Perplexity Computer: An Orchestra of 19 AI Models
Perplexity has officially launched Perplexity Computer, positioning it as a significant step forward in agentic AI. Instead of relying on a single, all-purpose model, this system breaks down complex user requests into subtasks and assigns each one to the most suitable of 19 specialized AI models. Our initial testing confirms this approach allows for nuanced and powerful project execution, from building competitor analysis dashboards to coding live satellite tracking apps from a single prompt.
The system runs in a sandboxed cloud environment with persistent memory, remembering past projects and user preferences. It's a direct competitor to platforms like OpenClaw but with a more accessible user interface. Access is currently available for Perplexity Max subscribers at $200 per month, which includes a consumption-based credit system.
The core idea behind Perplexity Computer is that the future of AI isn't one monolithic model, but a team of specialists working in concert. The platform leverages models like Claude Opus 4.6 for reasoning, GPT-5.2 for long-context tasks, and Gemini for deep research.
The Broader Rise of Agentic AI Workers
Perplexity isn't alone in pushing the boundaries of autonomous AI. Several other platforms released features this week aimed at making AI agents more capable of interacting with digital environments and performing complex, multi-step tasks. These updates signal a clear industry trend away from simple chatbots toward true digital assistants.
Tool | New Agentic Feature | Key Detail |
|---|---|---|
Claude Cowork | Scheduled Tasks | Users can automate recurring tasks with the |
Cursor Cloud Agents | Dedicated Cloud VMs | Agents now operate in full dev environments to run software, test changes, and return artifacts like videos and logs. |
Gemini on Android | In-App Autonomy | Can now complete multi-step tasks inside third-party apps like Uber and DoorDash, currently in beta. |
Claude Code | Remote Control | Start a coding session in the terminal and continue controlling it from any web or mobile device. |
New Wave of Developer and Creative Tools
The developer community received several new resources. Apple released a Foundation Models SDK for Python, enabling developers to perform on-device inference with its framework. On the agent side, Cognition launched Cognition for Government, bringing its Devin AI coding agent to U.S. agencies, and also previewed Devin 2.2, which can self-verify and auto-fix its work.
For creatives, Quiver AI opened public beta access to Arrow 1.0, its model for generating SVG vector graphics from text. Adobe also entered the fray with Quick Cut, a new Firefly tool designed to turn raw video footage into a usable first cut, streamlining the editing process.
Specialized AI Applications Hit the Market
Beyond the major agentic platforms, a host of specialized tools also launched, targeting specific use cases from productivity to safety.
Productivity & Organization: Wispr Flow offers advanced dictation that polishes speech into clean text, while TabAI helps organize tasks from multiple sources. For designers, Boards by Variant AI turns curated inspiration directly into design concepts.
Business & Enterprise: Rowspace creates an intelligence layer from an investment firm's scattered documents. Reloop helps create ad campaigns without complex prompting, and Protex provides AI video analytics for CCTV to detect safety hazards.
Gaming & Music: Moonlake can turn a text prompt into a playable 3D game, while Music Arena added Google's Lyria 3 and ElevenLabs Music v1 to its blind-comparison platform for AI music generation.