Perplexity introduces autonomous AI teammates for workplace tools

Perplexity has rolled out a new feature called Teammates, aiming to turn conversational AI into independent software workers that complete background tasks inside everyday office software.

Announced on 5 November 2026, the tool moves past standard search and question answering by giving each agent its own designated role, memory across past interactions, and a dedicated virtual environment known as Perplexity Computer.

How the agents operate

Instead of waiting for a user to prompt them inside an open browser tab, Teammates can be connected to internal business systems like Slack, GitHub and Linear. When activity occurs in those channels, the assigned agent detects the event, interprets the context, and gets to work.

Perplexity designed the system to manage complex instructions by dividing them into individual steps. A lead agent can launch subsidiary agents in parallel on separate virtual computers to handle different parts of an assignment simultaneously. The primary agent then collects and checks those contributions before presenting the final result back to human workers.

By default, each agent runs on Anthropic's Claude Opus 5.5, though users can swap in different underlying models depending on the specific job.

Access and security controls

Perplexity launched the tool as a research preview for individual Perplexity Max subscribers using the web app, with mobile support planned. The tier carries an access fee of 167 US dollars per month.

To address enterprise security concerns, Perplexity isolates each agent on its own virtual machine with a private browser profile and separate cookies. Connected login credentials remain encrypted so the models cannot view raw account secrets.

What this means for Australian businesses

Small and medium Australian firms often struggle with administrative handoffs between systems, such as tracking project updates in Linear, following code merges, or triaging routine operational questions in Slack. Dedicated agent setups offer a way to assign recurring monitoring work without hiring specialised technical staff.

However, business owners should consider practical boundaries before connecting autonomous agents to operational software:

• Start by testing the agents on internal discussion channels or non-critical tracking tasks before giving them write access to production workflows.
• Check data boundaries and review what company files or customer records the connected tools can reach.
• Audit the subscription cost against the actual staff hours saved, especially given the monthly subscription fee per user.
• Establish human review points so that outputs produced across background threads receive sign-off before reaching external clients.