AI-native browsers are on the rise
Two browsers have launched within months of each other, both putting an AI assistant at the heart of the browsing experience: ChatGPT Atlas from OpenAI and Comet from Perplexity. Instead of treating artificial intelligence as a separate add-on, these applications bring the language model into the space where we read, work, and research.
This is the “AI-native browser” concept. We stop jumping between tabs, copying content into ChatGPT, and pasting the answer back. The assistant understands what is on screen and responds in real time. At the same time, early reactions show we are at the very beginning — this is still a transitional phase that needs polishing.
ChatGPT Atlas: an assistant in every tab
- ChatGPT embedded in the interface – Atlas can summarise an article, answer questions, or even book an appointment without leaving the current page.
- Contextual memory – the browser remembers what you have been working on and can resume previous threads, such as job postings from last week. All memories can be viewed, archived, or deleted.
- Agent mode – the automation feature from ChatGPT arrives in the browser but with safeguards (no installing extensions, no access to files outside the browser, pauses on banking or email sites).
- Roadmap – multi-profile support, developer tooling, and deeper integration for apps built on the ChatGPT SDK are on the way. OpenAI also hints at new standards, like ARIA tags, to help sites cooperate with agents.
Atlas is available today for macOS users on every ChatGPT plan, including Business, Enterprise, and Edu. Versions for Windows, iOS, and Android are in development, and the setup wizard lets you import bookmarks, passwords, and browsing history.
Perplexity Comet: browsing through conversation
- An assistant always in reach – Comet lets you ask follow-up questions, compare sources, and delegate tasks (emails, summaries) without leaving the page.
- Session memory – context carries across tabs and sessions, making multi-step research easier.
- Automation of common tasks – emailing, summarising, or comparing products can be triggered with natural language instructions.
- Distribution and performance hurdles – the invite-only model and slower performance cause some users to return to Chrome or Edge. Corporate environments also report blocked pages when accessed through Comet.
Comet is currently available to Perplexity Max subscribers via an invite-based rollout, with broader access planned for the summer.
The direction is clear, but we’re still in transition
Atlas and Comet illustrate that browsers are starting to:
- Share context with the assistant – AI sees what the user sees.
- Learn our habits – memory helps resume paused tasks.
- Automate key actions – the agent can walk through a series of forms and clicks.
- Force new standards on the web – semantic markup and accessibility become more important so agents can understand pages.
Yet it is hard to claim this is the moment we all switch from Chrome or Safari:
- user experience and performance need more work,
- limited availability (invites, missing Windows build) slows adoption,
- corporate IT departments are wary of approving an unfamiliar browser,
- users remain sceptical about real productivity gains.
Do we really want AI to see everything?
The deeper AI integrates into the browser, the more pressing the question becomes: are we ready to let an assistant have full visibility into our habits, history, and sensitive data? Atlas and Comet promise convenience, but each benefit implies the assistant:
- analyses the same content we see,
- keeps context from previous sessions,
- can carry out multi-step actions on our behalf.
Before handing over control, it’s worth defining boundaries: when does AI gain access to data, how often do we clear memories, which tasks do we still perform manually, and where do we genuinely need automation? Otherwise, it’s easy to slip from “helpful co-pilot” into “agent that knows too much.”
What it means for teams
- Browser makers must focus not just on rendering pages but on context management and safety.
- Site owners should prepare content for agent consumption (semantics, accessibility, clarity).
- Developers gain a new playground for integrations that tap into memory and automation.
- Users need education and trust — in the tool and in their own digital hygiene habits.
Summary
Atlas and Comet are early but significant steps toward “AI-first” browsers. They set the direction, yet they do not deliver a drop-in replacement for today’s tools. We still need progress in performance, governance, and culture around working with agents.
This is the perfect time to experiment — and to ask uncomfortable questions. If an agent is going to live inside our browser, we need to decide on the ground rules before handing over the keys.