(MENAFN- Khaleej Times)

For years, we've had AI in our browsers - think Copilot in Edge, or AI summaries in Perplexity. But OpenAI wants something different: AI as your browser. Enter ChatGPT Atlas, the company's new Chromium-based browser that promises to help you browse, research, write, and even automate tasks - without leaving your tab. I spent the past few days using Atlas for work, research, and daily browsing. Here's what stood out, what fell short, and whether this new AI companion can really replace your trusty Chrome.

For someone who's been using Chrome for more than a decade, Atlas feels instantly familiar - it's Chromium under the hood - but cleaner. The left panel hosts ChatGPT, always present yet unobtrusive. There, you get all your conversations.

Recommended For You UAE President offers condolences to Saudi King over passing of Princess Nouf bint Saud

On the right, a tab opens when you click 'Ask ChatGPT'. You can type, talk, or highlight text and ask it to summarise, explain, or even rewrite what's on screen.

The setup was smooth. I signed in with my OpenAI account, imported my bookmarks, and was up and running in under five minutes.

The overall vibe is“digital assistant on the go.” The question is whether that assistant is useful enough to justify its presence. Sometimes yes, sometimes no.

Coming back to the sidebar on the right, where you can highlight text, ask for a rewrite, or request a quick summary of an article, it usually delivers something readable and fast. For routine things - pulling background info, creating quick summaries, or rewriting copy to a different tone - it can genuinely speed up parts of your workflow. That's the core promise delivered. It eliminates the part where you seek an AI tool in a different window to summarise or get information.

But it also introduces new habits. I found myself pausing more often: should I ask Atlas for a rewrite, or just do it myself? Will the summary be accurate? The features are attractive and, only when accurate, genuinely useful. When they're wrong, they're not just annoying, they erode trust.

Agent Mode

This is the feature that sounds most futuristic: you can ask it to perform tasks that usually need a dozen clicks like find a product, add it to cart, even draft an email. It's the browser equivalent of saying,“Handle this for me.”

But here's the catch: Much of the truly useful automation sits behind upgrades. I hit“upgrade” or“limit reached” nudges repeatedly.

It's undoubtedly a smart concept, if accurate (again), almost like dictating a task to a digital intern, but the constant reminders to upgrade make the experience feel gated at times.

Testing its accuracy

This is where Atlas lost me. I started testing accuracy with everyday queries relevant to my beats and I wish I hadn't.

Example: I asked for the best movies of 2025 and Atlas returned Ghost of Yotei as a film - that's a game I've been playing, not a movie. The model had clearly mixed up my browsing signals and produced a confidently wrong answer. That kind of cross-pollination between personal browsing context and factual answers is alarming.

Worse: I asked Atlas to summarise a YouTube trailer (it suggested doing so), and the summary contained dramatic dialogue that simply doesn't exist in the trailer.

When Atlas is right it's helpful and time-saving. When it's wrong, it's actively harmful, and because the UI packages the claims so authoritatively, it's easy to mistake manufacture for fact unless you double-check sources.

But it's still early; currently only available on MacOS. Windows and mobile users will have to wait, and some features (like Agent Mode) are clearly in beta.

I'd say it is ambitious and useful in parts, but not yet trustworthy enough to replace the current workflow we depend on.

MENAFN29102025000049011007ID1110266396