Perplexity Launches 'Computer' AI Agent
AI startup Perplexity has launched “Computer,” a multi-model AI system designed to autonomously create and execute entire workflows. The platform represents a move toward enterprise-grade agentic AI, allowing businesses to automate complex, multi-step processes without direct human intervention.
Perplexity's "Computer" works by orchestrating 19 different specialized AI models, delegating subtasks to the best system for the job. For instance, it might use Anthropic's Claude Opus for reasoning, Google's Gemini for deep research, and its own models for other functions, acting as a manager for a team of AI specialists. The system was launched on February 25, 2026, and is designed to handle long-running projects that can continue for hours or even months. Tasks are executed asynchronously in a secure cloud environment, which includes a browser, filesystem, and tool integrations with platforms like Gmail, Slack, and Notion. Perplexity was founded in August 2022 by a team with experience at top AI labs: CEO Aravind Srinivas (OpenAI, Google), CTO Denis Yarats (Meta), CSO Johnny Ho (Quora), and President Andy Konwinski (Databricks). The San Francisco-based company has attracted high-profile investors, including Jeff Bezos and Nvidia. Following a $500 million funding round in June 2025, the company's valuation reached $14 billion. Unlike some open-source AI agents that run locally on a user's device, Perplexity's system operates entirely in the cloud. CEO Aravind Srinivas has highlighted this as a key safety feature, comparing local-access agents to potential malware risks. Access to "Computer" is available to subscribers of Perplexity's "Max" tier, which is priced at $200 per month. This subscription includes a monthly credit system to pay for the agent's usage-based tasks. This launch places Perplexity in the competitive enterprise agentic AI market, a sector valued at over $3.6 billion in 2024. This market focuses on creating autonomous systems that can plan and execute complex business workflows with minimal human oversight.
