The agent wars just got a third heavyweight, and it's the one that built its business on search, not chat.
The Summary
- Perplexity launches "Teammates," always-on AI agents that handle multi-role work within a shared workspace, pulling from existing projects and memory
- Teammates layer on top of Perplexity Computer (the device-level agent system) to tackle "ongoing, large-scale work" through collaboration
- Direct competition with OpenAI's Dots and Anthropic's Claude Duet, but from a company that started with answers, not conversations
The Signal
Perplexity just made the smart move: skip the chatbot phase entirely and go straight to the work layer. While OpenAI and Anthropic spent years teaching people to talk to AI, Perplexity built a search product that already understood tasks. Teammates aren't a pivot. They're the logical next step for a company that's been answering questions with citations since day one.
The architecture matters here. Teammates sit on top of Computer, Perplexity's device-level agent system. That's not just branding. It means these agents inherit local context (files, apps, workflows) while operating as persistent collaborators in a shared workspace. You're not spawning a new agent every time you need something done. You're assigning work to an entity that remembers what happened yesterday.
"The difference between an assistant and a teammate is whether it disappears when you close the window."
Compare that to the competition:
- OpenAI's Dots: personal AI entities that learn your preferences over time
- Anthropic's Claude Duet: collaborative AI that works alongside you in real-time
- Perplexity's Teammates: role-based agents that handle simultaneous jobs with workspace memory
The key differentiator is "always-on" paired with "multi-role." Most AI agents today are either persistent OR specialized. Teammates claim to be both. One agent handles research while another drafts and a third coordinates, all pulling from the same context layer, all running whether you're watching or not.
The Implication
If Teammates deliver on simultaneous multi-role work, we're looking at the first real test of whether businesses want collaborative AI or just faster autocomplete. The workspace integration piece is critical. Companies that adopt this aren't just trying a new tool. They're restructuring who does what around agents that don't sleep.
Watch for pricing. Perplexity has always played the value card against OpenAI and Anthropic. If Teammates come in cheaper while delivering comparable output, the enterprise AI stack gets a lot more competitive. And if they don't, we'll know pretty fast whether "always-on" is a feature or just marketing.