In the world of technology, the burgeoning possibilities of artificial intelligence (AI) are sparking both excitement and skepticism. Recently, Aravind Srinivas, the CEO of Perplexity AI, shared his insights on the future of AI agents, prefacing his remarks with a dose of realism amid the heightened expectations associated with these intelligent systems. Srnivasan emphasized that while AI agents hold the potential to significantly alter workflows in various sectors, their deployment is likely further away than many enthusiasts believe.
Srinivas pointed out challenges in developing fully functional AI agents and highlighted the constraints posed by ubiquitous operating systems such as iOS. In his view, the optimism surrounding the imminent arrival of these agents may need reevaluation. “Anyone saying agents will work in 2025 should be skeptical,” he noted, indicating that various technological hurdles must be overcome before AI agents can be a reliable addition to daily tasks.
Perplexity AI is steering its efforts towards creating AI agents capable of performing tasks that traditionally require human input, such as booking travel arrangements and managing personal preferences. During a recent discussion, Srinivas explained that these agents would be tailored to individual user requirements—be it selecting a specific hotel room with a favored view or booking activities that align with the user's interests.
“We’re going to be building agents that let you book stuff, buy stuff, make travel plans, book hotels,” Srinivas elaborated. His vision for these agents aligns closely with the role of executive assistants (EAs), albeit at a significantly lower price point. Currently, human EAs can cost anywhere from $100,000 to $200,000 annually—an expense not feasible for many users. With AI agents expected to deliver roughly 80% of the value of human assistants at a fraction of the cost, Srinivas believes there is ample market opportunity. “If that’s mainly available for 80% of the value at 10 to 100x lower price, that’s a pretty good business to build,” he stated.
Another critical aspect of Perplexity AI’s strategy involves developing a bespoke browser named Comet. Announced last month, Comet is intended to enhance the functionality of AI agents, especially when navigating restrictive operating systems like iOS. By utilizing a browser, the AI agents can operate seamlessly across various applications, thus addressing challenges inherent in traditional app ecosystems. “The plan is to develop an operating system with which you can do almost everything,” Srinivas explained.
This browser would enable deeper integrations, allowing for automated tasks without requiring users to jump through hoops or repeatedly enter personal information. “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,” Srinivas said during the Upfront Summit.
The approach to AI agents proposed by Perplexity AI offers a refreshing perspective amidst a landscape where exaggerated timelines for AGI (Artificial General Intelligence) have become commonplace. OpenAI predicts that AGI could be developed before 2027, while prominent figures like Softbank’s Masayoshi Son suggest it might arrive even sooner. Srinivas challenges these predictions and emphasizes the need for tempered expectations surrounding AI agents, which currently face critical issues of security and autonomy. Dealing with sensitive tasks such as payment processing raises valid concerns regarding user security, particularly when agents access confidential data like credit card information.
Moreover, there are fundamental operational hurdles that need attention. There’s a possibility that some websites could prevent AI agents from functioning in the same manner as human users, which raises questions about the reliability and versatility of AI workflows. As such, Srinivas believes that approaching the development of AI agents with cautious optimism is critical. The immediate focus for Perplexity AI lies in identifying and implementing straightforward use cases to fortify user experience without over-promising on capabilities.
“It’s not going to work this year,” Srinivas asserted, reiterating that while the potential of AI agents is immense, the journey to their realization is a multifaceted and complex endeavor. He expressed hope that Perplexity’s innovations could ultimately lead to a more accessible and efficient digital experience for everyone, illustrating a long-term vision that prioritizes both accessibility and functionality.
Aravind Srinivas’s insights shed light not just on the operational challenges but also on the broader vision within the AI domain, positing that AI systems can be designed to alleviate the daily burdens faced by individuals and businesses alike. As industries anticipate a future characterized by innovation, it becomes paramount to evaluate technological advancements with both enthusiasm and caution.
