carefully controlled demo inside a convention hall.

First off: Gemini is much slower than you, or me, or most anyone at using their phone. If you need to order an Uber right this second, you’re still the best person for the job. Before you write it off, though, remember that task automation is designed to run in the background while you do other things on your phone. Even better, it keeps working while you’re not looking at your phone, so you can do things like check that your passport is in your bag for the 10th time.

But if you’re curious, like I am, you can watch the whole thing happen. While it’s working, text appears at the bottom of the screen indicating what Gemini is doing. Stuff like “Selecting a second portion of Chicken Teriyaki for the combo,” which it did when I directed it to order my dinner on Saturday night. Watching Gemini figure things out on the fly honestly kinda rules. I asked for a chicken combo plate; the menu presented options in half- portion increments, so it correctly added two half servings of chicken.

It’s for the best that when you start an automation with Gemini, the default behavior is for it to run in the background. You have to tap a button and open another window if you want to watch Gemini working through the task. And it can be excruciating. Watching the computer try to find a side of greens on a menu in Uber Eats when it’s sitting right there at the top of the screen is like watching a horror movie and knowing the murderer is in the closet right next to the protagonist. I mean, except for the murder part. Gemini made a couple of wrong turns as it put together my teriyaki order, which it eventually figured out on its own, but the whole episode took about nine minutes. Not ideal.

Gemini is supposed to carry out your task right up to the point where it’s time to hit confirm and order your car or dinner so you can double-check its work. This, I think, is the only sane way to use this feature right now, and I don’t mind the added friction of completing the order. In the tests I’ve run over the past five days, I’ve never had it go rogue and finish my order for me. And it is surprisingly accurate; I’ve had to make very few adjustments to the final order. If it fails — which I have seen happen a couple of times — it tends to be within the first minute or two when something about the app needs my attention, like giving it permission to use my location, or changing the delivery location to home rather than Nevada, which was the last place I used that app. I had to figure out what the problem was in cases like this, but once it was sorted out I was able to restart the automation without an issue. 

Here’s the one that really got me. I put an event on my calendar for a flight to San Francisco the following day (a pretend trip for me, but real flight details). I gave Gemini a vague prompt to schedule an Uber that would get me to the airport in time for my flight tomorrow. Because Gemini has access to my email and calendar, it can go find that information. It did need a little extra guidance — possibly because the flight wasn’t in my email like it expected. But with that, it found the flight information, suggested leaving by 11:30 or 11:45AM (logical timing for a 1:45PM flight given I live close to the airport), and asked if I wanted to schedule a ride for one of those times. I confirmed the time, and it went about setting up the ride in about three minutes with no further input required on my part.

It’s a little more impressive when you consider that Uber doesn’t even refer to it as scheduling a ride — you reserve a ride. That’s the key difference between the digital assistants we’ve been using and the AI assistants emerging now. Being able to use natural language when talking to the computer makes a huge difference when you’re controlling your smart home or placing your dinner order. If the computer is going to get tripped up and ask for clarification when you forget that the restaurant calls your meal a “plate” and not a “combo,” or if you ask for “slaw” instead of “shredded cabbage,” then it’s no more useful than the assistants we’ve been using for the past decade to set timers and play music. 

That said, watching Gemini tap and scroll around Uber Eats makes one thing painfully obvious: If you were designing an application for AI to use, it would look nothing like the ones we have today. You know, apps designed for humans. An AI assistant won’t be tempted by a big ad in the middle of a page to save 30 percent on your order. An appetizing, well-staged photo of the dish it’s ordering isn’t any more convincing than a low-quality one. You would give it a database, not a bunch of clutter to weed through — something the industry is working toward in Model Context Protocol, or MCP.

An AI model reasoning its way through a human-centric interface feels like the most impractical and brittle way to place a pizza order. It does hit a snag occasionally, and it’s not great at telling you why it couldn’t do something. This version of task automation feels like a stopgap until app developers adopt more robust methods: MCP or Android’s app functions. Google’s head of Android, Sameer Samat, told me recently that Gemini takes the reasoning approach in the absence of the other two. Maybe this version of task automation is our preview of what’s possible, or a way to prod developers into adopting one of the other methods. Either way, this feels like a notable first step toward a new way of using our mobile assistants — awkward, slow, but very promising.

Photography by Allison Johnson / The Verge