Two thousand dollars a year. That's the price of admission to Perplexity's newest product, assuming you pay up front. Go month to month at $200 and the bill climbs to $2,400.
For that you get a crew of AI workers that never log off. Perplexity calls them Teammates, and they launched this week for subscribers on its top-tier Max plan.
The pitch is easy to say and hard to verify: give an agent a standing job, walk away, come back when it needs you.
What a teammate actually does
Teammates live inside Perplexity Computer, the company's cloud-based assistant for multi-step tasks. Computer already handles a chain of chores when you ask. Teammates are meant for the chores that never end.
You assign each one a role. One might watch a stream of competitor news. Another might keep a research file current. A third might draft the weekly summary nobody on your team wants to write. They all run at once, in the background, around the clock, and they tap you on the shoulder only when a human decision is required, as Engadget reported.
Each teammate gets its own entry in Computer's sidebar. Open it and you see a thread: what the agent is doing now, what it finished earlier. The same thread is where you hand over extra work, redirect it, or ask for a status update. It's closer to messaging a colleague than filling out a form.
Perplexity also says the agents can pull in your existing projects, skills and memory from the tools you already use there. They can be reached in Slack, too, which matters more than it sounds. Plenty of people will never open a dedicated AI app on a Tuesday afternoon, but they'll answer a message in a channel they already live in.
Availability is narrow: the web and iOS, and only for Max.
The manager-of-managers problem
This is the part of the design that deserves slow reading. Teammates don't only take orders from you. They coordinate with one another, and they can supervise one another too.
As the workload grows, a teammate can recruit additional teammates and plan out what the newcomers should do. Your "team" isn't fixed by you. It grows to fit the pile of work, and the agents decide how.
That's a genuine step past the usual chatbot arrangement, where a person prompts, a model answers, and the person decides what happens next. Here you set a goal and the software draws its own org chart underneath it. Picture a kitchen where the head chef can hire extra cooks mid-service, and you only hear about it when a plate comes back wrong.
It's also where things could get strange.
Think about delegation among people. A good manager hands off work but also knows how a project is going, who is stuck and which corner got cut. That awareness comes from friction: meetings, questions, the occasional awkward silence. An agent hierarchy that runs 24/7 removes most of that friction by design. Which is the selling point, and also the risk.
If teammate A hires teammates B and C, and B quietly misreads the assignment, who notices? Perhaps A does. Perhaps the thread you check on Friday morning looks tidy, with a long list of completed steps, and the mistake sits three layers down. Perplexity hasn't said how it surfaces errors inside these chains, or how much of the sub-agent activity you can actually inspect. Those details will matter more than any feature list.
Notifications are the other pressure point. The promise is that you get pinged when a decision is needed, but "needed" is a judgment call the agent makes for you. Set the bar too low and you've hired an assistant who interrupts you twelve times an hour. Set it too high and it commits you to something you'd have vetoed.
There's also a cost question the company hasn't answered publicly. Agents that spin up other agents consume compute, and a flat $200 a month has to cover whatever they choose to do overnight. Whether heavy users hit limits, and what happens when they do, is unclear.
I'd call the supervision loop the whole product. The rest is packaging. Get the oversight right and Teammates look like a real shift in how a solo professional or a small team gets work done. Get it wrong and you've bought an expensive way to produce confident-sounding mistakes at 3 a.m.
A crowded field, a steep price
Perplexity isn't alone here. As The Verge noted, the new agents arrive alongside similar offerings from OpenAI, called Dots, and from Anthropic, called Claude Duet. All of them promise multiple AI workers handling tasks in parallel, around the clock.
The notable difference, per The Verge, is price. Perplexity's version requires the $200 Max plan, which the outlet described as much more expensive than the rivals' approach.
That's a bold place to plant a flag. Perplexity built its name as a search-style answer engine, and the jump from a free or cheap tool to a $2,000-a-year commitment is steep. It narrows the audience to people who already treat AI as a core work instrument: analysts, consultants, founders, researchers who bill by the hour and can do the math on time saved.
For them, the math might work. Reclaim ten hours a month and $200 is easy to defend. For everyone else it's a hard sell, especially when the product is new enough that nobody has a track record to point to.
There's some strategy in it too. Computer, Perplexity's agent system, has been expanding for months. The Verge's coverage trail shows a Portable Computer announced in August that runs entirely on device, and a version of Computer that can run fully offline on a Mac, with a catch: you need a lot of RAM. Teammates are the next rung. Perplexity is stacking capability onto one platform and hoping it becomes where people do their daily work, not just where they look things up.
What the CEO is betting on
Aravind Srinivas, Perplexity's chief executive, laid out the long game in a post on X. "Over time, we envision every org having hundreds of teammates working for each person," he wrote.
Hundreds. Per person.
It's an extraordinary number, and worth sitting with. Today most professionals might juggle a few software assistants, each with a narrow job. Srinivas is describing a world where one employee directs a workforce larger than most departments. Whether that sounds inspiring or slightly alarming depends on where you sit, and on what happens to the people those agents are standing in for.
Nobody at Perplexity has said how many Teammates a Max subscriber can run today, or what the practical ceiling looks like. For now, the hundreds are a destination, not a feature.
What to watch from here
Three things will tell us whether this works.
First, reliability over time. A demo of an agent finishing a task is easy. A teammate running unattended for six weeks without drifting off course is a different test, and early adopters will report on it fast.
Second, pricing pressure. If OpenAI and Anthropic keep their offerings cheaper, Perplexity will have to prove its version does something the others don't, or lower the bar for entry. A rollout beyond Max, or to Android, would be the first sign the company wants a wider audience.
Third, trust. The first widely shared story of a teammate making a costly call without checking in will shape how cautiously everyone uses these tools. It will happen to someone. The only open question is how Perplexity responds when it does.