Hire an agent for a job, not a chat.
Every morning I do the same three things.
I open my inbox and sort what needs a reply from what can wait. I check what changed overnight — a price, a post, a ticket. And I write the first drafts of the day before the day starts writing them for me.
None of that is hard. All of it is the same, every day. And every AI chat I have ever used forgets it by lunchtime.
That is the gap Perplexity tried to close on November 4, when it launched Teammates.
A teammate is not a better answer. It is a worker you hire for one job, and it keeps that job until you take it away.
The morning test
Here is how I think about it, because this is the work I actually do.
I run a small content company. Every morning, someone on the team checks the same things: what came in overnight, what moved, what needs a draft before noon. It is real work, it repeats, and it never makes it into a strategy deck because it is too boring to describe and too constant to skip.
A chat window cannot hold that job. You can ask it to draft a reply. You cannot ask it to notice, on its own, that a regular customer stopped buying — and then go read that customer’s order history and support tickets, draft an offer, and keep following up until the work is done. That is Perplexity’s own example from the launch, and it is the right example, because the hard part was never writing the offer. The hard part was noticing, remembering, and coming back.
A teammate is built for the noticing and the coming back.
What a teammate actually is
The plain version, from Perplexity’s own help pages:
You give a teammate a responsibility — triage bug reports, track your metrics, research a market — and it plans the tasks and follows through until the result is ready for your review. It keeps its role until you change it. It schedules its own follow-ups. It remembers what happened between conversations.
It watches the tools you connect, like Slack, GitHub, and Linear, and starts work when something arrives. Each teammate runs on its own isolated computer with its own browser, so it can work with apps and websites directly. Teammates in the same project share context and files, and they can message each other when new information arrives.
Perplexity’s help page gives the kind of instructions that make this concrete. “Monitor #acme-alerts. Reply to each bug report with a likely cause and a suggested fix.” Or: “Watch Linear for updates in my areas and send me a weekly summary of my work. Flag stale projects and refresh their tickets.”
Notice what those instructions have in common. They name a source, an output, and a schedule. A job, not a question.
The catch, stated plainly
None of this is open to everyone, and that matters more than the demo.
Teammates are a research preview, and the preview is for individual Max subscribers only. Not Pro. Not Enterprise by default. Max is $200 a month. The feature is on the web first, with iOS coming. Perplexity’s own help page says the feature is still under development and behavior may change.
Get Rembiq’s stories in your inbox
Join Medium for free to get updates from this writer.
New teammates run on Claude Opus 5.5 by default, and you can pick a different model per teammate. Complex projects use more of your monthly credit allocation — Perplexity says so directly — especially when agents run in parallel, work with large files, or run frequent automations.
So the honest price of a teammate is not $200 a month. It is $200 a month, plus the credits a busy teammate burns, plus the fact that you are testing a preview. If you run a business where one missed follow-up costs more than that, the math is easy. If you do not, it is not your product yet.
The companion piece: the job stays on your machine
There is a second half to this story, and it arrived a couple of weeks earlier.
On October 20, Perplexity made its Computer agent run fully local on Mac. The model, the harness, and the task itself run on your machine — on supported Apple silicon Macs, with 36 GB of memory or more recommended. With your permission, it works across local files and Mac apps like Notes, Mail, Messages, and Numbers, without sending personal data to the cloud. It launches with two local model options, Qwen 3.6 or PPLX Pro, Perplexity’s post-trained version of Qwen. Local tasks do not use Computer credits.
That is available to Pro, Max, and Enterprise subscribers — a wider door than Teammates.
Put the two together and the direction is clear. One announcement says the agent can hold a job for weeks. The other says the agent can do the job on your machine, with your files staying yours. The next leap is not a smarter answer. It is a worker that shows up every morning, and does its work where you can see it.
The test I would run
I have not run this test yet, and I am not going to pretend I have. Here is the test, so you can run it — or watch me run it and judge the result.
Give one teammate one read-only job for one week.
Pick something small and checkable. Mine would be this: every weekday at 9 a.m., read the new messages in one Slack channel, and draft replies from our own written policies — drafts only, nothing sent. Or: watch one competitor’s pricing page and note every change, with a screenshot and a one-line summary.
Read-only on purpose. No sending, no buying, no posting, no deleting. The teammate can draft, flag, and summarize. I do the sending.
Then keep a simple log for the week. Four columns: what it caught, what it missed, what it asked me, and what it did that I would not have done. At the end of the week, answer one question: did it save me more mornings than it cost me in checking its work?
That log is the whole review. A teammate that catches three real changes and asks one good question is worth keeping. A teammate that drafts ten confident replies, two of them wrong, is a morning chore with a nicer interface. The difference only shows up if you measure it — and the measurement has to be yours, on your work, not a launch video.
If you run it, start narrow. Perplexity’s own setup asks for a name or role, instructions, and a schedule. Write the job the way you would write it for a new hire: the source of the work, the expected output, when it repeats, and who sees it. Then connect only the one tool it needs, and check the permissions before you approve them. A teammate’s access follows the permissions of the plugins and projects you give it, and its credentials stay encrypted — it cannot view them. That is worth knowing before you hand it the keys, not after.
Why this matters even if you never buy Max
You may never pay $200 a month for this. Most people will not, and most should not — not for a preview.
But the shape of the product is the point. For three years, the AI industry sold us a better answer box. You ask, it answers, you ask again. Teammates, and the local Computer mode beside them, sell something else: a named worker, with a named job, that remembers yesterday and shows up tomorrow.
That changes what you ask for. You stop asking “can it write this?” and start asking “can it own this?” Those are different questions, with different answers — and the second one is the one that decides whether AI actually takes work off your plate, or just gives you a faster way to do it yourself.
Hire for the job, not the chat. Then check the work for a week, and believe the log, not the launch.
Sources: Perplexity Help Center, “Get started with Perplexity teammates” (updated Nov 5, 2026); Perplexity, “Perplexity Computer runs fully local on Mac” (Oct 20, 2026). Teammates announced Nov 4, 2026.
