Perplexity launches Teammates, always-on AI agents for work

What happened

Perplexity has launched its own work-focused AI agents, called Teammates, placing the startup in direct competition with OpenAI and Anthropic. According to The Verge, the agents are always on and can take on several roles within a single workspace.

A central feature is continuity. Teammates can draw on projects, skills and memory from the Perplexity tools a user already works in, so each agent does not start from scratch. They are also designed to work with Perplexity Computer, the company's AI system that can deploy agents on a user's device.

The reporting available here is a summary, so details such as pricing, availability and rollout timing were not in the material reviewed.

Why it matters

The significance lies less in the feature list than in the architecture. An agent that carries context from existing projects and memory moves AI from a tool people query on demand to a participant in ongoing work. For organisations, the question shifts from "which assistant should we buy?" to "how is our institutional knowledge structured so that agents can act on it safely and accurately?"

Agents that run across a shared workspace also change how work is handed over. When several roles can be assigned to automated teammates, approval routes, escalation paths and accountability need redesigning, not just licensing. Digital transformation leads should treat this as an operating-model decision as much as a software purchase.

The Renascence take

Most coverage will focus on the head-to-head with OpenAI and Anthropic. The more consequential question is what happens to the people who work alongside these agents every day.

Memory-enabled agents are only as useful as people's willingness to trust them. Colleagues and customers need to see what an agent remembers, why it acted, and how to correct it, or continuity becomes a liability rather than a convenience. Treat the agent's context as a service-design artefact: map who can see what, where handoffs happen, and where a human must still make the call. Operators who design those moments deliberately will earn adoption, while those who simply switch the agents on will inherit confusion dressed up as productivity.