The race to build AI that does real work, not just answers questions, took another step this weekend. Perplexity has launched Teammates, a feature that lets users hand an AI agent an ongoing role, such as watching bug reports or tracking a market, and have it plan, carry out and report back on the work over time. It is available now as a research preview for individual Perplexity Max subscribers on the web, according to IT Brief.

Here is how it works, who can use it, and what to keep in mind.

From prompts to roles

Most AI chat tools work one request at a time: you ask, it answers, and the job is done. Teammates is designed to take ownership of work that continues. A user assigns a role, and the agent plans the tasks, sees them through and returns results for review.

When the goal is bigger, a teammate can break it into separate tasks, figure out which ones can run at the same time, and start additional agents to handle them. Each agent works on its own computer and finishes its piece before the results are collected and checked. After the work is done, a teammate can keep monitoring the project and make updates when something changes.

Built on Perplexity Computer

The feature builds on Perplexity Computer, the company’s existing system for using tools, files and code. Teammates adds support for ongoing assignments, larger projects and coordination among several agents.

Within a project, teammates share context, files and instructions while keeping separate roles, and they can message each other when new information arrives. Output can include reports, pull requests or configuration changes, all presented for a human to review.

What it can connect to

Through Perplexity’s plugin system, teammates can connect to services including Slack, GitHub, Linear and Gmail. Users ask a teammate to connect a service from inside the chat, approve the permissions, and then tell it what to watch.

For apps without a dedicated connector, teammates can also use their own desktop and browser to work with websites directly. Users can step in and steer the work at any point.

Who can use it, and what it costs

Access is limited for now:

  • It requires an individual Perplexity Max subscription.
  • Pro subscribers are not included.
  • Enterprise Pro and Enterprise Max are not included by default.

Usage runs on Perplexity’s credit system. Max plans include a monthly credit allowance, and more complex work can use more credits, especially when several agents run in parallel, handle large files or codebases, or run automations frequently. Users can also choose which AI model a teammate uses.

Security guardrails

Teammates inherit the permissions of the plugins and project settings they are given. Saved credentials are encrypted, and each teammate runs on an isolated computer with its own desktop, browser session and cookies. A teammate’s chat is visible to its owner, while shared chats are read-only for others.

What to keep in mind

Agent tools like this are powerful, but they work best with clear instructions and a human checking the results. A few sensible habits:

  • Start with a narrow, low-risk role before handing over anything important.
  • Grant only the permissions a task truly needs.
  • Watch credit usage, since parallel agents can add up quickly.
  • Review outputs before acting on them, especially changes to code or settings.

The bigger picture

Teammates is part of a broader shift across the AI industry toward systems that manage persistent assignments across many tools while keeping a person in the loop before anything is finalized. Expect more products that look less like chatbots and more like digital coworkers.

The wrap

Perplexity is betting that people want AI that keeps working after the chat ends. For now it is a preview for Max users, but it offers a clear look at where everyday AI tools are heading.

Image: Backlit Laptop Keyboard by Gre123gr via Wikimedia Commons, CC0 1.0