AI's Real Battleground Is User Experience, Not Model Superiority
"In my 2.5 years of running Perplexity, one thing I have realized is no one has a long-term model advantage. The product is the ultimate place to compete on," declared Aravind Srinivas, co-founder and CEO of the AI search company challenging Google's dominance.
End of Miles reports that Srinivas believes the true competitive edge in AI lies not in having the most advanced language model, but in building superior products and experiences around those models.
The leveling effect of open source
During a candid Reddit AMA (Ask Me Anything) session, the Perplexity chief explained why he remains unfazed by competitors with seemingly more advanced AI models. His perspective challenges the conventional wisdom that proprietary AI models like GPT-4 or Claude provide sustainable competitive advantages.
"Open source models always keep catching up or improving on the closed ones, with better cost-effectiveness and speed and customization." Aravind Srinivas, Perplexity CEO
This observation comes as Perplexity reveals it's currently A/B testing a post-trained version of DeepSeek-v3 that appears to outperform GPT-4o, highlighting the company's strategy of enhancing existing open-source models rather than developing their own from scratch.
Perplexity's advantage strategy
Rather than competing in the resource-intensive race to build foundational models, Srinivas outlined Perplexity's approach to gaining an edge in the competitive AI landscape.
"We have a pretty good post-training and inference team... We will be post-training on top of open source models, and making them cheaper. We will be top-of-the-market for search-grounded LLMs in terms of accuracy, speed and costs." Srinivas
The Stanford AI researcher turned CEO emphasized that Perplexity isn't trying to have "the best model at any point," but rather developing "the ability to post-train and improve upon anything that exists out there within a few months."
Why product experience trumps model performance
When asked about competition from tech giants integrating AI search capabilities, Srinivas remained confident in Perplexity's approach of focusing on user experience rather than raw model capabilities.
The entrepreneur highlighted how this strategy shapes Perplexity's roadmap, with investments in specialized capabilities like browser agents that leverage their unique position in the market. "We are also investing in post-training for browser agents - something that will be unique to us and matter more to us over time," he noted.
"We also won't be doing pre-training. We will be post-training on top of open source models, and making them cheaper." Srinivas
The shifting AI landscape
Srinivas' comments reflect a broader shift in how AI companies approach competition as the technology matures. With multiple capable models now available, differentiation increasingly comes from how these models are implemented and the quality of the user experience built around them.
This perspective aligns with Perplexity's recent moves to expand beyond traditional search into browser integration, mobile assistants, and specialized vertical applications – all areas where user experience design becomes as critical as the underlying AI performance.
"If the progress in AI reasoning models continues, search itself might be more for the curious minded. And people are going to be using AIs largely for workflows and tasks which will require plenty of search in the loop, but will move to the abstraction of asking AIs to do work for you, instead," the CEO concluded, pointing to a future where AI assistants handle increasingly complex tasks.
