Perplexity's AI Agents Keep Working When You Log Off

Perplexity's new 'Teammates' feature lets AI agents coordinate on tasks, remember context, and keep working even after you close the tab.

Perplexity has introduced a significant evolution for its AI platform with a new feature called Teammates. According to reporting from The New Stack, this system allows users to deploy multiple AI agents that can work together on complex tasks. The most notable capability is their persistence: these agents continue their work even after a user closes their browser tab. They can remember the context from previous sessions and operate asynchronously without needing constant human prompting. This marks a departure from the typical request-and-response model of most AI assistants, moving towards a system where AI can be assigned a project and left to execute it over an extended period. The feature is designed to handle collaborative, multi-step processes, fundamentally changing the user's relationship with the AI from a simple tool to a persistent digital collaborator.

The core innovation behind Teammates lies in its underlying infrastructure, which is built to manage long-running, stateful processes. Unlike a standard chatbot session that ends when the user leaves, these persistent AI agents require a backend that can manage their state, track progress, and coordinate their actions over time. The New Stack highlights the key engineering challenges involved, such as keeping everything in sync when multiple agents work concurrently. This involves managing a shared project state, where one agent's output becomes another's input. Furthermore, the system must handle complex permission models, ensuring different agents have the appropriate access levels for their specific roles within a task. This architecture is more akin to a distributed system or a workflow orchestration engine than a simple large language model API, requiring robust job queues, state databases, and sophisticated logic for coordination and error handling.

This development places Perplexity at the forefront of a broader industry trend toward more autonomous AI systems. While many companies have demonstrated agent-like capabilities, most have been confined to single-user, single-session interactions. The introduction of persistent, multi-agent, asynchronous workflows represents a major step towards creating truly autonomous systems that can manage complex projects with minimal human oversight. This capability moves beyond simple task automation, like writing an email, into the realm of project execution, such as conducting market research over several days by gathering data, analyzing it, and compiling a report. This shift aligns with the long-term vision for AI agents as proactive partners rather than reactive assistants, a goal shared by research labs and startups across the AI landscape. The emphasis on coordination and persistence is a key differentiator in a crowded market.

For technical leaders and development teams, the launch of Teammates serves as a clear signal of where the industry is headed. The immediate implication is the need to think about how to integrate and manage these persistent AI agents within existing workflows and systems. CTOs and infrastructure teams must consider the new operational complexities, such as monitoring, debugging, and securing autonomous processes that run 24/7. Security teams, in particular, face a new paradigm; an agent with persistent access to internal systems or data represents a novel and potentially significant attack vector that requires new security models and oversight mechanisms. The key questions to watch are how platforms like Perplexity will provide tools for governance, observability, and control, allowing businesses to safely harness the power of these autonomous systems without introducing unacceptable risks. The ability to audit an agent's actions and intervene when necessary will be critical for enterprise adoption.

This move from session-based, synchronous AI to persistent, asynchronous agents is a major architectural shift. For developers and infra teams, it signals the need to build systems that can manage long-running, stateful AI tasks, including complex state synchronization and permission models.

Persistent AI agents can automate more complex, multi-step business processes that were previously too long for a single session. This could unlock new efficiencies in research and project management, but also introduces new operational and security oversight challenges for companies.