One of the biggest wishlist items for the average user when it comes to AI is the “personal assistant.” The dream is the Star Trek-style computer where you just ask for something and the computer does the rest. Maybe not as far as a replicator giving you “Tea, Earl Grey, hot” but something to manage the day-to-day would be excellent.
The problem is apps don’t play nice with one another. They aren’t meant to given they are owned by different companies. For years, if you wanted, say, attachments from your mom sent to your Outlook to be stored in your Dropbox, you had to use some type of connector like Zapier or Make. Costly and inefficient.
Recently, OpenClaw threatened to blow that world wide open. It’s an incredible tool from a developer now working for OpenAI that can connect all your various apps and technology together. In simple terms, you give it permission to take control of a computer that has all your apps and password access. It then follows your commands to do anything from sending someone an email to building complicated tools for managing your money. It’s part of the agentic AI movement that many believe will lead to startling breakthroughs in how we manage our lives.
Here’s the problem for most of us: it’s insanely complex and doesn’t always work the way you want. It’s a bit like a beta version of a brand new software. Unless you are comfortable with code and bugs and all of that nonsense, don’t bother.
Which is why when I heard about ChatGPT’s browser, Atlas, I was intrigued. Could AI control my browser in a meaningful way?
Actually, yes!
Atlas works just like ChatGPT but paired with its own browser built off of the open source chromium browser platform. So, it looks and acts just like a regular web browser, but it has a sidebar with ChatGPT built into it.
In its most simplistic, it will help explain what is on the page. Think summarizing a very long article or explaining a medical study in layman’s terms. It can also handle much more complex searches than Google. While Gemini, Google’s own AI, is great, if you prefer using ChatGPT (as I do) and have already sunk a lot of time into it (as I have), this is the better option.
If you’ve ever used ChatGPT to build an itinerary, for example, you know it can really help sort out all kinds of things from restaurants and attractions to accommodations. But, with the browser, it can expand upon that by finding very specific things based on your prompts.
“Find a pet-friendly rental on Padre Island for seven days from May 11-17 and make sure it is on the beach, not a condo, and is less than $3,000 for the week.”
It does that sort of thing really well, but then you can take it a step further which is where “agent mode” comes in. This is like asking the browser to do the work for you. You give it tasks to accomplish and it handles it like booking a trip or making reservations at a restaurant or finding and summarizing news about a specific topic for you.
It’s like agentic AI-lite. It does a lot of the same things, but it needs more of your attention. It won’t operate on its own and it only accesses the things you provide access to. For example, if you wanted it to open your bank account and transfer money to a friend for you, it cannot do that unless you provide access to the bank account. And it will ask – sometimes repeatedly – if you are sure.
That makes it a little more cumbersome than traditional agentic AI systems, but a lot less complicated and certainly safer.
Like all AI chatbots, it is only as good as what you tell it to do. The more specific and targeted the request, the better the results. Additionally, it is a bit slow and does not handle multiple tasks at once well (usually it will need to pause during particularly long requests to confirm information or need additional instruction). But because it can control your browser window, it can control anything you can through a browser interface which, for many of us, is almost everything.
It’s not quite the Star Trek future, but it unlocks some key functions you don’t get with a standalone chatbot, and you don’t have to have a degree in computer science to figure it out.
