Perplexity Launches Teammates, a $200-a-Month AI Agent System for Ongoing Work

Perplexity has introduced Teammates, a system of personal AI agents intended to carry out ongoing work with limited supervision. The feature operates within Perplexity Computer, the company’s cloud-based assistant for multi-step tasks, and is initially restricted to Perplexity Max subscribers on the web and iOS.

The basic pitch is less “ask one chatbot a question” and more “delegate a continuing workload to several software workers.” A user can give individual Teammates distinct roles or assignments, let them work at the same time, and follow each agent’s activity through its own sidebar entry in Computer. The agents are described as capable of collaborating, coordinating other Teammates and adding more agents as the workload grows.

That is a meaningful expansion in ambition from a conventional chat interface. It also comes with a very high entry cost: Perplexity Max costs $200 per month, or $2,000 per year.

What Teammates are designed to do

Teammates build on Perplexity Computer, which is already positioned as an agentic assistant that can perform tasks involving multiple steps. In this context, agentic means the software is intended to pursue a task through a sequence of actions rather than simply returning a single reply to a single prompt.

The new layer is about persistence and scale. Rather than keeping all work in one conversation, users can assign separate responsibilities to separate agents. Those agents can run at the same time, continue working in the background around the clock and alert the user when a decision requires human input.

The available description does not enumerate specific job types, integrations or the boundaries of what an agent may do. What it does establish is a workflow model: set a role, assign work, inspect progress, redirect the agent when needed and make decisions when the system asks for them.

Each Teammate has a dedicated entry in Computer’s sidebar. That is important because ongoing automated work can become opaque very quickly. A separate thread gives the user a place to see what the agent is currently doing, what it has completed, provide further instructions, request an update or tell it to change direction.

How a multi-agent setup changes the interaction

A single AI assistant is generally a turn-by-turn experience. The person asks, the model responds, and the person decides what comes next. Teammates shifts the emphasis to assignment and oversight.

Related coverage includes Perplexity Introduces Teammates AI Agents for Ongoing Work.

In the announced setup, one agent can work on one responsibility while another addresses a different one. They can also work together, and an agent can manage other agents to help keep the overall task moving. Perplexity says Teammates can bring in additional Teammates and plan work for those newcomers as more work is assigned.

That makes the feature sound closer to a digital project structure than a standalone text generator. But the distinction should not be overstated. The supplied details describe organizational capabilities, not a guarantee that the agents will independently resolve every complex task. The product’s own design still leaves a person in the loop: Teammates will notify the user when decisions are necessary.

That notification step is arguably the central practical feature. Automation is most useful when it surfaces the moments that need a person’s judgment rather than making it easy to forget that judgment is required. The announcement does not specify how those decisions are prioritized or presented, so users evaluating the service will need to assess whether its alerts are timely and sufficiently clear for their own work.

What “ongoing work” means here

The phrase ongoing work matters because it separates Teammates from a one-off answer engine. The agents are designed to operate in the background, with the user able to revisit their threads, add assignments, check status and alter their direction. Perplexity also characterizes the system as having additional capabilities for long-running, larger-scale workloads beyond the underlying Computer product.

There is a difference, however, between the ability to keep a task active and the ability to make every decision safely or correctly without oversight. The information currently available does not detail accuracy controls, review systems, permissions, data handling, recovery from mistakes or the kinds of actions an agent can take on a user’s behalf. Those omissions do not prove that such controls are absent; they simply remain unaddressed in the disclosed description.

For potential users, that means the sensible question is not only whether several agents can be created. It is whether the work is visible enough to audit, whether a human can intervene at the right time and whether the value of persistent delegation matches the cost.

The Max-only price changes the audience

At $200 monthly, Teammates is not being positioned as a casual add-on. The annual option reduces the stated total to $2,000, but it remains a premium service aimed at people or organizations that expect recurring work to justify a substantial subscription.

That price also frames the feature as a test of outcomes rather than novelty. A user does not merely need to find the agents interesting; the system has to make a sustained workload easier to manage. Parallel assignments, status visibility and decision notifications are potentially valuable, but the benefit will depend on how much useful work can continue without constant correction.

The exclusive rollout also limits who can evaluate the feature directly. At launch, access is available only to Max subscribers using the web or iOS. No availability beyond those platforms is specified.

Analysis: the real product is supervision at scale

It is tempting to read any AI-agent launch as a promise that the software will replace a whole team. The details here suggest a more measured interpretation. Teammates are presented as agents that can be organized into a team-like structure, but their workflow explicitly includes human decision points, supervision and the ability to redirect work.

In other words, the key proposition is not fully hands-off operation. It is the ability to supervise several active assignments from one interface. The sidebar threads are therefore more than a layout choice: they are the record of what each agent has done and the control surface for assigning more work or changing course.

That focus on visibility is likely to matter as much as the agents’ ability to delegate among themselves. If multiple agents are operating simultaneously, users need to know which agent owns which responsibility, what has already been completed and when intervention is needed. Perplexity’s stated design addresses those needs at a high level, though it does not yet provide the finer operational details needed to judge how well the model works in practice.

The launch arrives amid a broader public argument about AI infrastructure and its impact, an issue reflected in recent protests at a Microsoft construction site in Amsterdam. Teammates is a product announcement rather than an answer to that debate, but it illustrates why the conversation has widened beyond simple chatbots. Systems designed to keep working, coordinate with other systems and request human decisions when necessary raise more operational questions than a one-message tool.

What is confirmed, and what remains unclear

Confirmed details

• Teammates are Perplexity’s personal AI agents for ongoing work.
• They run inside Perplexity Computer, the company’s cloud-based, multi-step agentic assistant.
• Users can assign individual roles or tasks to different Teammates.
• Agents can work simultaneously, collaborate and manage other Teammates.
• Teammates may add further agents and organize tasks for them as workloads increase.
• They can work in the background continuously and notify users when decisions are needed.
• Each agent receives its own sidebar entry in Computer for monitoring, follow-up work, course changes and status requests.
• The initial availability is limited to Perplexity Max subscribers on web and iOS.
• Max is priced at $200 per month or $2,000 per year.

Not specified in the available information

• The exact types of tasks and external services Teammates can access.
• How users control permissions, review actions or correct errors.
• Any rollout plans for platforms beyond web and iOS.
• Whether all Max subscribers receive identical limits or capacity.
• How decisions are escalated when multiple agents are involved.

For now, Teammates is best understood as a premium attempt to turn an AI assistant into a persistent, multi-agent workbench. Its appeal will rest on whether people with repeatable, ongoing assignments find the coordination tools and background operation worth the Max subscription—and whether the promised human checkpoints provide enough control over work being delegated to software.