Aravind Srinivas: Perplexity's Race to Build Agentic Search

Srinivas says Perplexity is investing in Comet, a browser that combines navigation, search, and an assistant. He imagines users running several tasks at once, with access to sources such as email, calendars, and other accounts. The browser could also let an agent work across tabs, fill forms, and handle recurring research tasks.

Perplexity began with natural-language search over relational data from sites such as Twitter, LinkedIn, and GitHub. The team saw sustained use, then expanded the idea to web search that returned answers with sources. Srinivas recalls nearly 700,000 queries on New Year’s Eve as the point when user interest convinced him the product had potential.

Srinivas expects well-funded companies to copy successful products, so Perplexity focuses on answer and task accuracy, speed, and coordinating tools. He says the browser is a harder product to copy than another chat tool, while acknowledging that OpenAI and Anthropic could build browsers too. His advice to founders is to expect competition and keep shipping.

Perplexity requires employees to use at least one AI coding tool, with Cursor common internally and some teams also using GitHub Copilot. Srinivas says these tools can turn an algorithm implementation and experiment from three or four days into about an hour. He still values software engineering, infrastructure, and distributed-systems skills, especially for production work.

Srinivas sees subscriptions, task-based pricing, and a share of purchases as possible revenue sources, though he expects Perplexity may never match Google’s profits. For accuracy, he points to better search indexing, stronger page snippets, and models that can reason through multiple steps. He also describes partnerships across hotels, maps, shopping, finance, and sports, and says distribution should expand through both overlapping and new audiences.