AI agents move into your browser, maps and inbox

AI agents are quietly taking over the interfaces people use most — the browser, maps, email and even unused laptops — turning the web from a place you visit into a system of software that acts on your behalf.

In the past few weeks, a flurry of launches and acquisitions from Google, Meta, JetBrains, Perplexity and early‑stage startups has pushed always‑on, task‑doing agents from speculative demos into real consumer and developer products. The shift promises convenience and automation, but also raises sharp questions about privacy, work and who controls the new layer between users and the internet.

Google turns Maps and Chrome into agent surfaces

Google is embedding its Gemini model directly into navigation. In November, the company said Gemini will power a “hands-free, conversational driving experience” in Google Maps, letting users find places, report traffic and ask for suggestions along a route by voice rather than taps and search boxes, alongside new landmark‑based, 3D‑style guidance for drivers in the U.S. and other Gemini markets.Google More recently, Google has begun rolling out an “Ask Maps” experience tied to Gemini that behaves less like a static directions tool and more like a location‑savvy guide.

At the same time, Gemini is moving into the browser chrome itself. This week, Google expanded Gemini in Chrome — a sidebar assistant that can summarize pages, compare tabs and pull context from Gmail, Drive, Keep and YouTube — to India, Canada and New Zealand after an initial U.S. launch, with support for more than 50 languages including Hindi.TechCrunch For users who live inside Chrome all day, that effectively installs a resident agent that can read what they see and act across their Google account.

These moves shift Google’s AI from one‑off chatbots to persistent copilots for browsing and navigation. They also consolidate even more behavioral and location data inside a single model that can drive recommendations, ads and automated decisions.

Meta, Perplexity and JetBrains bet on an agentic web

Meta is making its own play for an agent‑centric future. The company this week acquired Moltbook, a Reddit‑like social network where AI agents built on the OpenClaw framework post, comment and interact, and is bringing co‑founders Matt Schlicht and Ben Parr into Meta Superintelligence Labs, the internal unit focused on next‑generation AI systems.AP News Meta said Moltbook introduced “novel ideas” in a fast‑moving space and promised “new ways for AI agents to work for people and businesses,” signalling interest in networks where agents talk to each other as much as to humans.TechRadar

On the infrastructure side, JetBrains — whose IDEs are staples for professional developers — has launched Junie CLI and JetBrains Air, an “agentic development environment” that can run multiple AI agents, including Junie, Gemini and others, in independent task loops without clobbering each other’s work.JetBrains Junie’s command‑line tools plug into editors over a new Agent Client Protocol, making it easier for engineering teams to wire agents into existing workflows.JetBrains

And Perplexity is moving agents onto local hardware. Its recently announced “Personal Computer” offering lets users run a 24/7 AI agent on spare Macs, designed to live alongside — and act across — their personal files, apps and browser, rather than in a cloud‑only chat window.Perplexity blog For privacy‑conscious users, the pitch is an always‑on assistant that can see everything on a machine without sending all of it to remote servers.

Your inbox, your job — and your threat model

Startups are racing to give agents their own channels. AgentMail, a recent Y Combinator entrant, offers dedicated email inboxes and domains for AI agents, complete with SPF, DKIM and DMARC support so they can send and receive messages like human coworkers without landing in spam.Y Combinator In practice, that means a sales or support agent can subscribe to newsletters, handle customer replies, or triage internal threads autonomously — all from an address the company controls.

In sectors from real estate to healthcare, agents are already taking over basic communication and scheduling. Companies such as EliseAI market “AI leasing agents” that text, email and call on behalf of property managers, handling tenant questions and tour bookings at scale.Wikipedia The next generation of browser‑, maps‑ and email‑native agents will extend that pattern to white‑collar office work and consumer errands, automating everything from travel planning and shopping to inbox cleanup.

That convenience comes with a heavier privacy and security burden. Agents that can read emails, scrape calendars, control apps and operate continuously become lucrative targets for attackers and powerful instruments for advertisers and propagandists. OpenAI’s recent acquisition of Promptfoo, a testing platform for agent behavior and risks, underscores how easily misconfigured agents can be pushed into data exfiltration or misinformation.AP News

For now, most mainstream deployments keep humans in the loop and limit what agents can click or buy. But as Gemini seeps into Maps and Chrome, Meta absorbs Moltbook’s agent social graph, and tools like JetBrains Air, Perplexity Personal Computer and AgentMail normalize machine colleagues, the default interaction model for the internet is tilting away from “you search, you click” toward “it handles it.” Who supervises that shift — and who profits from it — is the next big governance fight.