The Rise of AI Agents: Perplexity’s CEO Defines the Next Web Battleground

Generative artificial intelligence is rapidly reshaping how we interact with technology and information. Among the companies at the forefront of this transformation is Perplexity, an AI-powered answer engine that has emerged as a notable alternative to traditional search methods. By tapping into the capabilities of AI, Perplexity aims to provide users with direct, synthesized answers rather than just lists of links.

Since gaining prominence, Perplexity has seen significant growth, indicating a strong user appetite for AI-driven information access. The service currently handles hundreds of millions of queries each month. This rapid rise has also brought scrutiny, including legal challenges regarding alleged copyright infringement and accusations of generating content too similar to existing articles or even fabricating information. Despite these challenges, the company continues to grow, reportedly seeking investment that could value it substantially in the market.

Perplexity is actively expanding its presence, developing AI assistants for mobile devices and even working on its own web browser. Recent partnerships, such as bundling Perplexity with new smartphones from a major manufacturer and integrating with payment platforms for easier transactions, signal a strong push towards embedding AI assistance into daily digital workflows. Discussions with other leading device makers further underscore the ambition to make AI assistants widely accessible.

A key figure driving this vision is Aravind Srinivas, the CEO of Perplexity. In recent discussions, Srinivas outlined his perspective on the evolving digital landscape, identifying AI agents as the critical frontier. For him, these intelligent assistants are not just niche tools but represent the fundamental next step in how people will engage with the web and their devices.

The Agentic Future: More Than Just an App

Srinivas posits that AI agents are poised to become the “killer app” for virtually everything online. This perspective goes beyond viewing agents as limited to specific tasks like e-commerce. Instead, he sees agents as facilitators that can tailor user experiences to individual preferences. Some users might prefer to deeply research and shop, while others might want the agent to handle the entire process. The power of agents, according to Srinivas, lies in their ability to adapt to this spectrum of needs, providing the optimal experience for each person.

However, the notion of autonomous agents taking actions, especially involving transactions, raises immediate questions about reliability and error handling. What happens if an agent makes a mistake, such as purchasing something unintentionally? Srinivas suggests that the digital economy, much like historical markets adapting to new technologies, will evolve. Merchants and consumers will learn and adjust to the capabilities and limitations of agents. He stresses that, like any successful technology throughout history, AI agents must prioritize security and have robust mechanisms for resolving errors. These foundational principles, he argues, are non-negotiable for the widespread adoption and trust in agentic systems.

Integrating AI into Mobile Experiences

A significant part of bringing AI agents into widespread use involves deep integration with personal devices. Partnerships with device manufacturers, such as the collaboration to bundle Perplexity with millions of smartphones, are seen as crucial steps.

This kind of partnership is important because it offers a native, seamless way to introduce trustworthy AI capabilities to a vast audience. By being readily available on devices out of the box, more people can experience the enhanced possibilities that AI-powered search and assistance can offer beyond traditional methods.

While Perplexity is focusing on building its AI assistant and answer engine rather than developing its own hardware, being pre-installed on devices from major brands is a powerful distribution strategy. As device manufacturers increasingly incorporate various AI assistants, Perplexity aims to differentiate itself through its core strengths.

The key differentiator, according to Srinivas, is accuracy and trust. In a world where multiple AI assistants might reside on a single device, the reliability of the information and actions provided becomes paramount. An assistant that frequently errs is not truly useful, and one that is misleading or overly eager to please becomes manipulative and potentially dangerous. Srinivas argues that inaccurate AI has a negative compounding effect, where small errors can lead to larger issues or a complete loss of user confidence. In contrast, Perplexity has consistently focused on developing AI models and assistants centered on accuracy and verifiability, aiming for a positive compounding effect as trust builds over time.

Navigating the Landscape of Accuracy and Criticism

Despite the stated focus on accuracy, Perplexity, like other AI services, has faced criticism regarding issues such as generating incorrect information or closely paraphrasing copyrighted content. Srinivas acknowledges this criticism, viewing it as an essential part of the continuous improvement process. He contends that while errors occur, they represent a small fraction of results compared to the noise and SEO manipulation found in traditional search engine results pages.

“We welcome this criticism, because it’s the best way for us to continually improve.” - Aravind Srinivas

While specific data on error rates was not provided, the company indicates that reliability is a constant area of focus and is steadily improving. The feedback received from users and critics is seen as indispensable in the ongoing effort to enhance the system’s accuracy and build trust.

Regarding the use of information from copyrighted news articles, particularly in features that synthesize content, Perplexity’s stance has been previously articulated. The company points to its technical approach, including adherence to standard web protocols like robots.txt, which website owners can use to specify how web crawlers should interact with their content. This technical compliance is presented as the company’s response to concerns raised by publishers.

The Crucial Hurdle: App Accessibility

A significant technical and strategic shift involves enabling AI to take actions beyond simply answering questions. This transition from providing information to executing tasks is what truly defines an “agentic” system. Whether called agents or assistants, their core functionality lies in their ability to sequence different tools and perform actions on behalf of the user. This is a driving force behind Perplexity’s development of its own browser and mobile assistants.

A major challenge in realizing the full potential of AI agents is their ability to interact seamlessly with other applications on mobile platforms. Srinivas highlighted the complexities of this issue.

Comparing major mobile operating systems:

• iOS: While the Apple SDK provides access to core native apps like Mail, Calendar, and Reminders, allowing agents to draft emails, schedule meetings, and set reminders relatively easily, it is particularly restrictive when it comes to accessing and interacting with third-party applications. Agents on iOS often can only link out to specific content within an app, requiring the user to manually complete the action there.
• Android: Android’s platform is not fundamentally different in that AI also faces limitations accessing most apps. However, Android’s SDK allows third-party app developers to build their applications in a way that makes certain functionalities accessible to external systems, including AI agents. This structural difference can enable more direct interactions, such as an Android-based agent being able to display a song within Spotify, whereas on iOS, it might only be able to provide a link that the user must then open and play manually.

This indicates that the constraint isn’t solely the intelligence or capability of the AI model itself, but also the willingness and technical implementation of app developers to create APIs (Application Programming Interfaces) that allow AI agents to access information and execute actions within their apps without requiring the user to open the application directly.

Imagine the potential if widely used services like ride-sharing apps, food delivery platforms, or reservation systems offered robust APIs to AI agents. An agent could access real-time pricing and availability (e.g., checking if a premium ride option is only slightly more expensive), or quickly filter options and initiate an order based on stored user preferences, bypassing the need to navigate through the app’s interface manually. This seamless integration could significantly simplify tasks, offering a much faster and more direct user experience compared to the current process of searching within one app and then switching to another to complete a task. The future development of AI agents is therefore heavily reliant on how readily app makers open up their ecosystems.

The Browser Wars and the Future of the Open Web

Beyond mobile platforms, the traditional web browser remains a critical piece of infrastructure for AI agents. Perplexity is developing its own browser, partly to provide users with a reliable front end to perform tasks themselves if the AI agent doesn’t meet their needs. This also allows the system to learn from user interactions and improve over time, much like autonomous driving systems learn from human interventions.

This focus on the browser led to a notable statement from Srinivas regarding the potential future of the dominant browser, Chrome. While not expressing a desire to acquire Chrome, he stated that if regulatory actions were to force Google to divest the browser, Perplexity would be open to the possibility of running it. However, he strongly believes Google should not have to divest Chrome, citing its close ties to the open-source Chromium project, which Google manages effectively and which forms the basis for other browsers like Microsoft Edge and Brave.

The prospect of Chrome changing hands prompted a pointed critique regarding other major players in the AI space. Specifically addressing the possibility of another company, OpenAI, taking control of Chrome, Srinivas expressed significant concern.

“Giving ownership of Chrome or Chromium to a company like OpenAI would be a disaster, because open source and OpenAI are an oxymoron at this point.” - Aravind Srinivas

This statement highlights a perceived conflict between the principles of open source development, which Chrome/Chromium largely represent, and the perceived more closed nature of certain AI companies. While OpenAI has announced intentions to release open-source models in the future, the historical and current perception held by some is that their core operations and most advanced models are proprietary. Handing control of a foundational piece of web infrastructure like Chrome to such an entity is seen as a potential threat to the open web ecosystem.

Looking at other potential stewards for Chrome, Srinivas was also critical. He suggested that Microsoft might “spoil” Chrome, drawing a parallel to the development of its own browser, Edge. Transferring control to Meta was also viewed negatively, essentially swapping one large company for another that faces its own accusations of monopolistic behavior. This underscores the complexity and sensitivity surrounding the control of key internet infrastructure in the age of AI.

Early Wins: Identifying Initial Use Cases for Agents

Despite the current challenges and ongoing development required for AI agents to reach their full potential, there are specific areas where they are expected to become useful relatively quickly. Srinivas identified several initial use cases where agents can immediately enhance the user experience.

These early applications tend to focus on automating repetitive, time-consuming, or information-gathering tasks that currently require significant manual effort:

• Personal Information Retrieval: Agents can significantly improve the ability to find specific pieces of personal information that might be scattered across various digital sources. This could include tasks like finding “the article I read last week about this one particular company,” which often involves sifting through browser history or saved links.
• Information Summarization: Agents can condense large amounts of information from digital feeds or documents. An example provided is summarizing a social media feed to quickly grasp trending topics without getting sidetracked by the platform’s infinite scroll. This saves time and helps users focus on essential information.
• Automated Scheduling and Task Management: Agents can handle routine administrative tasks that involve interacting with calendars, email, and reminder systems. Scheduling meetings, finding alternative times when conflicts arise, or setting reminders for specific follow-ups are all tasks that can be automated, freeing up user time and reducing cognitive load.

Srinivas refers to these types of tasks as “boring things” that are ripe for automation. By successfully handling these initial, practical use cases, AI agents can demonstrate their value to users and build the trust needed for tackling more complex actions in the future. This gradual approach, starting with automating tedious but common tasks, is seen as the path forward for mainstream adoption of agentic capabilities.

Conclusion

The landscape of how we access and interact with online information is undergoing a profound transformation, driven by advancements in artificial intelligence. Perplexity, as a prominent player in AI-powered search, is positioning itself at the center of this evolution, specifically championing the rise of AI agents.

Aravind Srinivas, the company’s CEO, views these agents not as supplementary tools but as the fundamental architecture for the next era of web interaction. His vision extends beyond single-use applications, seeing agents as intelligent assistants capable of tailoring experiences across a vast range of tasks, from finding information to completing complex transactions.

Realizing this vision requires overcoming significant technical and ecosystem challenges, particularly gaining seamless access to the functionalities within existing applications and navigating the complexities of different mobile operating systems. It also necessitates building robust systems that users can trust, addressing critical concerns around accuracy, bias, and error handling.

Furthermore, the strategic competition for control over key digital infrastructure, such as web browsers and platform access, is intensifying. Perplexity’s perspective highlights the potential implications of this battle for the principles of the open web, raising concerns about consolidation of power in the hands of a few dominant players.

While AI agents are still in their nascent stages, much like AI search was only a few years ago, the potential for automating tedious tasks and providing highly personalized digital experiences is significant. As models improve and platforms become more open, the capabilities of these agents are expected to grow dramatically. The push to integrate AI assistants deeply into mobile devices and online workflows underscores the industry’s belief that agentic AI represents the inevitable future of digital interaction. The competition to define and control this future is clearly underway, making AI agents the next crucial battleground for the web.