Perplexity Launches Autonomous AI Agent
AI search company Perplexity has launched "Computer," a new multi-model AI system. The platform is designed to autonomously create and execute entire business and technical workflows with minimal human oversight, signaling a major step in AI-driven process automation.
Perplexity's "Computer" isn't a single large language model but an orchestration layer that coordinates 19 specialized AI models. This system acts as a "general-purpose digital worker" by breaking down complex requests into smaller subtasks and assigning each to the most appropriate AI agent for execution. The platform moves beyond the company's origins as a conversational "answer engine." While the original Perplexity synthesizes web results into direct answers with citations, "Computer" is designed to execute multi-step tasks end-to-end, such as conducting research, creating a spreadsheet with the data, and integrating it with tools like Gmail or Slack. Founded in August 2022, Perplexity was started by engineers with backgrounds at top AI labs including OpenAI, Google Brain, and DeepMind. CEO Aravind Srinivas, a UC Berkeley Ph.D., previously worked as a research scientist at OpenAI. The San Francisco-based company has seen a rapid rise in valuation, reaching a reported $20 billion by September 2025. It has attracted significant funding from high-profile investors including Jeff Bezos, Nvidia, and IVP. The system is designed for long-duration tasks that can run for hours or even months, operating in a secure cloud sandbox with access to a browser and filesystem. When it encounters a problem, it can create new sub-agents to find solutions, like researching API keys or coding a necessary app. This launch places Perplexity in the growing field of "agentic AI," where the focus is shifting from the power of individual models to the ability of systems to coordinate multiple AIs to achieve complex goals. The aim is to have the AI operate software and interfaces just as a human would. Perplexity "Computer" is a paid service available to subscribers. The pricing model is usage-based, providing a monthly credit allowance for task execution.
