Perplexity’s new Teammates product turns its AI assistant into a set of persistent workplace agents, a shift from asking a chatbot questions to assigning software continuing responsibilities. The company describes these agents as digital coworkers that can take on projects such as advertising campaigns, coding work and routine daily tasks, then report progress through services including Slack. Unlike a conventional chat session, an agent can retain relevant context and operate in the background, potentially handling several jobs at once. The launch puts Perplexity squarely in the contest to make artificial intelligence useful not merely as a source of answers, but as an active layer in how organizations get work done. That promise reflects a broader industry effort to move generative AI from novelty and experimentation into repeatable business workflows.
The product builds on Perplexity’s existing identity as an AI-powered search and answer service, but its ambition is different: the agent is meant to act, not just explain. A user can hand over a goal, such as preparing campaign materials or working on a coding project, and expect the system to make progress beyond a single exchange. Persistent context matters because workplace tasks unfold over days, depend on prior decisions and often require updates rather than a one-time response. Perplexity says Teammates can remember context, collaborate and send updates in Slack, placing the agent closer to the flow of work employees already use. The practical test will be whether these features save time without adding another tool employees must constantly supervise.
That distinction is central to the emerging category known as AI agents. Most chatbots respond to a prompt and stop; an agent is designed to break a goal into steps, use software tools, and continue working toward completion. In theory, this can take routine coordination off a person’s plate, from gathering information to drafting or updating materials. In practice, the systems can misunderstand ambiguous instructions, make errors or take actions that need human approval. Persistent memory may improve continuity, but it also raises questions about what information is retained, who can inspect it and how users correct outdated or mistaken assumptions. The product’s value will depend less on impressive demonstrations than on reliable behavior under ordinary conditions.
Perplexity is entering a crowded market where major technology companies are trying to turn AI assistants into business products. Microsoft, Google, OpenAI and specialist software vendors are adding agent-like capabilities to tools that already handle documents, communication, coding and customer operations. That distribution is a major advantage: companies may favor systems embedded in their existing platforms over a new destination service. Perplexity, by contrast, can compete on its search heritage and on the idea of agents that work across tasks, rather than simply assisting inside one application. The challenge is to demonstrate secure integrations, dependable results and clear administrative controls, while convincing organizations that a newer provider can meet demanding workplace requirements.
For businesses, the most attractive use cases are likely to be bounded tasks with clear goals and reviewable outputs. A marketing team might ask an agent to assemble campaign drafts; a developer could delegate routine coding work; an operations group might use one to summarize recurring updates. Such delegation could help small teams stretch limited staff and free experienced employees for decisions that require judgment. But an agent’s output is not automatically trustworthy, and organizations will need processes for checking facts, permissions and changes before they reach customers or production systems. The near-term model is therefore less “replace the employee” than “give the employee another worker to manage,” with productivity gains depending on good oversight and task selection.
The Slack connection highlights both the promise and the friction of this approach. Updates in a familiar collaboration channel can make an agent’s work visible and let colleagues respond without switching applications. Yet placing automated coworkers in the same channels as people can also create noise, unclear responsibility or misplaced confidence in a polished status report. Teams will need to decide which work agents may perform independently, when they must ask for approval and how their actions are logged. For workers, that may mean learning to write goals, provide relevant context and evaluate results, rather than simply composing clever prompts. These are management and workflow changes as much as technical ones.
Perplexity has not, in the announcement details available, established that Teammates can handle every complex assignment safely or independently; the launch is better read as a bid to shape expectations for workplace AI. The crucial measures will be accuracy, uptime, security, integrations and the cost of reviewing an agent’s work, alongside whether it delivers measurable time savings. Persistent agents could become a meaningful new interface to software if they reliably coordinate tasks across tools, but they could also remain an appealing demonstration if supervision consumes the time they promise to save. For users, the opportunity is practical assistance that follows work over time; for employers, it is a new balance between automation and accountability. Teammates makes that bet explicit: the next phase of AI competition is about who can turn a capable model into a dependable colleague.
