ChatGPT Atlas: Your Browser as a Personal AI Assistant

Last week I wrote to you about Google AI Studio and how it allows you to create applications with practically no code. So far, the easiest way to create a web app with AI for non-programmers.

ChatGPT Atlas? Well, I don’t have the same level of enthusiasm. But it can be a useful tool that’s worth trying out.

I’ve spent the last week diving into ChatGPT Atlas, OpenAI’s new AI-powered browser, and on one hand, it surprised me.

Today, I’ll break down how ChatGPT Atlas aims to change the way we work, shop, research, and plan with real-world examples, including my own hands-on tests.

How ChatGPT Atlas Turns Your Browser Into a Personal Assistant

What impressed me the most was this: Atlas isn’t just another browser with a chatbot tacked on. It’s a fundamentally different approach where the AI can actually browse websites, fill out forms, compare products, and perform multi-step tasks while you supervise and approve everything.

Imagine having a smart assistant that lives in your browser and remembers everything you’ve researched—of course, only if you allow it to.

Agent Mode: A Feature Worth Mentioning

“Agent Mode” allows Atlas’s ChatGPT to take control of your browser and perform tasks that would normally take you more time. So, things like comparisons, searches, analyses…

What makes it different:

You see a cursor moving and working in your tab in real-time.
You can intervene or stop it at any time. - It performs multi-step tasks just like a human would.
Every action that requires login or confirmation waits for your approval.

My Personal Test: Atlas vs. Regular ChatGPT

To see the difference, I ran the same prompt in both Atlas and regular ChatGPT:

“Around November 1st, I’m planning a visit to Gardaland in Italy. Make me a 3-day travel and activity plan that would be interesting for a 10-year-old child.”

And the result?

Agent Mode in Atlas:

Automatically opened the Gardaland crowd calendar.
Checked the actual expected crowd levels for early November.
Pulled real-time information directly from the source.
Created a plan based on the actual crowd data.

Regular ChatGPT:

Did a web search and found useful information.
Noticed that last year on October 31st, Gardaland was open until midnight.
Suggested we could go then if we decided to.

chatGPT Atlas

Both worked. Each had its strengths; for example, ChatGPT created a more detailed schedule and transport plan, while Atlas was better at targeted activity planning and checking occupancy. For these purposes, I’ll probably still use ChatGPT directly.

I also tested this prompt: “Act as the best agency I could hire to review and suggest improvements for my website: ideasuniverse.com.”

“Agent Mode” independently browsed through various subpages, checking meta tags, loading speeds, and structure. It was like watching an SEO consultant go through their checklist, only faster and without the hourly rate.

chatGPT Atlas

A great feature is definitely: Atlas can see images on websites. This means you can ask it questions about visual content, for example: “What design patterns does this site use?” or “Analyze the layout of this landing page.” This is great when you don’t know exactly what to ask or need inspiration. Just point it to a page with interesting visuals and ask for ideas or analysis.

A new way of searching?

Dr. Lee, an academic researcher, needed to prepare a literature review on the ethics of gene editing. She asked Atlas: “What are the main ethical dilemmas in gene editing with CRISPR technology?”

Atlas immediately served her a concise summary. She then followed up with the question: “Find me recent scientific articles on the misuse of gene editing.”

Atlas opened relevant search results and directed her to key articles. She then read the full texts herself. The workflow? Atlas outlined the topic and directed the search, while she did the in-depth reading with classic browsing.

Shopping without the tab chaos

Comparing products used to mean juggling multiple tabs and spreadsheets. Atlas does this differently.

“Agent Mode” can:

Fetch product specifications from multiple websites.
Compare user ratings.
Track price changes.
Recall products you viewed last week.
Add products to the cart (with your permission).

Someone shopping for a laptop ordered Atlas to compare processor speed, screen resolution, and battery life. The AI gathered the data, highlighted key differences, and even marked models that were recently discounted.

The memory feature that gets little talk

Here’s another key advantage: Atlas can remember your browsing history—but only if you explicitly allow it.

Let’s say you were researching competitors on Tuesday. On Friday, you can simply say: “Show me the competitor analysis I was doing on Tuesday,” and Atlas will serve you the relevant pages and notes.

It works like a personal librarian who knows exactly where you’ve stored everything. No more digging through your history or guessing where you saw something.

Privacy controls are built-in:

You can turn the memory on or off at any time.
A full “incognito” mode is available.
You decide what it remembers.
By default, your data is not used to train AI models.

Will I be using it?

The basic principle of Atlas is simple: to shorten the time from information to action.

Will I be using it?

Honestly, I don’t know.

You certainly have more control over the context (information) from which you want an action. Because you can go to a website and Atlas manages the site like a human.

Install Atlas for yourself (currently available only for Mac), explore the assistant in the sidebar, and ask it to summarize the page you’re on. Then gradually test “Agent Mode” with tasks you can supervise.

Decide for yourself.

The most important skill? Trust, but verify. Let the AI do the heavy lifting, but always review and confirm the final steps.

Talk to you soon,

Primož