Former Twitter CEO Parag Agrawal Predicts AI Agents Will Drive a 'Parallel Web'

Former Twitter CEO Parag Agrawal believes the internet is approaching a fundamental structural shift where AI agents, rather than humans, become the primary drivers of web interaction. Speaking on the Training Data podcast, Agrawal outlined a vision for a "parallel web" designed specifically for machine consumption, suggesting this transition could occur within the next 12 to 24 months.

Agrawal, who has since founded Parallel Web Systems, argues that the current mechanisms of the internet are ill-equipped for the generative AI era. He specifically contends that traditional search mechanisms—which rely heavily on human click data to determine relevance—are insufficient for AI agents. In a sharp critique of current search paradigms, Agrawal described human click data as a "bug" rather than a feature when applied to the needs of autonomous AI systems.

The Shift to a Machine-First Web

This perspective comes as large language models (LLMs) increasingly challenge the traditional link-based economy of the internet. For decades, the web has operated on a "pull" model, where humans search for information and click through to publishers. Agrawal envisions a transition to a "push" model, where AI agents efficiently retrieve and utilize data without the need for human-centric interfaces. To facilitate this, Parallel Web Systems is developing technology intended to enable AI agents to search and navigate the web more effectively than current tools allow.

Redefining Internet Economics

The transition to an AI-driven web threatens the existing economic model of the internet, which is built almost entirely on human attention and advertising revenue. If AI agents replace humans as the primary visitors to websites, the traditional traffic-based monetization model for content creators and publishers will collapse.

To address this disruption, Agrawal suggests a move toward new incentive alignment frameworks. He proposes using Shapley values—a method from cooperative game theory used to distribute total gains among players based on their marginal contribution—to ensure content owners are fairly compensated when their data is used by AI agents to generate value.

The Road Ahead

As the industry moves toward this agentic future, the primary challenge remains the tension between AI companies and the publishers who provide the underlying data. While Agrawal predicts a rapid shift toward a parallel web, it remains to be seen how content owners will react to a model that removes the human visitor from the equation. The coming two years will likely determine whether the web evolves into a collaborative ecosystem for agents or a fragmented landscape of walled gardens.