Perplexity Computer: the age of the digital worker

On February 25, 2026, Perplexity launched Computer — not hardware, despite the name, but a cloud system that turns an objective ("build a website," "produce this research report," "deploy an app") into decomposed tasks, delegates them to specialized sub-agents, and returns finished outcomes rather than drafts. Claude Opus 4.6 sits at the core as the reasoning engine; around it, a model-agnostic harness routes work to Gemini, Nano Banana, Veo 3.1, Grok, and ChatGPT 5.2 as the subtask demands. Each task runs in an isolated environment with a real filesystem, a real browser, and 400+ tools — asynchronously, in parallel, spawning its own sub-agents when it hits problems, and only checking in when it genuinely needs a human.

The build story is as telling as the product: Comet, Perplexity's browser, took six months; Computer took two, built largely with Claude Code — and the product helped finish itself, animating its own logo and modifying its own codebase. It wasn't on the roadmap until December 2025, when frontier capabilities made it suddenly feasible.

Three readings for IT leaders. First, the model is no longer the product — Computer treats models like an OS treats drivers, swappable components under an orchestration layer, which is where the value now migrates. Second, build-vs-buy got existential: when a subscription gives any knowledge worker autonomous digital workers on demand, the strategic question shifts from "should we adopt agents?" to "where on the autonomy spectrum, and who builds our orchestration layer?" Third, operations get rewritten — Gartner expects 40% of enterprise applications to embed agents by the end of 2026, and monitoring, access control, and audit trails all have to account for non-human actors that work around the clock and spawn sub-processes.

For engineers, this is the "agent architect" pattern arriving at consumer scale: delegate, review, own. Multi-agent orchestration is becoming what microservices were to the 2010s — complete with emerging protocols (MCP, A2A), cost patterns like plan-and-execute, and a tooling gap around observability and governance that someone will fill. Gartner logged a 1,445% surge in multi-agent system inquiries in eighteen months. That's not a trend; it's a phase transition, and Computer is its clearest consumer-grade evidence yet.