Perplexity Announces 'Computer,' an AI Agent That Assigns Work To Other AI Agent

Source: Slashdot

Overview

joshuark shares a report from Ars Technica: Perplexity has introduced Computer, a new tool that allows users to assign tasks and see them carried out by a system that coordinates multiple agents running various models. The company claims that Computer, currently available to Perplexity Max subscribers, is “a system that creates and executes entire workflows” and is “capable of running for hours or even months.”

How Computer Works

The user describes a specific outcome—e.g., “plan and execute a local digital marketing campaign for my restaurant” or “build me an Android app that helps me do a specific kind of research for my job.” Computer then:

1. Ideates subtasks.
2. Assigns each subtask to one or more agents.
3. Runs the models Perplexity deems best for those tasks.

The core reasoning engine currently runs Anthropic’s Claude Opus 4.6. Other models are used for specialized tasks:

• Gemini – deep research
• Nano Banana – image generation
• Veo 3.1 – video production
• Grok – lightweight tasks where speed matters
• ChatGPT 5.2 – long‑context recall and wide search

All tasks execute in the cloud with pre‑built integrations. “Every task runs in an isolated compute environment with access to a real filesystem, a real browser, and real tool integrations,” Perplexity says.

Model‑for‑Task Approach

This best‑model‑for‑the‑task strategy differs from some competing products, such as Claude Cowork, which relies solely on Anthropic’s models.

Users have previously combined multiple models and tailored them to specific tasks, often using protocols like MCP (Model Context Protocol) to give models access to local data and applications. Perplexity Computer takes a different approach—centralizing the orchestration in the cloud—while still aiming to let AI agents operate on users’ files, services, and applications.

Related Concepts

A related effort is OpenClaw, which can be seen as an immediate predecessor to this concept: OpenClaw article.

All links and references are retained from the original article.