DooDooLamb News
Gemini task automation is slow, clunky, and super impressive
Brief published March 23, 2026 ยท Original source published March 21, 2026
Original reporting by Allison Johnson at theverge.com.
Automated brief. Verify important details at the original source.
Google's AI Spends Nine Minutes Ordering Dinner, Somehow Still Beats Your Average Date
Google's Gemini AI can now control your phone apps directly, turning what should be a thirty-second food order into a nine-minute digital performance art piece. But hey, at least it's consistent with every other Google product launch: technically impressive, practically frustrating, and definitely not ready for your actual life.
The Verge's David Pierce has been testing Gemini's new task automation feature on Android phones, watching Google's AI fumble through DoorDash orders like a well-meaning but extremely methodical grandmother trying to figure out "the Uber Eats." The feature, currently limited to a handful of food delivery and rideshare apps, represents Google's latest attempt to prove that AI agents can handle the mundane digital tasks that consume our lives. The results are exactly what you'd expect from a company that somehow made messaging more complicated every time they tried to fix it.
Task automation, for those blissfully unaware of the latest AI buzzword bingo, is essentially giving an AI the ability to tap, swipe, and type on your behalf across different apps. Think of it as a robot butler for your phone, except instead of serving you tea on a silver platter, it's spending nearly ten minutes trying to add extra sauce to your burrito order. The technology represents a significant leap from traditional AI assistants that could only tell you the weather or play your Spotify playlist. Now they can actually interact with app interfaces the same way humans do, which means they can also get confused by bad UX design the same way humans do.
Pierce's testing reveals the beautiful absurdity of our AI-powered future. Gemini successfully completed tasks like ordering food and booking rides, but with the speed and grace of someone wearing oven mitts trying to thread a needle. The AI carefully reads through menus, methodically selects options, and double-checks every step with the kind of thoroughness that would make a compliance officer weep with joy. It's the digital equivalent of watching your most careful friend order at a restaurant, except your friend doesn't need to parse natural language instructions and translate them into precise UI interactions while running on a phone processor.
The technical achievement here shouldn't be undersold. Getting an AI to understand app interfaces, navigate complex menus, handle edge cases like sold-out items, and complete multi-step transactions represents genuine progress in AI capabilities. This isn't just a chatbot regurgitating training data; it's a system that can perceive, reason about, and manipulate digital environments in real-time. The fact that it works at all is legitimately impressive, even if "working" currently means "eventually gets the job done while you question your life choices."
But the gap between "technically possible" and "practically useful" remains Grand Canyon-wide. Nine minutes to order dinner isn't just slow; it's slower than calling the restaurant directly and explaining your order to someone who doesn't speak your language. It's slower than walking to the restaurant. It's approaching "I could have grown the ingredients myself" territory. The AI's careful, methodical approach makes perfect sense from a reliability standpoint but completely misses the point of convenience.
Google's positioning this as an early glimpse of AI agents that will eventually handle complex, multi-step tasks across our digital lives. The company envisions a future where you can ask your AI to "plan a weekend trip" and watch it coordinate flights, hotels, and dinner reservations while you focus on more important things like arguing about JavaScript frameworks on Twitter. It's a compelling vision, assuming you're comfortable with an AI that currently needs fifteen minutes to add guacamole to your order making travel arrangements that could strand you in Cleveland.
The reality is that we're watching the awkward adolescent phase of AI automation, complete with all the growing pains that implies. These systems work, sort of, sometimes, if you're very patient and have realistic expectations and maybe a backup plan. For developers, it's fascinating to watch AI tackle the same interface frustrations that plague human users. For everyone else, it's a reminder that the future arrives not with a bang but with a very, very slow loading screen.
The most honest take might be that we're witnessing something genuinely significant happening in slow motion, like watching evolution through a microscope. Today's nine-minute dinner order could be tomorrow's instant everything, assuming Google doesn't get distracted and pivot to making AI-powered smart glasses that nobody asked for.