Perplexity has launched Computer, an agentic AI platform that orchestrates 19 different models to execute complex workflows autonomously. Unlike their Model Council feature (which cross-validates search queries), Computer is a full digital worker: it can research, code, deploy applications, and manage ongoing projects for hours or even months without intervention.
This represents a significant step beyond chatbots and copilots. We are now seeing the emergence of autonomous agent platforms that can handle end-to-end project execution, not just answer questions or suggest code snippets.
How Computer Orchestrates 19 Models
The core architecture treats AI models as specialized workers rather than general-purpose assistants. When you submit a task, Computer's orchestration layer (powered by Claude Opus 4.6) analyzes the request and routes subtasks to the most appropriate model.
The model roster covers diverse capabilities:
- Claude Opus 4.6: Core reasoning and task decomposition
- Gemini: Deep research and information synthesis
- GPT 5.2: Long-context interactions and extended web search
- Grok: Fast, lightweight processing for simple tasks
- Nano Banana: Image generation
- Veo 3.1: Video production
The system can spawn subagents for specific tasks like data scraping, chart parsing, or code generation. These subagents work in parallel and merge their outputs into a final deliverable.
This automatic model selection is the key differentiator. If one model excels at code but struggles with math, Computer routes code tasks to that model while delegating math to another. The user does not need to understand model strengths or manually select providers.
Persistent, Asynchronous Execution
What distinguishes Computer from most AI tools is its ability to run autonomously in the background. You can assign a task before leaving for the night and return to completed work.
The platform maintains persistent memory across sessions. Your preferences, project context, and conversation history carry forward. This continuity makes Computer suitable for ongoing projects rather than one-off queries.
Tasks execute in a sandboxed cloud environment, isolating security risks from your local network. This addresses a legitimate concern with autonomous agents: the blast radius of mistakes or exploits is contained.
One reviewer reported building two branded micro-apps, four research packets, and an automation workflow overnight, with working code deployed to GitHub in under 30 minutes from concept. These are not cherry-picked demos but practical examples of what autonomous execution enables.
400+ App Connectors
Computer integrates with the tools enterprises already use. Supported connectors include:
- Productivity: Gmail, Outlook, Google Calendar, Slack, Microsoft Teams
- Development: GitHub, Linear, Jira
- Documents: Notion, Confluence, Google Drive, OneDrive, Dropbox
- Data: Snowflake, Databricks, Salesforce
This integration layer transforms Computer from a standalone agent into a workflow automation platform. It can pull context from your existing systems, execute tasks, and push results back to appropriate destinations.
For organizations already invested in tool ecosystems, this connectivity reduces friction significantly. You are not asking employees to learn a new interface for every task. Computer operates within their existing workflows.
Pricing and Availability
Computer is currently available to Perplexity Max subscribers at $200 per month. The subscription includes 10,000 monthly credits, with early adopters receiving a bonus 20,000 credits (valid for 30 days).
This marks Perplexity's first implementation of per-token billing for consumers. Users can set spending caps and select specific models for subtasks, providing cost control for intensive workflows.
Pro and Enterprise tier access is planned following load testing. The staged rollout suggests Perplexity is being cautious about infrastructure capacity, which is appropriate given the compute intensity of multi-model orchestration.
Comparison with Claude and OpenClaw
Computer occupies interesting competitive ground. It relies on Claude Opus 4.6 as its reasoning backbone, effectively positioning Claude as infrastructure rather than competitor.
The complementary use case is clear: Claude Cowork (Anthropic's agent offering) excels at deep document work and extended reasoning. Computer excels at orchestrating diverse models for researched, built, and deployed projects.
Where they compete is in agentic capabilities. Claude now manages subagent teams internally. Computer manages subagents across 19 external models. The architectural difference matters for specific use cases but may not matter for typical users.
OpenClaw, the open-source alternative, offered local execution and direct messaging integration but faced ecosystem collapse. Provider bans, account suspensions, and security concerns made it impractical for production use. Perplexity's secure cloud sandbox avoids these risks at the cost of desktop control capabilities.
Implications for AI Practitioners
Several observations stand out for those building AI systems:
Multi-model orchestration is becoming productized. What required custom engineering last year is now a subscription feature. The barrier to running coordinated model workflows has dropped dramatically.
Autonomous execution is the new frontier. The shift from synchronous chat to asynchronous task execution represents a fundamental change in how we interact with AI. Tools that can work while you sleep have different utility than tools that require real-time attention.
Pricing models are evolving. Per-token billing with spending caps gives users granular cost control. This addresses a legitimate enterprise concern: unpredictable AI spending. Expect more platforms to adopt similar models.
Security sandboxing is essential. As agents gain more capabilities, isolation becomes critical. Perplexity's cloud sandbox approach is one solution. Others will emerge for on-premise and hybrid deployments.
Practical Considerations for the UAE Market
For enterprises in the UAE and broader Middle East, several aspects of Computer are worth noting.
The cloud-hosted model eliminates local infrastructure requirements. Organizations can access frontier AI capabilities without building GPU clusters. This is particularly relevant for mid-sized companies that cannot justify dedicated AI infrastructure.
The app connector ecosystem covers enterprise tools common in the region: Microsoft 365, Salesforce, Slack. Integration with existing workflows reduces adoption friction.
The $200 monthly price point positions this as a professional tool. For knowledge workers whose time costs significantly more than $200 per month, the productivity gains from autonomous task execution can justify the investment.
Looking Forward
Perplexity pulled a planned demo of Computer after discovering late-stage flaws, a reminder that autonomous agents remain brittle. The technology works impressively in many cases but fails unpredictably in others.
This brittleness will improve. The underlying models are getting better. Orchestration techniques are maturing. Within 12 to 18 months, platforms like Computer will likely be substantially more reliable.
For now, Computer is best suited for practitioners comfortable with iteration, those who can review outputs, catch errors, and refine workflows. It is not a replacement for human judgment but an amplifier of human capability.
The era of autonomous AI workers has arrived. It is imperfect but improving rapidly. Those who experiment now will be better positioned as the technology matures.