Google just rolled out Auto Browse, an AI agent that promises to handle your digital chores by taking control of Chrome and clicking through tasks like booking tickets and shopping. But a hands-on test by Wired reveals the tool's critical flaw - it lacks common sense. The feature, available now to $20-a-month AI Pro and AI Ultra subscribers, can technically navigate websites and execute commands, but it stumbles on decisions that humans make instinctively, raising questions about whether AI agents are ready to manage our daily browsing.
Google is making its biggest bet yet on AI agents reshaping how we use the web. Auto Browse, which launched this week to US subscribers of the company's AI Pro and AI Ultra plans, promises to handle the tedious clicking and scrolling that fills our days. Book concert tickets, shop for clothes, plan camping trips - all while you sit back and watch the bot work.
But Reece Rogers' hands-on test for Wired reveals a tool that's technically proficient but fundamentally flawed. The AI can navigate websites and execute commands, sure. What it can't do is think like a human.
Rogers discovered this the hard way when he asked Auto Browse to book two symphony tickets with specific requirements - not orchestra seating, not the cheapest available, and next to an aisle. The bot delivered exactly what was requested. Almost. The two $185 seats were indeed by an aisle and outside the orchestra section. They were also in separate rows, meaning Rogers' partner would spend the concert staring at the back of his head.
"It's the kind of common sense decision that I didn't even consider including as part of my prompt," Rogers wrote. The miscommunication highlights a critical gap between what AI agents can technically accomplish and what users actually need.
Google has positioned Auto Browse as part of a broader vision to fundamentally alter how we interact with the web. The feature joins AI Overviews in Search and Gemini integrations in Gmail as tools designed to remove users from firsthand experiences in the name of efficiency. Chrome, the world's most popular browser by far, gives Google enormous leverage to reshape these daily habits.
Accessing Auto Browse requires a $20-a-month subscription to AI Pro or AI Ultra plans. Users activate it through the Gemini chatbot sidebar in Chrome, where a sparkle icon summons the AI assistant. Once enabled, any direct request automatically triggers Auto Browse to start clicking through websites on your behalf.
The experience of watching the bot work is, as Rogers describes it, "a bit bizarre." Auto Browse first uses Gemini 3, Google's latest model, to strategize and define goals - similar to reasoning models that talk through steps before acting. Then the automated clicking begins, with each action logged for users to review.
Rogers tested three scenarios that Google highlighted in press briefings as ideal use cases. The symphony ticket booking failed on nuance. A fashion shopping task on Depop fared slightly better - the bot successfully filtered for men's XL leather jackets and added three options to the cart. But it simply grabbed the first three search results without any apparent curation or taste.
"Picking the top three search results doesn't inspire confidence that it made any kind of qualitative judgement," Rogers noted. For a shopping task that's supposedly about personal style and preference, Auto Browse acted like a search engine with a cart button.
The camping trip research proved most disappointing. Rogers asked for five campground options within three hours of San Francisco, available for four nights in April or May. After 15 minutes of processing - an eternity in AI time - Auto Browse checked actual availability for just one location in Point Reyes, then suggested Rogers visit Reserve California himself to check the others. So much for automation.
Google has built safeguards into Auto Browse that flag sensitive actions like purchases or social media posts, requiring user approval to proceed. A persistent warning in the sidebar reads: "Use Gemini carefully and take control if needed. You are responsible for Gemini's actions during tasks."
That disclaimer points to deeper concerns about security and trust. Generative AI tools remain vulnerable to prompt injection attacks on malicious websites, where hidden commands could divert the bot from its intended task. The potential vulnerabilities in Auto Browse haven't been fully examined by outside researchers yet, but the risks could mirror other AI tools that take computer control.
Handing financial information to an AI agent that demonstrably lacks common sense raises obvious red flags. Rogers approached testing with "a healthy sense of skepticism" and anxiety about "the havoc it could potentially wreak" with his credit card.
The performance issues Rogers documented suggest Auto Browse isn't ready to be trusted with consequential tasks. It can technically execute commands and navigate websites, but it can't make judgment calls that humans handle instinctively. A true internet surfer, Rogers argues, would never just grab the first three Depop listings. They'd scroll and hunt for unexpected options, enjoying the discovery process.
"My meandering quests through the internet's backwoods are certainly not the most efficient journeys," Rogers wrote. "But they remain delightful to me, and I would never want to outsource that joy completely to a browser bot."
Google's vision of a post-click internet, where AI agents browse on our behalf while we're removed from the experience, faces a fundamental challenge. Efficiency matters less if the results are wrong or unhelpful. Rogers found himself doing more work, not less, after Auto Browse's attempts to help.
The limited release to paid subscribers suggests Google knows Auto Browse needs refinement before broader deployment. Whether the company can close the gap between technical capability and practical usefulness will determine if AI agents become genuinely helpful tools or expensive novelties that create busywork.
For now, Rogers plans to keep clicking himself. The future of browsing might involve AI agents navigating a web redesigned for their needs, but that future isn't here yet. Auto Browse can open tabs and fill forms, but it can't replace human judgment about where to sit at the symphony.
Google's Auto Browse represents an ambitious vision for AI-powered web navigation, but the execution reveals we're still in the early experimental phase of AI agents. The tool can technically perform clicks and navigate sites, yet it consistently fails at the intuitive decision-making that makes browsing useful. Until AI agents can match human judgment on everyday tasks like picking adjacent concert seats or curating shopping options based on taste rather than search rank, they'll remain expensive curiosities that create extra work rather than saving time. For $20 a month, Chrome users might be better off doing their own clicking - at least until Google's bots learn some common sense.
