In a recent conversation hosted by Y Combinator General Partner David Lieb, Aravind Srinivas—co-founder and CEO of Perplexity—sat down to discuss his foray into the AI world, the evolution of Perplexity from a scrappy prototype to a serious contender in the search space, and his vision for a more user-centric future of online information retrieval. Below is an overview of their discussion, capturing key insights and lessons for anyone curious about building transformative AI-driven products.
Aravind Srinivas began his AI journey during his undergraduate years in India before moving to the United States for his PhD at UC Berkeley. A pivotal internship at OpenAI exposed him to cutting-edge thinking on large language models (LLMs), unsupervised learning, and the pursuit of Artificial General Intelligence (AGI). This experience helped him realize that generative AI could fundamentally reshape core areas like search—arguably one of the most ubiquitous gateways to information in our daily lives.
Perplexity started with a bold ambition—to build an AI-powered search experience that could rival the depth and utility of Google. Initially, however, there was a lot of exploration:
Early public demos of Perplexity allowed users to quickly see how generative AI could provide concise, source-backed answers to queries. A few key features propelled user engagement:
A standout principle guiding Perplexity’s product design (one originally championed at Google) is that the user should never be blamed for an unhelpful result. Instead, the AI must clarify ambiguous questions, handle typos gracefully, and deliver more intuitive experiences. This mantra underpins Perplexity’s approach to product management, ensuring that every bug or mishap serves as a clue to improve the service, rather than an excuse to place more burden on users.
As Perplexity’s team grows, balancing speed of innovation with product reliability has become a top priority. Srinivas discussed how excessive processes can kill agility, but that some structure is necessary to keep bugs and regressions in check. A few of Perplexity’s strategies include:
Looking ahead, Srinivas believes AI’s transformative potential lies in orchestrating multiple capabilities under one roof—traditional knowledge graphs, specialized widgets, and language models—for a fluid, end-to-end user experience. This could mean, for example, allowing someone to discover the best product to buy, check real-time information like flight schedules, and act on that information (e.g., purchase or book) without leaving the AI-driven environment.
But the search landscape is competitive. Established players like Google and Microsoft Bing have massive user bases and significant resources. Aravind sees Perplexity’s edge in its dedication to product detail, user-centric design, and a culture of pushing constant, incremental improvements—all without the baggage of legacy ad-driven models and corporate inertia.
Perplexity’s story exemplifies the power of unwavering focus on user experience, combined with the innovative potential of rapidly advancing AI models. As Srinivas puts it, the true challenge is not just building smarter language models, but packaging that intelligence into a seamless product that learns, clarifies, and ultimately empowers users.
