Perplexity launched Teammates on November 4, 2026. That is in a post on the Perplexity blog. Teammates are agents that do not answer a single question, but keep taking on a piece of work. They plan the work, adjust when priorities change, and present the result for review. They launch first on the web, for Perplexity Max subscribers.
What exactly is new
Perplexity calls teammates "personal agents that can take on that long-running work". According to the post, they remember project goals, task history and shared files from one run to the next. Each new task builds on the work already done.
They also do not wait for a new question. They monitor ongoing work and act when something changes. Perplexity's example: a sales teammate notices that a regular customer has stopped ordering. It reviews that customer's order history and support issues, and creates a relevant offer or first flags what needs to be resolved. Along the way it shares updates and asks for approval.
You can add a teammate to a Slack channel. Colleagues can then hand it work or ask how things stand. Teammates can also pass work to each other.
How it is kept separate
According to Perplexity, each teammate runs in its own environment, with browser sessions separate from those of other teammates. Saved credentials are encrypted. The user can review completed work, take over the desktop, or redirect a teammate while it works.
What it means for a business owner
The most useful part of the post is not the technology, but the setup steps. When you create a teammate, you first connect the tools it needs; for GitHub you choose which repositories it can see. Then Perplexity asks you to describe the work: where it comes from, what it should deliver, and when the teammate should act.
Those are the questions you need to answer for any AI agent, whichever tool you choose:
- Where does the work come in? A mailbox, a form, a Slack channel.
- What should come out? A draft, a quote, a notification.
- When should the agent act, and when should it not?
- Which systems may it use, and which not?
Without those four answers you do not know what to expect from an agent, and you cannot check whether it does what was intended.
What you can do this week
Pick one task that comes back every week, for example following up on quotes that are still open. Write down the four answers above, on half a page. Add who checks the result before anything goes to a customer.
Perplexity builds that check in as well: teammates present their result for review, and in Perplexity's example the teammate asks for approval before it finishes. That is a good standard for any agent you consider. Ask a provider: where do I see what the agent plans to do, and where do I say yes or no?
That half page is useful right away, whether you try a tool like this yourself or ask someone to set it up for you.
Why this will come back
At the end of the post, Perplexity says that teammates' access and features will expand in the coming weeks. So more is coming. The questions stay the same: where does the work come from, what should it deliver, when may the agent act, and who checks it.
Want to know what that looks like in your business? Read how we set up AI automation.