Photo by Erik Mclean on Pexels
Microsoft Edge Copilot Mode (with Copilot Actions and Journeys)
Edge's opt-in Copilot Mode is one of the more sensible takes on the AI browser: multi-tab reasoning, resumable Journeys and agentic Actions bolted into a browser you may already use. It's still a staged, US-first preview, so treat the rating as provisional.
Pros
- + Opt-in, toggleable and layered into an existing browser rather than a separate product
- + Assistant can read and reason across multiple open tabs, not just one page
- + Journeys give resumable context for long-running research
- + Agentic Actions require explicit permission and visible consent
- + Actions bounded to open tabs and current session state
- + Genuinely useful chores targeted: unsubscribes, form filling, price comparisons, bookings
Cons
- − Shipped as a staged, experimental, initially US-limited preview
- − No published performance, reliability or model details to verify Microsoft's claims
- − Privacy trade-off is real: the assistant reads open tabs and can analyse browsing history
- − Security, privacy and 'anticipates your next step' claims are vendor assertions only
- − Journeys and history analysis reporting is second-hand and lower confidence
- − No pricing or availability specifics supplied, making value hard to judge
Rather than shipping a standalone "AI browser" and asking you to migrate your bookmarks, Microsoft has taken the quieter route: Copilot Mode is a toggle inside Edge that swaps the new-tab page for a combined search-and-chat box and parks a persistent assistant alongside your browsing. Turn it off and you're back to ordinary Edge. That restraint is the most immediately likeable thing about it.
The headline capability is that the assistant can actually see what you're doing — reading and reasoning across your open tabs, summarising pages, and (as of the October 2025 update) organising related activity into Journeys you can come back to later. Copilot Actions goes further, letting the browser act on your behalf with permission: clicking through pages, filling in forms, chaining steps for things like unsubscribing from newsletters, pulling together price comparisons, or starting a booking.
This is aimed at the tab-hoarders and researchers — people who end up with thirty open pages and no memory of why. If you keep a tidy five-tab browser and dislike assistants reading your screen, there is nothing here you need. Worth noting up front: this arrived as a staged, initially US-limited preview, so availability is the first hurdle.
Conceptually, this is one of the better-judged AI browser efforts we've seen described. Making it opt-in and reversible, keeping Actions bounded to open tabs and current session state, and routing agentic behaviour through visible consent flows are all the right instincts — an agent that can click buttons and fill forms on your behalf is exactly the sort of feature that should require you to say yes out loud. Journeys is the idea with the most obvious everyday payoff: resumable context is a real problem that bookmarks have never solved well.
The caveats are substantial, though, and most of them come down to maturity. Microsoft's own framing is experimental and staged, which means feature availability, reliability and polish are all moving targets. We also only have the vendor's word on quality: there are no published performance figures, no detail on which model does the reasoning, and no stated limits on how far Actions can go before it needs you. Claims that Copilot "anticipates what you may want to do next" and meets Microsoft's security and privacy standards are marketing posture, not verified behaviour. And the trade for all this is that an assistant is reading your tabs and, for Journeys, your browsing history — a genuine privacy decision, even with the controls in place.
Because there's no pricing information and no measurable specs to evaluate, this is an assessment of the design and the promise rather than a tested verdict on how well it performs. On that basis it earns a cautious recommendation: the architecture and consent model are sound, the use cases are real, and the cost of trying it is close to zero if you already run Edge. We'd want hands-on reliability data on Actions before going higher than a 3.5, and anyone outside the preview regions is simply waiting.