I tried an AI web browser, and I’m never going back

Gemini in Chrome, Comet from Perplexity and Dia from the Browser Company: which one should you use? (Photo: Reuters)
Gemini in Chrome, Comet from Perplexity and Dia from the Browser Company: which one should you use? (Photo: Reuters)
Summary

OpenAI’s ChatGPT Atlas, Perplexity’s Comet and others have built-in chatbots and agents that can take over your tabs.

I’ve been using several AI web browsers lately. I’m never going back to the boring old kind.

An AI browser has a built-in chatbot that can see what’s open in your tabs. As you surf, you can type requests like: “Explain this." “Is this the best price?" “Make it vegetarian." Your artificial-intelligence browsing assistant instantly understands the context.

Browsers with advanced task-performing “agents" can click and type in your tab to complete tasks like filling a shopping cart.

You probably haven’t changed your web browser in a decade, because there was no reason to switch unless you had to. Now, there’s a dizzying number of new—better—options. This week alone, OpenAI announced the ChatGPT Atlas browser and Microsoft Edge added an agent to Copilot Mode. Gemini in Chrome, Comet from Perplexity and Dia from the Browser Company also all became free to a wider audience recently.

I tested the two main flavors—the assistants and the agents—and found benefits and risks in each. As with all AI interactions, anything you prompt will be sent to a company for processing, and the bot can make mistakes. Here are the safe ways to get the most out of them, and my vote for which one’s the best.

Assistants: Gemini and Dia

When Gemini in Chrome is active, a rainbow pulsing light indicates the AI can see your tab.
View Full Image
When Gemini in Chrome is active, a rainbow pulsing light indicates the AI can see your tab.

AI browsers save you time, especially if you already use a chatbot. Gemini in Chrome offers an easy introduction for Google Chrome users who don’t want to download a new app. (It’s rolling out to U.S. desktop and Android users now.) A new sparkle icon in your browser bar opens Gemini in a floating window.

I asked Gemini to give away the ending to the 10-minute YouTube video, “Your Microwave’s Most Underrated Button." (Spoiler: the power button for cooking foods more evenly.) And when I was cooking an egg dish from an Instagram video recipe, the “Live" voice mode showed me how to make it without a steamer basket, no typing necessary.

Willing to venture beyond Chrome? You don’t have to take a big leap. The following AI browsers are built with Chrome’s underlying technology, which means you can move all of the same extensions and bookmarks over.

Dia, available for Mac users only, has a similar AI sidepanel powered by a mix of different models and impressed me with its programmable “skills." Budget Buddy found cheaper alternatives to an expensive espresso machine, and Color Analysis helped me pick out a new watchband based on my hair, eye and skin color.

Dia has a unique feature: programmable, repeatable ‘skills’ such as Budget Buddy, which suggests a less expensive option for whatever you’re shopping for.
View Full Image
Dia has a unique feature: programmable, repeatable ‘skills’ such as Budget Buddy, which suggests a less expensive option for whatever you’re shopping for.

The Daily Wrap skill scanned my browsing history to recap yesterday’s work and today’s tasks. It was unnervingly accurate—down to an unfinished email to a friend. While handy, it’s a stark reminder of the privacy trade-offs. For Dia to be most useful, it needs to see everything.

When the Memory function is turned on, your browsing activity is summarized on the company’s servers, then sent back to your computer for local storage. The Browser Company, which runs Dia and was recently acquired by corporate software giant Atlassian, says it doesn’t store the data, and its AI partners wipe queries after processing.

Agents: ChatGPT Atlas and Comet

AI browsers with agents can handle general chatbot stuff—and go a step further, tackling complex requests in the background. But at first, it’s hard not to watch them work.

I asked Agent Mode in ChatGPT Atlas, available on Macs, to find flights for a coming trip. As the tab window glowed orange, the bot searched for the right dates and identified the best options.

At times, Atlas struggled with clicking the correct button; it was like watching my toddler feed himself—inefficient but ultimately successful. The whole process took 16 minutes. Painfully slow, sure, but if I actually let the agent work while I did other stuff, would the duration matter?Atlas also has a few clever privacy-protecting settings: a blocklist for sites the agent can’t go to and a toggle in the URL bar to prevent ChatGPT from seeing the page while you prompt. For now, only paid ChatGPT accounts can use Agent Mode.

Perplexity’s Comet can perform similar actions for users without a paid plan. I tasked Comet and ChatGPT Atlas with filling a Whole Foods cart with healthy groceries for new parents. Both came up with good ideas—Comet even ensured each item was snackable one-handed—but they added items to my Whole Foods and Amazon carts, complicating delivery. Remember: It’s still early days for these agents.

Comet is limited to desktops for now, though Perplexity Chief Executive Aravind Srinivas told me a mobile app will be ready in less than two months. (Last year, The Wall Street Journal publisher Dow Jones sued Perplexity for copyright infringement.)

Microsoft Edge’s Copilot Mode now performs agentic “actions," albeit more limited than the others. You can set the agent’s restrictions from light (no permissions needed) to strict (always request permissions before acting). While the AI couldn’t add items to my Amazon cart, it scoured Wayfair for my perfect standing desk, searched Airbnb for rental properties and found Halloween costumes on Facebook Marketplace.

Gemini can’t perform agentic tasks yet, though Anthropic’s Claude extension adds agentic capabilities to Chrome for its top-paying Max subscribers ($100 a month).

Your privacy and security

With an AI browser, you’re sharing more than you usually would with an AI chatbot. Parts of the webpage are sent to a server for processing. While the companies attempt to redact personal data, it isn’t perfect. As a rule, avoid sharing medical, financial or proprietary corporate information while prompting. When possible, opt to store browsing activity locally, so it isn’t sent to company servers, and use incognito mode for sensitive queries.

It’s scary, having a bot bop around the web for me, and there are real risks. What if it navigated to a malicious website while I wasn’t looking? But I was quickly hooked on delegating tedious, low-stakes tasks like booking restaurant reservations and finding furniture with precise dimensions.

Use agents judiciously, and only on trusted sites. They are vulnerable to prompt injection attacks, where hackers hide malicious instructions for AIs to carry out.

The winner is…

So, which AI browser should you use? The app with the slightest edge….isn’t Edge (sorry). It’s Comet. The citation-driven chatbot, capable agent and especially the smart tab management won me over. ChatGPT Atlas, with its granular privacy controls, is a close second. And I’ll still use Dia’s shopping skills.

A caveat: This recommendation could change in the coming months as the landscape continues to evolve. The AI browser race is on.

Write to Nicole Nguyen at nicole.nguyen@wsj.com

Catch all the Business News, Market News, Breaking News Events and Latest News Updates on Live Mint. Download The Mint News App to get Daily Market Updates.
more

topics

Read Next Story footLogo