Perplexity has launched Teammates, persistent AI agents that take on ongoing work instead of waiting for the next prompt. The new feature sits inside Perplexity Computer and gives each agent a named role, a memory of past work and the tools to keep a project moving. Teammates are available first to Max subscribers on the web, with iPhone and desktop apps to follow, and they mark Perplexity’s biggest shift yet from answering questions to owning jobs.
What a teammate actually does
A teammate starts with a job, not a question. Users name it for the work it should own, connect the tools it needs, such as Slack, GitHub or Gmail, and describe what it should deliver and when it should act. Each teammate plans the work, coordinates Computer sessions and other teammates, checks its own results and returns work for review.
Teammates build on what Computer already knows. Each one inherits a project’s files, goals, skills and memory, so a sales teammate can reuse account research, follow an outreach skill and run a weekly automation that reviews the pipeline. It spots stalled conversations, drafts follow-ups for approval and carries what it learns into the next week’s work.
Each teammate runs in its environment, with browser sessions separate from other teammates and credentials encrypted and entered securely. Users can watch the work, take over the desktop or redirect the agent as it goes. The controls matter because teammates touch accounts, and Perplexity has spent the year making Computer acceptable in regulated workplaces, including a government security certification.
Proactive, not prompted
Teammates also act without being asked. They monitor ongoing work and follow up as conditions change, investigating a problem before a human notices it. A sales teammate, for example, notices when a regular customer stops buying, reviews order history and support issues, then drafts a relevant offer or flags the problem before anyone reaches out.
They also learn from corrections. Tell an operations teammate that an agency needs two weeks’ notice, and it builds that lead time into future schedules. That memory is the point: each task builds on the last instead of starting fresh, so the agent gets more useful the longer it owns a job.
Teammates that talk to each other
Teammates are built to work in teams, not alone. One can join a Slack channel, where colleagues hand it work or ask for a status update. Teammates can also pass work to each other: one finds bugs on a website and hands its findings to another, which prepares the fixes.
Perplexity pitches the same pattern across four functions. Marketing teammates manage posts, campaigns, ads and reviews while watching competitors. Sales teammates hunt opportunities in inquiries and past purchases. Support teammates triage complaints and draft replies. Operations teammates chase deadlines, vendors and paperwork, following up when a deadline is 72 hours away.
Our read
The pitch is a challenge to how assistants work today. Chatbots wait for prompts; Teammates are supposed to own outcomes, a harder promise, and a more useful one if it works. Perplexity bets the next step after a good answer is a reliable colleague, and users will pay for work that continues when they stop typing.
The timing is not accidental. Anthropic has put Claude inside its users’ bank accounts with a read-only finance feature, and OpenAI keeps pushing agents deeper into work. Perplexity’s answer is different: instead of one assistant that does everything, it offers a team of named agents, each with a narrow job and a memory of how it went last time.
The catch is the price. Teammates launch only for Max subscribers, Perplexity’s $200-a-month tier, while Free and Pro users must upgrade. Max includes a monthly credit allowance, with extra usage free during a launch promotion. That positions Teammates as a power-user product first, and its real test is whether teams find $200 a month cheaper than the hours it saves.
Why It Matters
If it works as described, the assistant stops being a tool and becomes staff you manage. That changes who buys AI: not just people chasing faster answers, but teams paying for work that happens while they do something else. It raises the stakes on trust: an agent that acts alone needs permission, oversight and a record of what it did.
Perplexity is also racing rivals moving the same way. OpenAI’s workspace agents, Microsoft’s Copilot Cowork and Anthropic’s Claude Cowork all chase persistent-agent work, and every major lab is adding memory and multi-step planning. Teammates is Perplexity’s bid to stand out with persistence and teamwork, rather than raw model power it does not own.
Background
Perplexity launched Computer, its multi-model agent platform, in February, and has extended it through the year with automations, e-commerce tools and a voice mode that listens while it talks. Teammates is the next layer: the same agent, given a lasting job instead of a single session.
Source: Perplexity’s Teammates announcement and Digital Trends’ report
0 Comments