Perplexity Teammates gives AI an ongoing job. Who can use it?

Perplexity’s November 6 release update introduces Teammates, agents assigned continuing responsibilities inside Computer. The practical change is that users can hand over a standing job, rather than initiate each task themselves. The initial research preview is for individual Max subscribers on the web; an Enterprise Max account does not automatically qualify.

That eligibility restriction comes from the setup documentation, last modified November 5. It is more specific than the broad launch message. Free and Pro users are outside the initial preview, and Enterprise Pro and Enterprise Max are not included by default.

The launch also received independent coverage from Engadget on November 6. That confirms it is a public product announcement, but the report is not a controlled test of whether an agent can reliably finish a week of work. UpThat has not tested Teammates.

What continues between tasks

According to Perplexity’s capability documentation, a teammate can divide a goal into parallel tasks, add agents as work grows and retain its role and memory between conversations. Teammates in one Project share context, files and instructions. Each has an isolated computer and browser; users can take over when needed.

Perplexity’s launch explanation describes monitoring, investigation and follow-through as one responsibility. Its sales example starts when a regular customer stops buying: the agent investigates account history and support problems, then prepares a response or flags issues before contact. Another example passes website bug findings to a teammate preparing fixes. These are vendor-described scenarios, not measured customer outcomes.

Editorial visualization

The useful distinction is continuity. A scheduled report repeats a specified output; this design proposes that the agent decide which intermediate work a continuing goal needs. That could save coordination time. It also makes the quality of its decisions between reports part of what a user must evaluate.

The same launch explanation says existing Computer tool and Project permissions carry over. Users can inspect completed work, redirect a teammate or take over its desktop. Feedback is intended to shape later tasks, rather than disappear at the end of a session.

Credits are not a fixed workload allowance

The credit documentation, updated November 4, lists 10,000 monthly credits for Consumer Max and a current conversion of 100 credits to $1. Its examples below describe general Computer tasks, not measured Teammates costs or a promise about how many ongoing jobs one subscription supports.

Perplexity says task complexity changes credit use. Monthly credits do not roll over. Without additional credits or auto-refill, active tasks pause when the balance runs out; auto-refill is off by default. A completed result that disappoints the user does not automatically qualify for a refund.

For a buyer, the meaningful measure is therefore accepted work per credit, including the time spent reviewing and correcting it. A faster stream of drafts is not necessarily a cheaper finished outcome. The published ranges cannot answer that question for a particular inbox, repository or campaign.

Give the first trial a narrow brief

Editorial visualization

The setup guide recommends reviewing requested permissions and selecting only the GitHub repositories the teammate needs. It says users can choose models; new teammates default to Claude Opus 5.5. It also warns that preview behavior may change.

A useful first trial would specify one source of work, one deliverable and an explicit point for human intervention. For example: monitor a chosen issue queue, draft a daily triage note, and seek review before changing code or contacting a customer. This is an editorial suggestion, not a claim that the product enforces that boundary automatically.

Judge that trial by missed issues, incorrect actions, review time and credits consumed, alongside completed tasks. Perplexity’s cited launch material supplies no controlled reliability or net time-saving results. Until those are measured for the actual job, persistent responsibility remains a capability to evaluate—not evidence that supervision has become unnecessary.