Perplexity Computer landed this week, and despite the name, it isn't a piece of hardware. It's a cloud-based agent system that Perplexity is positioning as a general-purpose digital worker, one that takes a goal and spends hours or months executing it by using the same software humans use.
What It Actually Is
Computer lives inside Perplexity's existing product. Subscribers to the company's top Max tier, currently priced at $200 per month, get access to it alongside a monthly allotment of credits for task execution. The pitch is straightforward: describe an outcome in natural language, and Computer breaks it into subtasks, delegates them to specialized sub-agents, and runs the workflow in the background.
In practice, that means asking Computer to research competitors and produce a deck might spin up one sub-agent to handle web research, another to draft slides, a third to format the output, and a fourth to deliver it. Perplexity describes the system as one that "reasons, delegates, searches, builds, remembers, codes, and delivers." Connectors ship for Gmail, Outlook, GitHub, Linear, Slack, Notion, Snowflake, Databricks, Salesforce, and various finance tools.
The Multi-Model Trick
The interesting architectural choice is that Computer isn't built around a single model. Perplexity uses Claude Opus 4.6 as its core reasoning engine but routes specific subtasks to other models based on their strengths. Gemini handles deep research, Grok tackles lightweight speed tasks, ChatGPT 5.2 is used for long-context recall, and specialized models are invoked for image and video generation.
This orchestration layer is where Perplexity is making its bet. If you believe no single model will dominate every task, then the company that does the best job of routing work across multiple models at runtime has a durable position. It's a different proposition than being the best model or the best embodied AI. It's about being the best conductor.
The Longer-Running Claim
Perplexity says Computer can sustain workflows for hours or months. That duration is notable. Most agent products today are measured in minutes. Agents that can hold state and execute across weeks open up categories of work that short-session tools cannot touch: ongoing research projects, recurring reports, longitudinal monitoring.
Whether the product lives up to that claim is a separate question. Long-running agents are where reliability issues compound. A workflow that works 95 percent of the time over a ten-minute task fails frequently at the scale of months. Perplexity will need to show that the system actually holds up in real customer use, not just demos.
Who This Is For
At $200 a month, Computer isn't competing for casual users. The target is professionals, small teams, and early adopters who can justify the cost by delegating hours of knowledge work per week. Founders running lean operations, analysts doing research-heavy roles, consultants producing deliverables. The kind of customer who already pays for a stack of SaaS tools and would happily trade some of that spend for something that actually finishes the work.
It's worth separating Computer from Perplexity's other recent push, the Comet browser. Comet is a free AI-native browser for individuals. Computer is a paid agent platform for getting work done. Different products, different customers, overlapping philosophy.
The Real Bet
The broader argument Perplexity is making is that the app store model is ending. If an agent can operate any software interface the way a human does, the value moves up the stack. You don't need a Salesforce client, you need something that can use Salesforce on your behalf. Multiply that across every tool in a typical knowledge worker's stack and the category of "software that humans directly operate" starts to shrink.
That future isn't guaranteed. Agents still fail in unpredictable ways. Enterprise customers care deeply about reliability, security, and audit trails, areas where autonomous systems remain immature. The companies that own the software Computer operates, Microsoft and Google in particular, won't passively let third-party agents become the primary interface to their products.
But the direction of travel is clear. Perplexity Computer is one of the more concrete versions of the agent thesis to reach market. The question now is whether customers actually trust it with real work.
