Perplexity Launches Comet: A Browser with Built-In AI Agents
Perplexity AI has launched Comet, an AI-native desktop browser that embeds autonomous agents directly into the browsing experience, enabling users to delegate multi-step tasks like form filling and purchase flows. The product enters public beta today for waitlist users.
Original sourcePerplexity AI's Comet browser moves the company's AI ambitions from search into the browser layer itself, embedding agents capable of autonomously navigating websites, completing forms, and executing purchase flows on behalf of users. Rather than operating as a browser extension or overlay, Comet integrates these capabilities natively into the browser's architecture, making agent-assisted browsing a first-class feature rather than a bolt-on.
The browser targets a category of tasks that current AI assistants handle poorly: workflows that require navigating real-world web interfaces — filling out multi-step forms, checking out on e-commerce sites, or moving through authenticated user flows. By owning the browser runtime, Perplexity can theoretically give its agents direct DOM access and persistent session context, advantages that extension-based competitors like Operator or browser-agnostic agent frameworks lack.
Comet enters public beta through a waitlist rollout, meaning broad availability is not yet here. The launch represents Perplexity's most direct challenge to Google's browser dominance — Chrome — and positions the company as a platform player rather than a search layer. It also puts Comet squarely in competition with Arc, Brave, and any browser that adds AI tooling, as well as OpenAI's own Operator product, which targets similar agentic use cases through a different delivery mechanism.
No pricing details have been announced beyond beta access, and the depth of agent reliability on real-world, adversarial web pages remains unverified. The core technical question — how well Comet's agents handle sites that actively resist automation — will determine whether this is a meaningful product shift or a compelling demo.
Panel Takes
The Skeptic
Reality Check
“This is the same category as OpenAI Operator, and Operator already demonstrated that agentic web automation breaks hard the moment a site throws a CAPTCHA, a non-standard checkout flow, or a dynamic element the model hasn't seen. Owning the browser runtime is a real architectural advantage over extension-based approaches — I'll grant that — but the gap between 'completes a form on a demo site' and 'reliably books a flight on a real airline website' is where every agentic browser product has quietly died. The tell will be whether Perplexity publishes a task completion benchmark on real-world sites, or whether the launch blog stays at the level of polished screen recordings.”
The Founder
Business & Market
“The strategic logic here is sound: Perplexity needs a surface it owns, because sitting on top of Google's browser while competing with Google's search is a losing position long-term. A browser gives them session data, usage patterns, and the ability to charge for value delivered at the transaction layer — think affiliate revenue on completed purchases, or a subscription tier that unlocks agent actions. The existential risk is that Google ships agentic browsing natively into Chrome before Comet gets distribution, at which point Perplexity is fighting for market share in a commodity browser market with no search moat to fall back on.”
The Builder
Developer Perspective
“The primitive here is a browser with a DOM-aware agent runtime baked in — not an extension, not a CDP hook, but native integration with the browsing context. That's architecturally interesting precisely because it sidesteps the permission and isolation headaches that make browser automation via extensions so brittle. What I want to know before I get excited: is there any API or scripting surface for developers to define custom agent tasks, or is this a closed product where Perplexity decides what the agents can do? If there's no programmable interface, this is a consumer product dressed up as infrastructure, and that's a much smaller bet.”
The Futurist
Big Picture
“Comet's thesis is falsifiable: the browser becomes the agent runtime, not just the document viewer, and whoever owns that runtime owns the transaction layer of the web. That bet pays off only if two things hold — agents get reliable enough that users actually delegate real tasks rather than just toy with the feature, and web publishers don't successfully block or degrade agent-driven traffic the way they've started blocking scrapers. The second dependency is underappreciated: if e-commerce sites start fingerprinting and blocking Comet's agents to protect their UX funnels, the whole value proposition collapses. The second-order effect if it works is more interesting than the browser itself: it shifts purchasing power away from Google's ad-driven discovery model toward whoever's agent completes the transaction.”