Perplexity Teammates brings always-on agents to shared workspaces

By Sophie Lindqvist

Perplexity has introduced Teammates, an always-on AI-agent system that runs multiple roles in one workspace. The company says each Teammate can take a different job at the same time, directly challenging OpenAI and Anthropic in workplace software.

What does Teammates actually do?

At the center is a shared workspace. Teammates can assume multiple roles there and draw on projects, skills, and memory already sitting in Perplexity tools. Perplexity describes the product as always-on, a term that implies continued availability but comes without a stated service target or duty schedule.

That last point matters. The announcement gives no duration, workload, response-time, or availability limits. Always-on therefore describes intended operation rather than a documented capacity commitment. It does not establish how many agents can run together, how long each task continues, or what happens when their instructions conflict.

The system also supports assigning different jobs to each Teammate and running those jobs simultaneously. Perplexity says this arrangement can complete "ongoing, large-scale work." The phrase sets a broad ambition, but the announcement provides no benchmark, completed customer deployment, or comparison showing how the agents divide complex work.

How does Computer fit?

Perplexity says Teammates "build on top" of Computer. The company describes Computer as an AI system that deploys agents on a device to complete tasks. Teammates adds collaboration, shared roles, simultaneous assignments, and access to existing project context, skills, and memory.

Perplexity does not specify which devices support Computer, which tasks run on the device, or whether project data leaves the device. That omission carries commercial weight. The announcement does not claim that device deployment lowers cost, improves privacy, or removes network requirements.

It gives no telemetry, retention, or security terms. Any such benefit would remain unsubstantiated until product documentation supports it. The commercial case also depends on task limits.

The announcement gives no price, plan, usage allowance, or service commitment for long-running agents. It also omits any agent-count ceiling. Those omissions prevent a meaningful comparison of operating scale.

How does it compare with OpenAI and Anthropic?

Perplexity positions Teammates against two named products: OpenAI's Dots and Anthropic's Claude Duet. The source treats them as similar because Teammates can assign different roles and run separate jobs simultaneously. It provides no feature-by-feature comparison with either product.

The source points to two features that set the comparison: multiplicity and persistence. Teammates assigns different roles and jobs to several agents, allows simultaneous work, and connects those jobs to existing Perplexity context. It does not show whether Dots or Claude Duet lack those functions, only that Perplexity considers its offering comparable.

That caveat prevents a categorical claim of technical leadership. Perplexity also gives no model names, context-window sizes, tool access, permissions, error rates, or task-completion measurements. It discloses no agent limit for a workspace, nor any benchmark for the claimed large-scale work. Multiple and ongoing remain product descriptors, not verified operating specifications.

The comparison also remains directional. Perplexity names two rivals, but it does not claim exclusive access to on-device deployment, shared memory, or simultaneous agent execution. No supplied evidence shows how its coordination layer performs against either rival in routine enterprise work.

What will customers be able to judge?

The announcement leaves basic buying information unanswered. It names no launch date, subscription tier, standalone price, free allowance, or enterprise contract. It also provides no supported geography, customer list, deployment scale, or release schedule. The supplied material therefore cannot establish commercial availability beyond Perplexity's product introduction.

This matters most for an always-on service. Users would need limits on compute consumption, background execution, data access, and task escalation. The announcement gives no figures for any of them. It does not explain whether existing Perplexity subscriptions include Teammates or whether access depends on Computer.

Perplexity may be trying to turn its broader assistant context into an execution layer for longer projects. That strategy would make projects, skills, and memory the connective tissue between individual agents and a shared workspace. The announcement, however, does not prove that those assets transfer cleanly across agents or produce better outcomes.

The commercial uncertainty is therefore practical, not merely promotional. Teammates might reduce the need to move project context between separate tools. The announcement does not quantify that saving, identify eligible workflows, or show that Computer can execute every task assigned to a Teammate.

What will decide uptake?

Competitive pressure is clear because Perplexity names two rivals and two adjacent products. Commercial pressure is less clear because Teammates arrives without pricing, benchmarks, or deployment evidence. That combination favors a cautious reading: Perplexity has described a plausible route from on-device agents to coordinated workplace labor, but has not yet quantified its capacity.

The next contest will depend on what users can run, what it costs, and how reliably the agents finish sustained projects. Until those numbers appear, Teammates offers an ambitious workflow claim rather than a demonstrated lead over Dots or Claude Duet.