AI in Microsoft Edge
The AI-powered mode allows Copilot to search across all your open tabs and handle tasks like booking a restaurant, and it brings the Copilot chatbot to your new tab page.
[…]
Copilot in Edge also supports voice navigation to locate information on a website or to open tabs with products to compare. Microsoft is also planning to let Copilot, with your permission, access your Edge browser history and credentials so the chatbot can book reservations on your behalf.
With Copilot Mode on, you enable innovative AI features in Edge that enhance your browser. It doesn’t just wait idly for you to click but anticipates what you might want to do next. It doesn’t just give you endless tabs to sift through but works with you as a collaborator that makes sense of it all. It keeps you browsing, cuts through clutter and removes friction to unlock your flow – all built to the highest Microsoft standards of security, privacy and performance trusted by billions of people and businesses worldwide – with you as the user always in control.
Whether this process would be quicker than submitting a search query, choosing a paddleboarding location from the results, and making an online reservation would depend upon how long the searcher considered the options and how easily the searcher could navigate the relevant sites. It also requires the AI model not to be confused by the web pages it scans to perform the tasks.
Copilot Mode in Edge can also handle other common LLM-oriented tasks like summarizing web pages and remembering browsing sessions by topic, so that these can be picked up where they were left off.
Of course Copilot Mode has privacy implications similar to those raised by browser extensions. Both potentially have access to everything going on within the browser and Copilot may get even broader data access through explicit tool use permissions. Microsoft however insists it can manage the privacy risks.
Color me skeptical about the idea that my web browser should be “working for me”, rather than serving as a tool for me to work with. The AI hype cycle is pointing to a future where automated agentic web browsers surf automated AI-generated websites.
[…]
I am reminded of the decade-ago Netflix strategy espoused by Ted Sarandos: “The goal is to become HBO faster than HBO can become us.” I think something similar is behind Microsoft trying to make Copilot front-and-center in Edge, and Google’s concurrent move to junk up Chrome with AI-generated suggestions. Their goal is to make their web browsers chatbots faster than OpenAI can make ChatGPT a web browser.
Microsoft has launched new features for its AI “Copilot Mode” in Edge, following a raft of recent new offerings from competing browsers.
[…]
Copilot Actions adds agentic capabilities, letting the AI handle tasks like unsubscribing from emails or making reservations. Actions also includes voice capabilities, allowing users to speak to their browser to open web pages or ask Copilot to find specific topics within articles.
Meanwhile, the Journeys feature organizes browsing history by topic and suggests next steps, making it easier to resume research sessions. It can also group past browsing sessions thematically.
But being able to start a conversation with a tab’s contents as context isn’t much more helpful than working directly with ChatGPT or Perplexity. As a result, I haven’t been tempted to switch to these browsers from Arc, where I can navigate fluidly among numerous websites with just a couple of clicks.
[…]
The new hotness is “agentic browsers,” which can perform tasks for you. I’ve struggled to come up with tasks that are even mildly realistic.
[…]
The Copilot chatbot in Microsoft Edge was noticeably faster than the others, even when I prompted it to think more deeply about the task and added three tabs with confirmation lists to the prompt. However, it never even came close to producing the correct results, whether I let it try the entire task at once (154, 174, 144) or asked it to count the number of minors on each list (228, 213, 181).
[…]
When it comes to system prompts, the anxious tone of Copilot’s internal responses suggests a “ship now, apologize later, if you’re caught” system prompt that, if reflected in a real-world workplace, would be problematic. Obviously, AIs don’t have feelings that can be hurt and won’t complain to HR, but such a culture tends to encourage people to cut corners and make poor decisions that compromise quality and customer service. If Copilot is any indication, the same is true for AIs.
Previously:
1 Comment RSS · Twitter · Mastodon
As awful and junked up as they have made Edge, the only compelling idea about this for me is that, theoretically, if you are paying for a licensed account you SHOULD be able to trust it with work documents and work information and have that information not siphoned into the training abyss. Which would make it far more useful if used solely for work.
It's the consumer side that worries me. I'm not using any of these AI browsers until I see how this shakes out. To me it just sounds like a security nightmare waiting to happen.
