Credit: Raghav Sethi/MakeUseOf

There’s been a wave of new “AI-powered” browsers lately, and I’ve tried pretty much all of them. Most of them promise a smarter, faster, more automated browsing experience, and end up feeling like the same browser with a chatbot glued on top.

I genuinely thought OpenAI would be the one doing something different. So when the new ChatGPT Atlas browser launched, I gave it a full week as my daily browser. But just like Dia, Comet, and the rest, it left me disappointed. After seven days of trying to make it work, I ended up uninstalling it. And my search for a browser that actually feels fresh continues.

Cool AI features don’t guarantee a better browser

AI for the sake of AI isn’t enough

Credit: Raghav Sethi/MakeUseOf

ChatGPT Atlas is pitched as the next leap in browsing, in the same category as things like Perplexity’s Comet. And on paper, sure, it sounds impressive. The browser pipes every search through ChatGPT, it has a sidebar that follows you from site to site, and it can even “act” on your behalf using its Agent mode. It is the same checklist of AI browser features we have seen a dozen times already. But once you use it for more than a week, the excitement wears off surprisingly fast. Because here is the thing: AI alone is not enough to make a browser better*.*

Right now, every new browser (Dia, Comet, etc.) is doing the exact same trick. They take Chromium and layer a chatbot on top of it. That does not make the core experience any smarter, any faster, or any more enjoyable. It just adds another UI element to something that was already fine. After a while, it starts to feel repetitive. I am not switching browsers just to have another place to ask an LLM questions. I can already do that anywhere.

Arc understood this. Before it added AI, it came up with genuinely useful improvements to how you work in a browser. Vertical tabs that made navigation feel cleaner, or Spaces that helped separate your work and personal browsing. Even when Arc added AI, the focus wasn’t on having conversations with an LLM. It was on making the browser itself smarter, like how it could automatically organize your tabs for you.

With Atlas, there is none of that. It feels like the whole browser exists just to say, “Hey, look, AI is everywhere now*.”* And that is where I check out. If a browser wants me to switch, it has to rethink the fundamentals. I am tired of having AI shoved down my throat. Everyone has that. I want to see something genuinely new again.

Agent mode feels slow and underwhelming

I ended up fixing its work anyway

Credit: Amir Bohlooli / MakeUseOf

The idea behind Agent mode is simple: instead of just answering your questions, the browser can actually do* *things for you. It can open sites, click through pages, fill forms, search for products, and supposedly complete tasks inside the browser without your input. On paper, that sounds like exactly where AI should be heading. A browser that does the boring steps for you.

But in practice, Agent Mode feels slow and clumsy. It constantly trips over incredibly basic things. It sometimes fails to recognize that a page has a scroll bar, so it just sits there staring at the top of a page forever. Other times, it starts doing something and then gets stuck midway, like it forgot what the task even was. And when it does manage to complete something, it has this habit of drifting way off instructions, like it is improvising instead of following what you actually asked for.

That is a huge problem, especially because Agent Mode is positioned as something that can help with everyday workflows, like adding items to your cart while shopping or filling data into forms. But if you eventually need to redo everything yourself anyway, what is the point of the agent being there at all?

My breaking point was a simple task. I asked Agent Mode to take some data from a Google Sheets file and paste it into a table in Google Docs. A task that would have taken me ten seconds, maybe less. After over three minutes of watching it slowly move through menus, it finally “completed” the task. Except it didn’t paste the data. It created a summary of the sheet and generated a report explaining what it thought the sheet was about.

And that is what makes Agent Mode frustrating. I could forgive it being slow if it were accurate. But in its current state, it’s wasting my time instead of saving it.

There are some pretty major security risks

OpenAI even admits there could be problems

Credit: Amir Bohlooli / MakeUseOf

The issue isn’t just that Agent Mode is unpredictable. When a browser is allowed to click, fill forms, and interact with websites on your behalf, the door opens to real security concerns. Because the agent reads the page to understand what to do, the page itself can influence the agent’s behavior.

This is what’s known as prompt injection. A website can hide text that tells the agent to take a different action than what you asked for. So you might say something simple like, “Compare these two phones, and add the better one to my cart.” However, if the site has hidden instructions directing the agent to add the more expensive item to your cart or navigate elsewhere, the agent might actually follow those instructions instead.

And because this whole concept of agentic browsing is still new, there isn’t a reliable, guaranteed way to prevent this yet. Even OpenAI openly notes on Atlas’ download page that using Agent Mode comes with risks and that you should be cautious.

These are the last kinds of worries I want to have when I’m just trying to browse the web. If a feature isn’t consistent, and it might put my data or accounts at risk, the trade-off just doesn’t feel worth it.

ChatGPT Atlas still hasn’t nailed the basics

And this is where everything comes full circle. For all the talk about AI-first browsing, ChatGPT Atlas still struggles with the fundamentals. Strip away the ChatGPT sidebar and the agent features, and what you’re left with feels like a very plain Chromium browser. There’s no proper multi-profile support, no thoughtful tab management features, and not even something as simple as vertical tabThat’sat’s the part that makes the whole thing frustrating. In trying so hard to integrate LLMs into every corner of the browser, the basics have been ignored. Browsers are tools we spend hours on every day. The core experience has to be solid before you start layering on futuristic ideas. But Atlas feels like an AI layer sitting on top of something that could just as easily have been a Chrome extensioThere’sre’s almost no meaningful customization either. You can change your accent color, sure, that’sat’s about it. Everything else feels unfinished, like the browser shipped before the foundation was actually ready. It needed more time, more polish, and more thought put into how people actually browse.

Looking for something new again

Safe to sI’veI’ve stopped using Atlas entirely. My browser of choice used to be Arc, until it was replaced with Dia, which just never lived up to what made Arc special. So for nI’veI’ve taken temporary refuge in ChromIt’s

It’s not exciting, butdoesn’tsn’t get in my wI’m I’m still hoping a browser comes along that actually rethinks how we browse instead of just adding an LLM everywhere. But until thI’llI’ll be waiting.