What Edge Copilot Mode Says About the AI-Powered Future of Web Browsing (Premium)

My recent experiences with Perplexity Comet were eye-opening. But there are a few other AI web browsers out there, like The Browser Company’s Dia and Microsoft Edge in Copilot Mode, and others on the way, like Opera Neon. Where do things stand today with these products, how do they compare with Comet, and is it reasonable to assume that all browsers will eventually work similarly when it comes to AI functionality?

Let’s start with Microsoft Edge and its new Copilot Mode.

?‍✈️ Copilot Mode

Copilot Mode is, for now, the least ambitious of the AI browser options. But to be fair to Microsoft, it’s still a big step forward from the AI experiences we see in other popular, mainstream browsers like Chrome, Safari, and Firefox. And it’s still early: Microsoft describes Copilot Mode as experimental, and as a “step toward building a more powerful way to pilot the web.” That language is interesting. It suggests that a more sophisticated experience is coming, but also that the current “copiloting” of Edge will give way to a new AI-centric browser, if not a completely different AI-centric experience.

But first things first. In its initial form, Copilot Mode is an experience you can toggle on and, if desired off. Assuming you’re on the latest Edge version, you will be prompted to do so. But you can find the controls for Copilot Mode under the new “AI innovations” page in Edge settings.

When you enable Copilot Mode, the Edge user interface changes ever so slightly: The Copilot button moves from the far right side of the Edge toolbar to the far left side of the address bar, and so this button has changed in Edge almost as much as the Taskbar-based Copilot shortcut in Windows 11.

You’re also given a quick tour of Copilot Mode that describes its major new features and changes. They are:

New modern homepage. This is a minimalist New tab experience, actually, that has the same light tan look and feel as the Copilot app in Windows 11 and is an oasis of calm compared to the default New tab experience in Edge. But I typically use a free extension called Bonjourr—which is similar to Momentum–as a New tab page in Edge and other browsers, and so using Copilot Mode overrides that. More on this homepage (New tab page) below.

Quick assist. If you use Copilot in Windows 11, you know that it wants to enable a Quick assist feature that lets you quickly access Copilot in a minimal window as you work by typing Alt + Space. Quick assist is a bit like that, but in the browser, and it appears when you click the newly relocated Copilot button or type Alt + C while in the browser. This is a simpler way to access Copilot while you browse so you can do things like summarize the current webpage, document, or video, or anything else you may do with Copilot. But if you like the old Copilot pane better, you can toggle that on instead.

Simple task handoff. Copilot Chat in the (new) New tab supports a new Actions mode that is to Copilot on the web what App actions are to Copilot and the other AI features in Windows 11. According to Microsoft, Copilot Actions include such things as making a reservation, shopping online, formatting a document, and compiling data, whatever that means. “It’s perfect for busy days when you want to delegate repetitive or manual steps,” it says.

Voice navigation. Is the case with Copilot elsewhere and other AI chatbots, you can interact with Copilot in Edge using your voice instead of typing. To do so, you have to click a microphone in Quick assist or the New tab interface. But there doesn’t appear to be a keyboard shortcut (at least not yet), as is the case with Copilot in Windows 11. (There, you can long-press Alt + Spacebar to enter Voice mode.) You can at least choose the Copilot voice you want.

Let’s take a closer look.

? The New tab page in Copilot Mode

We all work differently, but the lack of customization in this New tab page will be limiting for many. It consists of a large Copilot Chat interface and a handful of Quick link buttons for sites you frequently visit. You can’t add or remove Quick link buttons, Edge does that for you.

Copilot Chat will look familiar, but it’s missing a few features from Copilot in Windows 11, including the ability to upload files. In place of a model picker, which will finally go away with GPT-5, hopefully, Copilot Chat provides choices like Search & Chat (the default), Ask Copilot (Quick response in the Copilot app), Think Deeper (as in the Copilot app), and Action, which changes the user interface and is discussed below.

If you sign in, you’ll get a new choice, Smart (GPT-5), that is obviously tied to this week’s release of the new OpenAI GPT-5 models. In this mode, Copilot Chat will “think deeply or quickly based on the task,” meaning that it will choose the correct model as needed, rather than force you to choose one. So we can expect this interface to get simpler in the future, I think.

The most interesting thing about this New tab page is that it puts the mouse cursor/focus on Copilot Chat and not the Address bar. That is, if you type Ctrl + Tab and then start typing, as one does currently to initiate web search from the Address bar, you will be typing in Copilot Chat instead, to search or create whatever prompt. This bothered me at first, but if you do want to go straight to the Address bar, you can just type Ctrl + L (or Alt +D), as always.

But the idea here, as with Comet, is to present an AI-centric interaction rather than a search engine interaction that results in a list of blue links. With Comet, the Perplexity AI is accessible directly from the Address bar, which is more natural/normal to me, and you can choose to search Google or whatever default search engine you have via the standard Address bar-based dropdown. But in Edge, Copilot Mode is, well, a mode. And so this happens from Copilot Chat in New tab. That is, you can interact with Copilot, go to a specific URL, or search the web from there instead of the Address bar. (And if you do type in the Address bar directly, it works as before.)

This was likely a design choice, with the idea being that Microsoft’s more mainstream users would find changing the Address bar to be a bit too much, whereas Comet is more of a power user-type app. But it also feels transitionary, and I would be surprised if direct Copilot access from the Address bar didn’t happen at some point. Plus, Copilot Chat works much like Comet in that you can choose between search and Copilot in its dropdown should you need or want to.

As noted, you can’t really customize the New tab page’s Quick links, you can just turn them off. B

ut you can also keep using Copilot Mode but without this New tab page. To do that, open Edge settings and navigate to “Start, home, and new tab page.” Then, turn off the option “Copilot new tab page.” And, if you were using a custom New tab page extension like me, you can re-enable that in the Extensions interface. Then, you can just access Copilot via the Quick assist UI. I will definitely make those changes in time–this reminds me in some ways of default Windows 11 Taskbar buttons like Quick view and Search that can be removed once you learn the keyboard shortcuts–but am leaving the default UI on for now.

➕ Quick assist

Quick assist is the new Edge sidebar, or Copilot pane, and I kind of prefer it, though the term Copilot is suggestive of the side-by-side interface that the Edge sidebar/Copilot pane currently provide. And if you’re a heavy user of this functionality today, its shift to the top left of the browser window may be disconcerting.

You can always reenable the pane, of course. To do so, open Quick assist and then click the “Open in side pane” (pin) button in that UI. Now, it appears as before, and clicking the Copilot button or typing Alt + C will toggle this pane going forward.

From a usage perspective, I don’t believe Quick assist adds anything unique. It’s Copilot in the browser, as before, but with a new UI. And that new UI does cover up part of the page you’re viewing, come to think of it. This might be annoying if you’re interacting with that page in some way, but you can always toggle it on/off as needed–I latched onto that Alt + C shortcut pretty quickly–and you might argue that things like a summary don’t require you to view the full page anyway. But again, that sidebar is an option if you want it.

From a configuration perspective, Quick assist lets you choose from the several available Copilot voice, all vaguely annoying, and see your tracked product prices (described later). But that’s about it, aside from toggling between the normal pop-up UI in the top-left and the old-school sidebar pane.

? Actions

Copilot Mode’s New tab page and Quick assist are mostly just subtle UI changes. And the addition of GPT-5 support is happening across the Copilot stack, so to speak. But Actions are where Copilot on the web starts to get interesting. And this is this browser’s only chance, for now at least, to offer functionality that we see now in Comet and will see, soon, pretty much everywhere. This is the type of thing that will chip away at the very notion of “browsing.”

That Microsoft calls this “simple task handoff” speaks to the nascent nature of this functionality, of course. Technically, the first Copilot action in Edge was Voice navigation: With this feature, you can turn on use natural language to control the browser in basic ways, at least once you toggle the microphone on, which can’t be done by voice. You can say things like “open a new tab,” “navigate to Apple.com,” and “Open the MacBook Air page in a new tab,” and that all works normally, though the feedback (“OK, opening the MacBook Air page in a new tab”) is annoying and unnecessary.

Beyond that, I figured setting up at least one price tracker would be an obvious place to start with Actions. So I asked Copilot in the browser, via voice, whether it could track the price of a product for me. It told me no, it could not do that, but it could show me the current prices. So I disabled voice control and tried just typing, starting with “track the price of a Sonos Era 100 speaker.” Quick assist displayed some information about this product, and at the end of that, it asked if I “would like help setting up a price tracker” or compare this to similar products.

I was about to prompt it again when I realized a mistake: Quick assist, and thus Copilot, was still in a normal (Quick response) mode. So I chose the model switcher–that’s what this really is–and saw that there’s no “Action” choice. That’s only in the Copilot Mode New tab page.

As a quick aside, I feel like these things should always just work. That is, I should be able to ask with voice or via Quick assist, and “where” I am shouldn’t matter. Moving on.

In the New tab, I chose “Action” from the model picker there. As noted above, this changes the UI a bit, so that you get a dropdown with suggestions like “Book a vacation rental,” “Order a bouquet,” “Reserve a hotel room,” and “Book a cruise.” Those seem overly-ambitious to me, but I started with the tracker. With the prompt, “Create a price tracker for the Sonos Era 100.”

The first time I tried this, it failed. It also failed the second time, because Copilot had gone back into “Quick response” mode. But then I put it back on “Action” and tried again. And this time, it opened up a side pane inside the New tab page asking me if I’d like to use CamelCamelCamel for Amazon, Honey, or Keepa. And whether I had a preferred website or pricing criteria.

I responded with, “Track on Amazon.com and I’d like to know when the price is under $150.” And it said it would notify me when this condition was met if I confirmed the action.

So I said yes. And what happened next was quite interesting.

Right off the bat, let me just say that it failed, and repeatedly. Also that controlling the user interface of a browser window is temporary, that the future of this type of interaction will be the AI interacting not with a webpage but rather with back-end services. But that as an interim step, it’s fascinating that it can even do this, albeit somewhat poorly.

Here’s what happened. Edge opened what is clearly a remote session of an other instance of Edge, probably running in a Microsoft datacenter. And then it proceeded to interact with that webpage using a cursor to select items, and then scrolling, opening tabs, and entering text in fields as needed. In short, it navigated to Amazon.com, selected the site’s search box, typed “Sonos Era 100.” clicked the search button, and loaded the correct product page.

Then, it opened a new tab and navigated to CamelCamelCamel to initiate a price track. This is where it went south: CamelCamelCamel threw up a Captcha to prove that the browser was controlled by a person, and Copilot asked me whether I would complete it to continue.

So I clicked the “Take control” button and clicked the Captcha using the cursor in the remote view. Then I gave control back. It didn’t take, so Copilot tried itself, which is hilarious, but that failed too. Over multiple attempts, CameCamelCamel just refused to fall for this ruse. And so Copilot said it would “look for a Captcha-free price tracker.” Also hilarious.

It tried Keepa and another tracker, I lost track (ahem), failing each time because of a Captcha. And then it actually tried CamelCamelCamel again because AI is stupid and doesn’t understand what they say about trying the same thing over and over again and expecting different results. In any event, I put a stop to this silliness. Time elapsed was over 20 minutes. During which I could have easily created this tracker myself a few times over.

Please don’t misinterpret that. This will work. It’s early, and to be fair here, Action is described as a lab feature of a Microsoft Edge feature, Copilot Mode, that’s described as experimental. But this experience serves to highlight a few salient points. The current delta between dream and reality in this space. Microsoft’s almost reckless need to advance its AI functionality regardless of its current efficacy. And the competitive nature of this market, where Microsoft seemingly felt compelled to ship this in its current state because of competition real (Comet), imagined (the rumored OpenAI browser), or about to appear (Dia for Windows).

❓ A few final thoughts … or questions

This experience was, in some ways, amusing. What works now is hardly essential or all that different from using Microsoft Edge with the Copilot sidebar normally. What doesn’t work–Actions–is the meat of the matter, so to speak, the thing we need to keep an eye on. And then watch as it evolves beyond screen scraping and remote browser control into something more sophisticated. That’s inevitable. But the questions this raises, around timing and competitors trying to do the same things, are key. It’s not there yet. It will be. But when? And will we already be using Chrome (most likely) or a relative unknown like Dia or Comet instead by then anyway? What happens when OpenAI inevitably enters this space?

We can’t answer these questions yet, of course. All we can do is wait and watch. In the meantime. I will take a similar look at The Browser Company’s Mac-only Dia and see how that compares as well.

More soon.