If you read my stuff regularly, you know that we’ve been blessed with another tidal wave of “AI-powered” browsers lately, and like a glutton for punishment, I’ve test-driven basically all of them. You can read the test-results over here: I tried all AI “Frankenbrowsers” out there | LinkedIn, and then I wrote a follow-up piece about the security nightmare that comes bundled free of charge when you sign up for these Beta platforms.
Yet, still, I wasn’t ready to give up that easily. So, in a moment of optimism – or madness (you decide) – I decided to give Atlas one last chance to win me over. I ditched my reliable old workhorse, Edge, and forced myself to use Atlas for an entire week.
These things all pinky-promise a smarter, faster, more automated browsing experience, and somehow they all end up feeling like they’re the exact same browser with a chatbot awkwardly duct-taped to the side because all of them are built on the same platform – ChromeOS.
Before I started, I honestly believed that OpenAI would break the mold and do something actually different. So when they birthed Atlas, I committed to using it as my daily driver for a full week. But surprise, surprise – just like Dia, Comet, Fellou, and all the other AI browser wannabes, it managed to thoroughly disappoint me. After seven whole days of desperately trying to make this thing work, I gave up and uninstalled it. And so my noble quest for a browser that actually feels fresh instead of recycled continues into the abyss, and I simply returned to Edge, waiting for the industry to undertake a second attempt at dethroning the current masters of our digital universe.
More rants after the messages:
- Connect with me on Linkedin 🙏
- Subscribe to TechTonic Shifts to get your daily dose of tech 📰
- Please comment, like or clap the article. Whatever you fancy.
Slapping cool AI features on something doesn’t magically make it better
ChatGPT Atlas, gets marketed as the revolutionary next leap in browsing, lumped into the same category as stuff like Perplexity’s Comet. And yeah, on paper it sounds pretty darn impressive, I’ll give them that. The browser forces every single search through ChatGPT, it’s got this sidebar that stalks you from site to site, and it can supposedly “act” on your behalf using its fancy Agent mode. It’s literally the same boring checklist of AI browser features we’ve already seen a dozen times over. But once you actually live with it for more than a week, that initial excitement evaporates surprisingly fast.
Because the truth is that AI by itself is absolutely not enough to make a browser better.
At this point, every single new browser (Dia, Comet, you name it) is pulling the exact same tired trick. They grab Chromium, slap a chatbot on top of it, and call it a day. That doesn’t magically make the core experience any smarter, any faster, or remotely more enjoyable. It just tacks on another UI element to something that was already working just fine. After a while, the whole thing starts feeling painfully repetitive. I’m not about to switch browsers just to have yet another place to pester an LLM with questions. I can already do that literally anywhere.
Arc actually understood this concept. Before they even touched AI, they came up with legitimately useful improvements to how you actually work in a browser. Vertical tabs that made navigation feel genuinely cleaner, or Spaces that helped you separate your work and personal browsing like a functional adult. Even when Arc finally added AI, the focus wasn’t on having cutesy conversations with an LLM. It was about making the browser itself legitimately smarter, like automatically organizing your tabs for you without being asked.
With Atlas, there’s absolutely none of that thoughtfulness. It feels like the entire browser exists purely to scream, “Hey everyone, look, AI is everywhere now!” And that’s precisely where I mentally check out. If a browser wants me to actually switch, it has to fundamentally rethink the basics.
I am genuinely exhausted from having AI aggressively shoved down my throat.
Literally everyone has access to that now. I want to see something genuinely innovative again, not this recycled nonsense.
Agent mode feels painfully slow and massively underwhelming
The grand idea behind Agent mode sounds simple enough, instead of merely answering your questions like a boring chatbot, the browser can supposedly actually do real things for you. It can open sites, click through pages, fill out forms, hunt for products, and theoretically complete entire tasks inside the browser without requiring your precious input. On paper, that sounds like exactly where AI should realistically be heading. A browser that handles all the boring, repetitive steps for you.
But in actual practice, Agent Mode feels excruciatingly slow and embarrassingly clumsy. It constantly trips over incredibly basic things that would make you cry. It sometimes completely fails to recognize that a page even has a scroll bar, so it just sits there staring blankly at the top of a page for all eternity. Other delightful times, it starts doing something halfway competently and then gets completely stuck midway through, like it suddenly forgot what the entire task even was. And when it miraculously does manage to complete something, it has this charming habit of drifting wildly off your actual instructions, like it’s improvising jazz instead of following what you specifically asked for.
That’s a catastrophically huge problem, especially because Agent Mode gets positioned as something that can supposedly help with everyday workflows, like adding items to your cart while shopping or filling data into tedious forms. But if you ultimately need to redo literally everything yourself anyway, what exactly is the point of having the agent there at all?
My personal breaking point was an absurdly simple task.
I politely asked (cause saying ‘please’ apparently helps, is what ‘they’ say) Agent Mode to take some basic data from a Google Sheets file and paste it into a table in Google Docs. A task that would have taken me roughly ten seconds, maybe even less. After over three agonizing minutes of watching it slowly, painfully move through menus like a confused sloth, it finally “completed” the task. Except – ding dong, plot twist! – it didn’t actually paste the data at all. Instead, it created a weird summary of the sheet and generated a whole report explaining what it personally thought the sheet was philosophically about.
And that’s exactly what makes Agent Mode so frustratingly useless. I could honestly forgive it being slow if it were actually accurate. But in its current pathetic state, it’s actively wasting my time instead of saving it.
There are some legitimately major security risks here
The issue isn’t merely that Agent Mode is wildly unpredictable. When you allow a browser to click things, fill out forms, and interact with websites on your behalf, you’re basically opening the door to some genuinely real security concerns, which I described exhaustively in my previous blog.
Because the agent has to read the page to supposedly understand what to do, the page itself can sneakily influence the agent’s behavior.
This lovely vulnerability is what’s known as prompt injection. A sketchy website can literally hide text that instructs the agent to take a completely different action than what you actually asked for. So you might innocently say something simple like, “Compare these two phones, and add the better one to my cart”. However, if the site has cunningly hidden instructions directing the agent to add the ridiculously more expensive item to your cart or navigate somewhere else entirely, the agent might actually follow those malicious instructions instead.
And because this whole trendy concept of agentic browsing is still painfully new, there currently isn’t a reliable, guaranteed way to prevent this exploit yet. Even OpenAI openly admits on Atlas’ download page that using Agent Mode legitimately comes with risks and that you should definitely be cautious.
These are precisely the last kinds of worries I want bouncing around in my head when I’m just trying to casually browse the web. If a feature isn’t remotely consistent, and it might actively put my data or accounts at genuine risk, the trade-off just absolutely doesn’t feel worth it whatsoever.
ChatGPT Atlas still hasn’t managed to nail the basic fundamentals
Cause it’s literally just Chromium with unnecessary extra steps
And this is where everything comes frustratingly full circle. For all the grandiose talk about “AI-first” browsing, ChatGPT Atlas still embarrassingly struggles with the fundamental basics. Strip away the ChatGPT sidebar and all the flashy agent features, and what you’re genuinely left with feels like an incredibly plain, boring Chromium browser. There’s no proper multi-profile support whatsoever, no thoughtful tab management features to speak of, and not even something as refreshingly simple as vertical tabs.
That’s the part that makes this whole thing so frustrating. In trying absurdly hard to cram LLMs into every conceivable corner of the browser, the actual basics have been completely ignored. Browsers are tools we literally spend hours on every single day. The core experience absolutely has to be rock-solid before you start enthusiastically layering on futuristic pipe dreams. But Atlas feels like a hastily-built AI layer awkwardly sitting on top of something that could just as easily have been a straightforward Chrome extension. There’s almost zero meaningful customization available either. You can change your accent color, sure, and that’s basically it. Everything else feels painfully unfinished, like the browser got shipped out the door before the foundation was actually ready. It desperately needed more time, more polish, and significantly more thought put into how actual people realistically browse.
Safe to say I’ve completely stopped using Atlas entirely. My browser of choice is still Edge, cause I’ve spent an eternity creating the perfect favorites structure, and I just love the workspaces and the side-bars, so for now, I’ve taken refuge in boring old Edge.
It’s not remotely exciting, but at least it doesn’t actively get in my way. I’m still desperately hoping a browser eventually comes along that actually rethinks how we fundamentally browse instead of just lazily adding an LLM everywhere possible.
Strangely enough, the industry uses the AI+ and +AI meme to explain to us that slapping AI onto an existing process (+AI) is not going to work, and that AI+ (redesigning old processes around AI) is the way to go. I use that concept in every AI project and it works. But apparently they don’t practice what you preach.
So, until that glorious day arrives that there’s an AI+ browser, I’ll be patiently waiting.
Signing off,
Marco
I build AI by day and warn about it by night. I call it job security. Big Tech keeps inflating its promises, and I just bring the pins and clean up the mess.
👉 Think a friend would enjoy this too? Share the newsletter and let them join the conversation. LinkedIn, Google and the AI engines appreciates your likes by making my articles available to more readers.
To keep you doomscrolling 👇
- I may have found a solution to Vibe Coding’s technical debt problem | LinkedIn
- Shadow AI isn’t rebellion it’s office survival | LinkedIn
- Macrohard is Musk’s middle finger to Microsoft | LinkedIn
- We are in the midst of an incremental apocalypse and only the 1% are prepared | LinkedIn
- Did ChatGPT actually steal your job? (Including job risk-assessment tool) | LinkedIn
- Living in the post-human economy | LinkedIn
- Vibe Coding is gonna spawn the most braindead software generation ever | LinkedIn
- Workslop is the new office plague | LinkedIn
- The funniest comments ever left in source code | LinkedIn
- The Sloppiverse is here, and what are the consequences for writing and speaking? | LinkedIn
- OpenAI finally confesses their bots are chronic liars | LinkedIn
- Money, the final frontier. . . | LinkedIn
- Kickstarter exposed. The ultimate honeytrap for investors | LinkedIn
- China’s AI+ plan and the Manus middle finger | LinkedIn
- Autopsy of an algorithm – Is building an audience still worth it these days? | LinkedIn
- AI is screwing with your résumé and you’re letting it happen | LinkedIn
- Oops! I did it again. . . | LinkedIn
- Palantir turns your life into a spreadsheet | LinkedIn
- Another nail in the coffin – AI’s not ‘reasoning’ at all | LinkedIn
- How AI went from miracle to bubble. An interactive timeline | LinkedIn
- The day vibe coding jobs got real and half the dev world cried into their keyboards | LinkedIn
- The Buy Now – Cry Later company learns about karma | LinkedIn
Portfolio Marco van Hurne 
Leave a comment