Perplexity Teammates Brings Persistent Work Agents to Max on the Web

Perplexity began rolling out Teammates to Max subscribers on the web on November 5. CEO Aravind Srinivas announced the launch as a move toward persistent agents that take responsibility for ongoing work. Native mobile and desktop apps are coming later, with no release date in that announcement.

The change is relevant to work that outlasts a single conversation. A useful assistant can help investigate a bug when asked. An ongoing responsibility also requires noticing the next report, remembering what changed and knowing when to involve a person.

What the launch promises

The official announcement says Teammates use existing Computer tasks, Projects, skills, memory and tools. They can monitor ongoing work, respond to changes, collaborate with other teammates and message coworkers in Slack. Perplexity also says they retain past work and apply feedback to future tasks.

That puts continuity at the center of the product. Users who already have relevant work in Computer may have less context to assemble from scratch. Whether that context is complete, current and appropriate for a particular job still needs checking.

The launch posts establish the announced capabilities and rollout, rather than independent evidence of reliability across long-running assignments. There is no measured task-success rate in those posts. Treat the first deployment as an evaluation of a specific responsibility, not a reason to delegate every unfinished project at once.

Persistence makes permissions more important

In its follow-up on controls, Perplexity says each teammate runs in an isolated environment and follows Computer's existing tool and Project permissions. Users can review its work, take over the desktop or redirect it while it works.

Those controls raise practical questions worth answering before a rollout. Which systems can the teammate read? Which can it change? Can it send a message on its own, and to whom? What should happen when the original goal is complete or someone changes the plan?

A sensible first assignment would be a narrowly defined monitoring job with a clear output. For example, ask it to track a chosen set of bug reports and prepare a triage summary for review. Specify what counts as a new development, where the supporting evidence should appear and which decisions remain with a person.

Then test the awkward cases. Introduce an outdated document, a duplicate report and a later instruction that changes the priority. Check whether the teammate preserves the distinction between an old assumption and the current request. Review the history before expanding its access or allowing actions with larger consequences.

Measure useful work, not activity

For a persistent agent, more messages and more completed subtasks can be misleading measures. The important questions are whether it caught something that mattered, saved someone time and avoided unnecessary or incorrect actions.

Track the time spent reviewing its work alongside the time saved. Include usage and credit consumption in that review rather than assuming ongoing monitoring has negligible cost. Teams should check the terms and controls available in their own account before putting a job on a continuing schedule.

Teammates makes Perplexity's move toward ongoing delegated work explicit. The strongest test will be a modest responsibility handled consistently, with a clear record and a person who can change its direction.

I'm Paras. I work at the intersection of AI, engineering and distribution.

I help AI teams explain new products through the workflows, limits and evidence that matter to the people using them.

Book a 30-minute call.