Perplexity Teammates Coordinate Ongoing Work Across Multiple AI Agents
Perplexity presents Teammates as work-focused AI agents that turn ongoing responsibilities into sequences of tasks and hand work between agents. Examples include preparing inventory reorders, coordinating a code fix with QA and handling customer-support tickets; proposals and work products are shown at review points for people. Built on Computer’s projects, skills and memory, Teammates can also apply feedback to future work and collaborate with colleagues in Slack, the company says.
Responsibilities give agents a sequence of work
Teammates are personal agents designed for work. Perplexity says a user can give one a goal or responsibility, after which it plans tasks, coordinates agents across parallel work, and can pass work between them around the clock. The store-manager example shows how that responsibility might be framed: not as a single question, but as an ongoing area of work involving stock, reorders, listings, customer emails, and shipping delays.
Kelly asks the store-manager teammate to handle inventory and reorders. It says it will compare stock with 90 days of orders and line up reorders with suppliers. Its checklist includes scanning inventory, flagging low-stock styles, reading an email from a supplier, checking lead times, and drafting reorders for Kelly’s review. Shopify and Gmail are shown as connected.
The sequence combines information gathering with preparation of a proposed action. The teammate does not present the reorders as approved; it drafts them for Kelly to review. In this example, the responsibility involves more than identifying low stock: the teammate uses order history and supplier information to prepare a next step for a person to inspect.
Handoffs connect the work, with review still visible
The software example makes a handoff explicit. An Engineer teammate reports fixing a password-reset crash and opening a pull request, then says it is sending the work to a QA teammate. QA reports testing the fix on desktop and mobile, says the crash no longer reproduces, and marks the pull request ready for review. A further screen presents the pull request and test results to a person.
That sequence presents coordination as part of the work: one agent makes a change, another tests it, and a person receives the outcome for review. Other examples also show work prepared for review, including inventory reorders and an ad budget proposal. The examples do not establish that every action requires approval; they show review points in these particular workflows.
Activity panels sketch the same model across business functions. An Ads Manager reviews competitor ads, drafts creative, and sends a budget proposal for approval. An e-commerce manager checks supplier lead times, adjusts reorder quantities, and queues back-in-stock emails. An Engineer watches an error rate before marking a pull request ready for review. These panels emphasize sequences of actions rather than isolated answers.
Customer support makes the division between routine handling and individual attention more concrete. A panel says the teammate is handling 214 tickets from the previous day. It shows cases involving exchanges, replacements, address changes, refunds, and reshipments, while noting that 18 need a personal reply. The display presents the teammate as handling many requests and setting some aside for individual attention.
Feedback and Slack are presented as separate capabilities
Teammates build on the projects, skills, and memory already in Computer. The product also says that feedback can shape future work. In one exchange, an Ads Manager teammate says it has prepared an expanded campaign and sent a creator’s contract extension to Legal. Emmanuel tells it to send contract changes to Legal every time and says it need not ask next time. The teammate acknowledges the instruction.
A separate Slack demonstration shows colleagues working directly with teammates in a channel named #legal. The Ads Manager sends Legal an expansion plan and draft contract extension for review. A Legal teammate then reads the extension, compares it with standard terms, flags non-standard clauses, and drafts a redline. The source does not show that Emmanuel’s earlier instruction caused or governed this Slack workflow; the two demonstrations illustrate feedback and Slack collaboration independently.
Perplexity describes Teammates as proactive: they can run automations and keep work moving on a person’s behalf. The examples give a sense of the kinds of work involved—campaign management, inventory, customer support, and code—but none depicts an automation starting or continuing without a request. The proactive claim is therefore a stated capability, distinct from the workflows shown on screen.