Fireside Chat, Interview
Aravind Srinivas: Perplexity's Race to Build Agentic Search
- Perplexity plans to rebuild its infrastructure over the next several months to daily scale to handle "10x" usage growth, operating in a high-velocity mode to overcome ongoing daily infrastructure constraints.
- The company is shifting its strategic focus from a search tool to an omnibox browser acting as a "cognitive operating system," with mobile browser versions planned for launch in "many months" to enable parallel asynchronous tasks like processing emails and calendars.
- Perplexity expects to grow subscription revenue to "at least... a few billions a year" and anticipates "usage-based pricing" for task completion as a significant, potentially higher-volume but lower-margin revenue driver normalized against human labor costs.
- The business strategy relies on building brand equity and narrative to secure a "right to survive" against well-funded competitors like OpenAI, Anthropic, and Google, who are expected to launch similar browser-based agents.
- Perplexity intends to expand partnerships with providers like Trip Advisor, Yelp, and Shopify to support AI agents in performing complex transactions, such as hotel bookings and shopping, though transactional revenue is noted to have historically lower margins than CPCs.
- The company aims to maintain a competitive advantage through superior accuracy, "multi-step reasoning" for complex queries, and improved web snippet capture to reduce hallucinations, leveraging the browser to create sticky network effects via stored history and agent memories.
- Long-term outlooks predict that within "five to 10 years," the web's power law will skew further, with well-known brands preserving organic visits while SEO-gaming sites struggle, and search behavior will shift so users "first go ask this app" before traditional engines.
- Future revenue models may include "transaction-based revenue" where the AI takes a cut of purchases, alongside "usage-based pricing" for recurring tasks, though the company acknowledges users may distrust ad-based AI and expect neutral alternatives.
- The browser approach is positioned to solve OS restrictions on mobile by acting as a universal agent layer capable of scraping data and performing tasks without requiring third-party "MCP servers," allowing generic operation across the web.
- Strategic risks include the difficulty of engineering mobile browsers, the potential for competitors to replicate features, and the challenge of establishing in-app network effects given that users can easily export chat history, countered by the browser's data stickiness.
- Perplexity anticipates that as coding tools improve, enduring value will depend on brand, speed, and accuracy rather than feature replication, with the browser serving as the necessary evolution to capture billions of daily queries similar to how Chrome succeeded Google.
- The company expects the web to become "quieter" with reduced traffic to individual websites, requiring a focus on direct organic visits for specific brands while users increasingly rely on AI for information and task execution.