Perplexity Launches Autonomous AI Agent
AI startup Perplexity has launched "Perplexity Computer," a multi-model AI system designed to create and execute entire workflows on its own. The platform moves beyond simple text generation to orchestrate complex sequences of tasks, signaling a shift toward more capable, self-directed AI agents.
Perplexity's new system is designed to act as a "general-purpose digital worker" that can operate the same interfaces as a human. Unlike single-task agents, it can manage workflows that run for extended periods, from hours to months, handling everything from research and coding to project management. The platform is built on a model-agnostic orchestration framework, meaning it routes tasks to the most suitable AI model from a pool of 19 specialized systems. This allows it to use different models for different subtasks, such as sending coding requests to Claude or research queries to Gemini, rather than relying on a single, all-purpose model. When the system encounters a problem, it can autonomously create "sub-agents" to find a solution. These sub-agents can perform tasks like researching new information, debugging code, or locating necessary API keys, only escalating to the human user when absolutely necessary. The company was founded in August 2022 by a team of engineers with backgrounds at top AI labs: CEO Aravind Srinivas (OpenAI, Google), CTO Denis Yarats (Meta), CSO Johnny Ho (Quora), and President Andy Konwinski (Databricks). Srinivas, born in Chennai, India, previously worked on projects like DALL-E 2. Perplexity has seen a rapid rise in valuation, reaching over $14 billion by June 2025. The startup is backed by prominent investors including Jeff Bezos, Nvidia, and Databricks. This launch places Perplexity in direct competition with other agentic AI systems from major players. OpenAI's "Operator," currently in research preview for Pro users, can also control a computer's mouse and keyboard to perform web tasks. Anthropic's Claude is also being developed to automate tasks within virtual desktop environments.
