Dev.to AI 🤖 Ai 👁 0 📖 7 min read

Perplexity Teammates Turns AI Into Coworkers That Don't Clock Out

Perplexity just shipped the thing every AI company has been circling for months: not a smarter chatbot, but a coworker that keeps working after you log off. It's called Teammates. CEO Aravind Srinivas announced it on X

Perplexity just shipped the thing every AI company has been circling for months: not a smarter chatbot, but a coworker that keeps working after you log off.

It's called Teammates. CEO Aravind Srinivas announced it on X on 5 November 2026, and the idea is simple enough to say in one sentence — you give an AI agent a role, standing instructions, goals and permissions, and it owns an ongoing piece of work. Not a single question. A job.

I've not used Teammates myself — it launched yesterday, and it's behind Perplexity's most expensive plan — so this is a source-checked breakdown, not a hands-on review. Here's what's real, what's promised, and what I'd actually do with it.

The shift: from answering to owning

Every chatbot you've used works the same way. You ask, it answers, the conversation ends. If the work continues tomorrow, you start over and re-explain everything.

A teammate is built the other way round. You create one, give it a role — say, triaging GitHub issues, or watching a Slack channel for support tickets — and it keeps going across conversations, with memory. It watches your connected tools, acts when something changes, and hands work to other teammates when a task crosses into their territory.

That's the real difference from the agents you've already seen. OpenAI shipped Dots on 29 September. Anthropic has Claude Duet in research preview. Both point the same direction. But Perplexity's version has one detail the others don't: each teammate runs on its own isolated computer, with its own desktop, browser and cookies. It reuses your existing Projects, skills, memory, plugins and tools — but it does the work in its own machine, not inside your chat window.

And here's the number Srinivas put behind it: Perplexity has already created around 10,000 Teammates internally in the last 30 days. That's roughly 20 times the company's human workforce. His stated vision is "every org having hundreds of teammates working for each person."

Read that as a claim about where he thinks this goes, not as proof it works. Ten thousand internal agents is a big number. It's also Perplexity's own number, about its own tool, with no independent check. Still — a company dogfooding its own agent system at 20x headcount is a stronger signal than a launch video.

How it actually works

The pieces, as Perplexity describes them:

  • You define the teammate. Role, standing instructions, goals, permissions. New teammates use Claude Opus 5.5 by default, but you can pick a different model per teammate or per task.
  • It has its own computer. Built on Perplexity Computer, each teammate gets an isolated desktop with its own browser and cookies. Saved credentials are encrypted, and teammates can't view them — a real design choice, given how much damage a credential leak does when an agent has standing access to your tools.
  • It watches your tools. Connect Slack, GitHub, Linear and the like, and the teammate acts when something changes — a new issue, a message in a channel, a ticket moving state.
  • It lives in Slack. You can add teammates to Slack channels and tag them, so a human colleague hands them a task the same way they'd hand it to a person.
  • Teammates divide work. Multiple teammates can split a job between them and pass tasks to each other.

That last one is the part to watch. One agent doing one job is a productivity tool. Several agents coordinating across a shared workspace is an org chart — and org charts are where things get interesting and where they break. Perplexity hasn't shown, publicly, what happens when two teammates disagree about a task, or who arbitrates when one hands off something the other can't do. That's a question for the first real users, not the launch post.

The catch: the price of admission

Here's where the honest part starts. Teammates is a research preview, and right now it's only for individual Perplexity Max subscribers, on the web. iOS is coming soon; native mobile and desktop apps come later.

Max costs $200 a month — $167 a month if you pay annually. That buys 10,000 Computer credits a month plus a one-time 35,000 credit bonus. There's free promotional usage at first, and after that it draws on your Max credits. Enterprise Pro and Max plans don't include Teammates by default.

So the honest math is this: the most interesting version of this idea currently costs $2,400 a year, per person, and it's a research preview on one platform. That's not a casual trial. If you're already paying for Max, trying Teammates costs you nothing extra beyond credits, and you should. If you're not, this launch is a reason to watch, not a reason to subscribe — not until there's evidence from real users that a teammate actually finishes work unsupervised without creating a mess a human has to clean up.

Teammates vs Dots vs Duet

This is the third entry in the same race in about five weeks, so the comparison matters more than any single launch:

Perplexity Teammates OpenAI Dots Anthropic Claude Duet
Announced 5 Nov 2026 29 Sep 2026 Oct 2026 (research preview)
Core idea Persistent coworker agents with roles, memory, own computer Persistent agents for ongoing work Persistent agents for ongoing work
Own isolated computer Yes — desktop, browser, cookies per teammate Not the headline feature Not the headline feature
Default model Claude Opus 5.5 (changeable per teammate/task) OpenAI models Claude models
Slack integration Add/tag teammates in channels Not the headline feature Not the headline feature
Availability Max subscribers, web only, research preview Rolling out Research preview

The honest read: nobody has won this yet. All three are research previews or early rollouts, all three are describing the same future in slightly different words, and the differences that will actually matter — reliability over weeks, not hours; how they handle mistakes; what they cost at real usage — haven't been measured by anyone outside the vendors. Perplexity's isolated-computer design and Slack-native handoff are genuinely distinctive on paper. Paper is where they stay until users report back.

One more piece of context: Business Insider reported back in July 2026 that Perplexity already had an internal Teammate tool, used mainly for debugging and system monitoring in software engineering. So this isn't a product built for the announcement — it's an internal tool being opened up. That's usually a better sign than a demo built for launch day. It's also worth remembering the internal tool was used for engineering tasks the company controls tightly; customer support triage in your Slack is a different environment.

Should you care? A quick checklist

  • You're already on Perplexity Max — yes, try it this week. Give one teammate one bounded job — triaging a low-stakes GitHub label, watching one Slack channel — and check its work daily for two weeks before giving it anything that touches customers or production.
  • You run a small team on Slack and GitHub — watch closely, but don't subscribe just for this yet. The Slack handoff is the most practical part of the launch, and if it works, it's the feature that changes how a team delegates. Wait for the first real reports.
  • You manage agents or automations already — the interesting question is whether teammates replace your current glue code or just add a new layer on top of it. If you're already running automations that work, a $200/month research preview isn't a reason to rebuild.
  • You're price-sensitive, or on Pro/free — this isn't your launch. Teammates doesn't exist for you yet, and there's no announced date for it to. The free promotional usage only applies once you're on Max.
  • You just want a better chatbot — this isn't that either. Teammates is about ongoing work, not better answers. If your problem is answer quality, a new model matters more than a new agent wrapper.

What I'd actually do

If I were on Max, I'd set up exactly one teammate with one narrow job and a short leash: read-only access where possible, a dedicated Slack channel where it reports what it did, and a human check before anything it produces goes to a customer. I'd run it for two weeks and judge it on one question: did it finish work I didn't have to redo?

If I'm not on Max, I'd do nothing — except watch for the first honest reports from people who are. The launch tells us Perplexity believes persistent agents are the product. The next month tells us whether they actually work when nobody at Perplexity is watching.

The honest summary: a real idea, from a company that's been running it internally at serious scale, wrapped in the most expensive paywall in the current agent race and a research-preview label. The isolated computer per teammate and the Slack handoff are the two details worth remembering. Everything else is a promise waiting for its first independent test.

Are you on Max? If you try Teammates, I'd genuinely like to know — did it finish a job cleanly, or did you end up redoing its work?

Sources: The Verge, "Perplexity launches its own take on AI agents for work", 5 November 2026; Perplexity launch post on X by Aravind Srinivas, 5 November 2026; Perplexity help center, "Get started with Perplexity teammates"; Perplexity product page, perplexity.ai/hub/products/teammate; CellCog, "Perplexity Teammates"; AI Affairs, "Perplexity launches Teammates agents for work". The 10,000-internal-teammates figure and the $200/month Max pricing are Perplexity's own disclosures and have not been independently verified. No hands-on testing was performed for this article. No affiliate links in this article — this breaking-news piece is not the daily buyer-intent post.

📰 Read the original article on Dev.to AI

Originally published by Dev.to AI. Aggregated on AIWithGhost for educational purposes — full credit and traffic to the original publisher.