ART ARGENTUM ANALYSIS

The Future of Web Search with AI Agents

Analysis of AI agents' impact on web search, based on "Parallel's Parag Agrawal: Building a New Web for AI Agents" | Sequoia Capital.

2026-08-25Sequoia CapitalParallel's Parag Agrawal: Building a New Web for AI Agents

OPEN SOURCE

SUMMARY

Parag Agrawal is spearheading a transformative approach to web search through Parallel Web Systems, emphasizing the importance of agent feedback over traditional human click data. This shift is designed to address the limitations of existing search technologies, anticipating a future where AI agents will dominate web queries and necessitate a complete overhaul of current business models.

Agrawal argues that the traditional web search infrastructure, which relies heavily on human interactions, is inadequate for the emerging needs of AI agents. He posits that agents will perform searches at a scale far exceeding human capabilities, which requires a new feedback mechanism and a reimagined approach to indexing and ranking information.

The economic implications of this transition are significant, as Agrawal warns that the ad-supported internet model may collapse when agents replace human users. He proposes a new compensation structure for content creators based on Shapley values, which aims to ensure that publishers are paid for the value their content provides to AI agents.

Agrawal's vision includes the development of a search agent product that allows for real-time web crawling and data collection, enabling incremental index building. This innovative approach is expected to enhance the efficiency of AI agents, allowing them to process queries in milliseconds and significantly increasing the volume of data handled.

Despite the promising advancements, Agrawal acknowledges the challenges of implementing a sustainable economic model that compensates content creators fairly. He predicts that within 12 to 24 months, a scalable system could emerge that provides real financial compensation to a wide range of content owners, reflecting a significant shift in how information is consumed online.

XDETAIL

INFO

YOUTUBE2026-08-25sequoia capital

Parallel’s Parag Agrawal: Building a New Web for AI Agents

STANCE

00:00

05:00

10:00

15:00

20:00

25:00

30:00

35:00

40:00

45:00

50:00

11 intervals • swipe left

Parallel’s Parag Agrawal: Building a New Web for AI Agents

sequoia_capital • 2026-08-25 12:00:23 UTC

Parag Agrawal is leading Parallel Web Systems to transform web search by prioritizing agent feedback over human click data. This shift aims to address the limitations of current search technologies and business models in…

FULL

00:00–05:00

Parallel Web Systems, founded by Parag Agrawal, aims to revolutionize web search by focusing on agent feedback rather than human click data, which they consider a flaw in traditional search models
Agrawal argues that agents will perform searches at a scale far exceeding human capabilities, necessitating a complete overhaul of search technology and business models to accommodate this shift
The current web search problem involves efficiently crawling, indexing, and ranking vast amounts of information to deliver relevant results to users, a process that is being reimagined for agent-based interactions
Agrawal emphasizes the need to build technology for future customers—agents—rather than relying on established practices from his time at Twitter, which was focused on human user engagement
The economic implications of this shift are significant, as the ad-supported internet model may collapse when agents replace human users, prompting the need for new compensation structures for content creators

Read full analysis

STANCE

STANCE MAP

Proponents of AI-driven web search

AI agents will significantly enhance search efficiency and volume
New economic models are necessary to adapt to the changing landscape

Skeptics of the ad-supported internet model

Challenges in implementing Shapley values for content compensation

Neutral / Shared

Agrawals predictions about the future of content monetization reflect ongoing uncertainties

FULL

05:00–10:00

Parag Agrawal is leading Parallel Web Systems to develop a search infrastructure that prioritizes agent feedback over traditional human click data. This approach aims to address the limitations of existing web search technologies in a future where AI agents will dominate web queries.

Agrawal argues that traditional web search infrastructure, reliant on human click data, is inadequate for the emerging needs of AI agents, which require a different feedback mechanism
He emphasizes that the ability to generate ratings data through advanced models allows for a more efficient and scalable approach to search indexing and ranking
Parallel Web Systems launched a search agent product before a full search engine, enabling real-time web crawling and data collection based on user queries, which allows for incremental index building
The company targets sectors where human labor is currently outsourced for data curation, such as insurance underwriting and sales data enrichment, demonstrating practical applications of their search agents
Agrawal believes that the shift towards agent-based search will necessitate new economic models, as the ad-supported internet may falter when agents replace human users

FULL

10:00–15:00

Parag Agrawal is leading Parallel Web Systems to develop a search infrastructure that prioritizes agent feedback over traditional human click data. The company aims to enhance AI agents' capabilities by optimizing search efficiency and reducing inference time.

Parallel Web Systems focuses on enhancing AI agents capabilities rather than merely developing large models, aiming to create systems that improve the performance of existing models
The company has optimized its search agent to process queries in just 20 milliseconds, significantly reducing the time required for inference and improving efficiency
Agrawal emphasizes the importance of organizing information effectively to maximize the utility of pre-trained models, addressing the challenge of selecting the best tokens from vast amounts of data
Using Parallels search technology allows agents to operate with fewer tokens, enhancing accuracy and speed while reducing costs, which is crucial for handling an infinite demand for relevant information
The shift towards agent-based search represents a new capability that surpasses traditional methods, such as relying on Google for every query, by providing more efficient and contextually relevant results

FULL

15:00–20:00

Parag Agrawal is developing a search infrastructure that prioritizes agent feedback over traditional human click data, aiming to enhance the efficiency of AI agents in web search. This approach addresses the limitations of existing search technologies and anticipates a future where agents will dominate web queries.

Agrawal highlights the need for intelligent compute allocation across different layers of the search system, emphasizing that agents require a different interface compared to human users who typically rely on keyword searches
While humans often exhibit laziness in query formulation, agents can handle more complex queries with fewer errors, allowing for more precise and efficient information retrieval
The existence of low-quality content on the web, often optimized for SEO, is a response to human laziness; however, agents can bypass this by directly accessing authoritative sources, thus improving the quality of information delivered
Agrawal illustrates the inefficiencies of traditional search methods, such as the time-consuming process of loading lengthy documents, and contrasts this with agents ability to extract relevant information quickly from authoritative sources
The architecture of Parallels search system involves multiple indexes that organize web content in various ways, enabling agents to craft tailored queries for different types of information

FULL

20:00–25:00

Agrawal discusses the complexity of optimizing search queries, emphasizing the need to refine vast amounts of data into concise, relevant excerpts for AI agents
He highlights the importance of incremental optimization in search technology, suggesting that while improvements are ongoing, there will be a point where further enhancements may not yield significant returns
The architecture of Parallels search system is designed to leverage multiple models and indexes, allowing for tailored information retrieval that surpasses traditional methods
Agrawal argues that existing web crawls may not be beneficial for building a new index, as they often focus on content already utilized in model training, which limits their utility
He believes that the future of AI agents will rely on accessing underutilized web content, which can provide unique insights and enhance the overall quality of information retrieval

FULL

25:00–30:00

Parag Agrawal is developing a search infrastructure that prioritizes agent feedback over traditional human click data, aiming to enhance the efficiency of AI agents in web search. The company has partnered with Google Cloud to provide search capabilities for enterprise agent APIs, integrating their search product with Google's Gemini models.

Parag Agrawal emphasizes the need for search infrastructure optimized for AI agents, suggesting that model companies should develop or acquire the best systems to support this shift
Parallel Web Systems has partnered with Google Cloud to provide search capabilities for enterprise agent APIs, integrating their search product with Googles Gemini models to enhance performance
Agrawal highlights the significant increase in search volume driven by AI agents, noting that a typical search agent can perform 5 to 20 searches in seconds, with potential for even greater multipliers in specific applications
The use of AI agents can transform traditional processes, such as risk assessment for small businesses, by enabling more frequent and efficient web searches, thereby increasing the overall volume of data processed
Agrawal anticipates that the economic model of the ad-supported internet will face challenges as AI agents become more prevalent, necessitating new compensation structures for content providers based on the value their pages deliver to these agents

FULL

30:00–35:00

Parag Agrawal is developing a search infrastructure that prioritizes agent feedback over traditional human click data, aiming to enhance the efficiency of AI agents in web search. He discusses the economic implications of AI agents on the ad-supported internet model, suggesting that traditional metrics of human attention are becoming obsolete.

Agrawal discusses the increasing efficiency of AI agents in performing web searches, noting that a single agent can conduct tens to hundreds of searches for tasks like meeting preparation, significantly multiplying the volume of data processed
He emphasizes the importance of balancing the value generated by these agents against their operational costs, suggesting that while current spending may seem irrational, it will eventually stabilize as the technology matures
The company initially focused on optimizing search quality and cost before addressing latency, leading to the development of their Turbo product, which is touted as the fastest and highest quality search agent available
Agrawal acknowledges that while AI traffic is growing, it is still early in the adoption phase, with some users experiencing search volumes that are 1000 times greater than traditional human queries
He raises concerns about the economic implications of AI agents on the ad-supported internet model, suggesting that the traditional metrics of human attention and engagement are becoming obsolete

METRICS

OTHER

1000Xtimes

details

CONTEXT: the increase in search volume by agents compared to traditional human queries

WHY: This indicates a significant shift in how web searches are conducted, highlighting the growing role of AI agents

EVIDENCE: I think today if you just look across it all of my agents, like I bet they're doing 1000X more than that.

FULL

35:00–40:00

Parag Agrawal is developing a search infrastructure that prioritizes agent feedback over traditional human click data, aiming to enhance the efficiency of AI agents in web search. He discusses the economic implications of AI agents on the ad-supported internet model, suggesting that traditional metrics of human attention are becoming obsolete.

This segment is mostly promotional material and adds little editorial content

METRICS

OTHER

7xtimes

details

CONTEXT: the expected growth rate of AI inference this year

WHY: Such rapid growth highlights the need for sustainable business models in the evolving digital landscape

EVIDENCE: let's say 7x this year and another 7x the next year

FULL

40:00–45:00

Parag Agrawal is developing a search infrastructure that prioritizes agent feedback over traditional human click data, aiming to enhance the efficiency of AI agents in web search. He discusses the economic implications of AI agents on the ad-supported internet model, suggesting that traditional metrics of human attention are becoming obsolete.

Parag Agrawal emphasizes the need for fresh data during inference for AI products, highlighting the importance of payments to content owners for their contributions
He discusses the challenge of attributing value to various content sources when an AI agent generates a response from multiple searches, advocating for a model that aligns incentives among content creators
Agrawal introduces Shapley values as a theoretical framework for determining the incremental value of content, although he notes the practical difficulties in computing these values in real-time
The computation of Shapley values can be costly, often exceeding the payments made to content owners, which complicates the economic model of compensating publishers
Despite the challenges, Agrawal believes that with the right data and models, it is possible to estimate the value of content accurately, fostering collaboration among content creators

METRICS

OTHER

10searches

details

CONTEXT: the number of searches an AI agent performs for a simple query

WHY: This indicates the potential scale of AI agent interactions compared to human users

EVIDENCE: even a very simple query will go do 10 searches

FULL

45:00–50:00

Parag Agrawal is developing a search infrastructure that prioritizes agent feedback over traditional human click data, aiming to enhance the efficiency of AI agents in web search. He discusses the economic implications of AI agents on the ad-supported internet model, suggesting that traditional metrics of human attention are becoming obsolete.

Agrawal emphasizes the importance of aligning incentives for content owners and AI agents, suggesting that Shapley values can help maximize participation by compensating publishers based on the unique value of their content
He notes that the economic model of the ad-supported internet is at risk as AI agents begin to dominate web interactions, necessitating a new approach to content monetization
Agrawal predicts that within 12 to 24 months, a scalable system could emerge that provides real financial compensation to a wide range of content creators, especially as agent usage on the web increases significantly
The transition to a parallel web requires content creators to consider both human and agent audiences, leading to a dual publishing strategy that optimizes content for both types of consumers
Agrawal reflects on the shift in content consumption, highlighting that earnings transcripts are likely being accessed more by agents than by human listeners, indicating a significant change in how information is consumed online

METRICS

OTHER

12 to 24 monthsmonths

details

CONTEXT: the predicted time frame for meaningful financial compensation to content owners

WHY: This timeline indicates a significant shift in the economic model for content creators as AI agents become more prevalent

EVIDENCE: by my calculations like 12 to 24 months from this math being able to give meaningful dollars for a very wide range of content owners on the web

FULL

50:00–55:00

Parag Agrawal is developing a search infrastructure that prioritizes agent feedback over traditional human click data, aiming to enhance the efficiency of AI agents in web search. He discusses the economic implications of AI agents on the ad-supported internet model, suggesting that traditional metrics of human attention are becoming obsolete.

Agrawal emphasizes that the primary audience for documentation and APIs is increasingly AI agents rather than human users, reflecting a shift in how information is consumed online
He envisions a future where agents operate in a parallel web that transitions from a pull model, where agents request information, to a push model, where the web proactively informs agents of relevant changes
The development of sophisticated multi-agent systems is on the rise, allowing agents to collaborate and orchestrate tasks, which could lead to more complex and efficient problem-solving capabilities
Agrawal predicts that within a few years, the web will enable agents to perform tasks based on real-time changes and insights, significantly enhancing their operational efficiency
He highlights the potential for a scalable system that compensates content creators financially as agent usage increases, with real payments expected to reach publishers within 12 to 24 months

CRITICAL ANALYSIS

Parag Agrawal's vision for Parallel Web Systems represents a significant shift in web search dynamics, focusing on AI agents rather than traditional human interactions. While the emphasis on agent feedback is innovative, the economic implications of this transition raise questions about the sustainability of the ad-supported internet model. Agrawal's reliance on Shapley values for compensating content creators introduces complexity, as the practical challenges of real-time computation may hinder implementation.

METRICS

other

1000X times

the increase in search volume by agents compared to traditional human queries

This indicates a significant shift in how web searches are conducted, highlighting the growing role of AI agents

I think today if you just look across it all of my agents, like I bet they're doing 1000X more than that.

other

7x times

the expected growth rate of AI inference this year

Such rapid growth highlights the need for sustainable business models in the evolving digital landscape

let's say 7x this year and another 7x the next year

other

10 searches

the number of searches an AI agent performs for a simple query

This indicates the potential scale of AI agent interactions compared to human users

even a very simple query will go do 10 searches

other

12 to 24 months months

the predicted time frame for meaningful financial compensation to content owners

This timeline indicates a significant shift in the economic model for content creators as AI agents become more prevalent

by my calculations like 12 to 24 months from this math being able to give meaningful dollars for a very wide range of content owners on the web

THEMES

#ai_startups#parallel_web#search_infrastructure#agent_feedback#ad_model#agent_search#ai_agents#content_monetization#content_value#economic_implications#intelligent_compute#parag_agrawal#search_agents#search_innovation#shapley_values#web_revolutionweb searchParallel Web Systems

DISCLAIMER

This analysis is an original interpretation prepared by Art Argentum based on the transcript of the source video. The original video content remains the property of the respective YouTube channel. Art Argentum is not responsible for the accuracy or intent of the original material.