Perplexity Launches Teammates, Always-On AI Agents for Ongoing Work
Perplexity has introduced Teammates, work-focused AI agents that the company describes as always-on and able to take on multiple roles within a workspace. The Verge reports that using Teammates requires a subscription to Perplexity's $200-per-month Max plan.
Teammates draw on existing projects, skills, and memory from Perplexity's tools. Users can assign each Teammate a different job, and those jobs can run simultaneously. Progress is visible through individual threads, which are running logs of each job, in the Perplexity app. Users can also interact with Teammates in Slack.
Teammates integrate with Perplexity Computer, a system that can deploy agents on a user's device to complete tasks. Perplexity says Teammates build on Computer by adding the ability to collaborate and to complete ongoing, large-scale work. The company's official X account states that Teammates take on jobs and work proactively using Computer tasks, skills, memory, and tools. Perplexity introduced Teammates on X.
On November 4, 2026, Perplexity published a blog post titled "Teammates are agents for complex, ongoing work." In it, the company describes Teammates as coordinating AI agents and managing ongoing projects.
The launch follows earlier work on shared workspaces. On October 12, 2026, Perplexity launched Projects, in which people and agents work together on long-running tasks inside Perplexity Computer. Perplexity's blog covered that release. Separately, Bloomberg reported on June 2, 2026 that Perplexity was building a platform that splits AI work between personal computers and cloud-based servers to manage demand and ease strain.
Teammates are persistent, multi-job agents with a thread-level audit trail and a Slack interface. They run on an agent runtime, the software layer that executes an agent's steps, and that runtime can execute on a local device.
In my view, the cost of entry is the most telling detail. At $200 a month, Teammates sit in the professional tier rather than the consumer one, which suggests the target is individual power users and small teams running multi-step workflows rather than broad enterprise deployment.
In my view, the open questions for teams adopting this kind of agent are about supervision rather than raw capability. Running several jobs in parallel multiplies the intermediate output a person must check, and a thread view is only as useful as the review habits around it. Memory and tool access, which Teammates use to work proactively, also widen what an agent can touch on a user's behalf, so permission scoping and data boundaries will matter as much as task completion rates.
The broader context here is a shift in how AI products are packaged. The chat window is giving way to agents that hold standing responsibilities, report back through the tools people already use, and carry context across projects. If that model works, AI could move from answering questions toward carrying a share of recurring work, and the measure will be how reliably those agents finish long jobs under human oversight.
