Why Perplexity is building Comet browser: Aravind Srinivasan on AI agents and app control

Last month, Perplexity AI announced that it is working on its own browser, Comet, a move that, according to Aravind Srinivasan, is essential for creating AI agents with enough control to function effectively—especially within restrictive ecosystems like iOS.

“The plan is to do develop an operating system with which you can do almost everything,” Srinivasan explained, highlighting why Perplexity is focusing on this approach. Unlike standalone AI apps, which are often limited by platform restrictions, a browser allows AI to operate seamlessly across different applications, facilitating deeper integrations and better automation.

The goal is to build AI-powered agents capable of handling everyday tasks—booking hotels, making purchases, filling out forms, and managing workflows—without requiring users to jump between multiple apps. “We are trying to figure out simple use cases like buying stuff without having to enter credit card details everywhere or manually filling in shipping information,” Srinivasan said at the Upfront Summit. The focus on simplifying workflows is a key part of Perplexity’s strategy. 

A browser-based AI assistant also solves a critical issue on platforms like iOS, where strict sandboxing prevents third-party apps from interacting freely with one another. By operating within a browser, Perplexity can bypass these limitations, ensuring its AI agents have the necessary access to streamline digital tasks.

Srinivasan is realistic about the timeline, cautioning that fully functional AI agents won’t arrive immediately. “Anyone saying agents will work in 2025 should be skeptical,” he admitted. However, Perplexity’s long-term vision is clear: to create an AI-powered browser that can act as a true digital assistant, reducing friction in online interactions and making everyday tasks more efficient.

This approach seeks to emulate the functions of an executive assistant but at a fraction of the cost, potentially providing significant value to users who cannot afford traditional assistants. “Personal EAs is like 100 to 200 thousand dollars a year, and so if that’s being made available for 80 percent of the value at ten to 100 x lower price, that’s a pretty good business to build,” he said. 

Source: https://www.businesstoday.in/