Back
Perplexity AILaunchPerplexity AI2026-05-30

Perplexity Launches Comet, an AI-Native Browser with Agentic Automation

Perplexity AI has launched Comet, a macOS browser that embeds an agentic AI assistant powered by its own LLM to autonomously complete multi-step web tasks, manage tabs, and summarize research in real time. A Windows version is planned for Q3 2026.

Original source

Perplexity AI has officially released Comet, a desktop browser built from the ground up around an integrated agentic assistant rather than layering AI onto an existing browsing architecture. Unlike browser extensions or sidebar copilots, Comet's assistant has native access to the browsing context, enabling it to execute multi-step tasks — filling forms, navigating between pages, synthesizing information across tabs — without requiring the user to manually hand off control at each step.

Comet runs on Perplexity's own LLM rather than routing through a third-party model provider, which has implications both for latency and for the company's unit economics. The browser is currently available for macOS, with a Windows release on the roadmap for Q3 2026. The agent can manage tab state intelligently, meaning it doesn't just open links but tracks context across sessions to support longer research workflows.

The launch puts Perplexity in direct competition with the browser layer itself — not just search or summarization tools. The company is betting that the browser is the right abstraction level for agentic AI: it already has permission to see every page, every form, and every click. Comet is the argument that controlling the shell, not just the extension slot, is what makes autonomous web tasks actually reliable.

This is Perplexity's most direct challenge yet to both Google and to general-purpose AI assistants like ChatGPT that operate above the browser layer. Whether Comet can deliver on the agentic promise at real-world task complexity — messy login flows, CAPTCHAs, inconsistent site structures — is the question that no demo has answered yet.

Panel Takes

The Skeptic

The Skeptic

Reality Check

The agentic browser category already has Arc, Dia, and the entire graveyard of 'AI browser' announcements from the last 18 months — Comet needs to show it survives contact with real-world web entropy, not just clean demo sites. The specific failure mode I'm watching for: multi-step tasks that break on any site using non-standard auth flows, CAPTCHAs, or dynamic JS rendering, which is most of the web that actually matters. Perplexity's 12-month kill scenario isn't Google — it's their own LLM not being good enough at grounded web tasks to justify switching browsers, at which point this is just a worse Chrome with a chatbot.

The Founder

The Founder

Business & Market

The strategic logic here is actually coherent: owning the browser means owning the session, the data, and the distribution channel without paying a platform tax to Google or Apple — that's a real moat if they can execute. Running their own LLM instead of routing through OpenAI or Anthropic is the margin play; every agentic task completed in-browser at proprietary model cost is a unit economics story that wrapper businesses can't replicate. The risk is consumer browser switching costs are brutal, and 'macOS only until Q3' means they're launching half a product into a market where the incumbent is free and already installed.

The Futurist

The Futurist

Big Picture

Comet's thesis is falsifiable and specific: the browser is the correct permission boundary for agentic AI because it already has ambient access to the full web context, making it more reliable than agents that operate via API or screenshot heuristics. The second-order effect that nobody is talking about is what this does to web analytics and advertising — if agents are completing tasks on behalf of users, the signal layer that the entire ad-targeting stack depends on collapses, and Perplexity ends up holding the only legible view of user intent. This bet is on-time, not early: the trend toward ambient compute with persistent context has been building for two years, and Comet is the first browser-layer product with a credible LLM stack behind it.

The PM

The PM

Product Strategy

The job-to-be-done Comet is hired for is 'complete this web task without me having to babysit every step' — that's a clean, singular hire with genuine demand, and the browser shell is a more complete solution to that job than any extension-based approach I've seen. The completeness question is the critical one: can a user actually replace Chrome with Comet today, or does 'agentic task completion' work on 60% of workflows while the other 40% silently fail mid-task? The macOS-only launch isn't a soft launch problem, it's a completeness problem — half the research workflows Comet is targeting involve cross-device handoffs that don't exist yet.

Bookmarks

Loading bookmarks...

No bookmarks yet

Bookmark tools to save them for later