AI4 min read

Perplexity brings always-on Teammates agents to the Mac, on the $200 Max plan

Persistent agents that own an ongoing responsibility and report back through Slack are now in Perplexity's Mac app, on the consumer Max tier at $200 a month. The business question for UAE and GCC IT leaders is not whether it works, but who governs the runtime, the data and the access those agents hold.

What happened

Summary of reporting by 9to5Mac

Perplexity has brought Teammates, its always-on agent feature, to its Mac app on 13 November, after the same feature appeared on the web the previous week. A teammate takes on a continuing responsibility, keeps its project context between runs and reports back, instead of starting from a fresh prompt each time (9to5Mac: https://9to5mac.com/2026-11-13/perplexity-brings-ai-agents-feature-called-teammates-to-the-mac/).

The agents can be reached through Slack, so colleagues can hand one work and ask for progress. Access sits on the consumer Max plan at $200 a month, which means the people running these agents are usually doing so on a personal subscription rather than under an enterprise agreement reached by procurement (iClarified: https://www.iclarified.com/103518/perplexity-launches-teammates-ai-agents-for-ongoing-projects).

The strategic move is Perplexity's shift from answer-on-demand search to persistent, task-oriented agents that run in the background and come back with results. Perplexity's own help centre describes the launch as a research preview for individual Max subscribers, with Enterprise Pro and Enterprise Max not included by default (https://www.perplexity.ai/help-center/en/articles/2026102601-perplexity-teammates).

For a technical leader in the GCC, the takeaway is not the launch itself but the arrival pattern: a capability that acts without being prompted, holds access to collaboration tools, and lands in an organisation through an individual expense claim well before it reaches a procurement conversation.

Read the original at 9to5Mac

The Azrty take

Always-on agents are now a priced product. In the GCC the defensible version is the one running on infrastructure you govern.

The question that matters for UAE and GCC organisations is not whether Teammates works. It is what happens when a product manager buys a $200 Max seat on a personal card and points a persistent agent at the company Slack. Perplexity's own help centre frames the launch as a research preview for individual Max subscribers, with Enterprise Pro and Enterprise Max not included by default, and says a teammate's access follows the plugins and Projects of the person who installed it (https://www.perplexity.ai/help-center/en/articles/2026102601-perplexity-teammates). The changelog that shipped the feature states that each teammate retains memory between conversations and can message coworkers in Slack (https://www.perplexity.ai/changelog/teammates-fully-local-computer-on-mac-and-voice-mode). There is no admin view of the agents, no retained audit trail and no data terms your legal team has seen. An agent that acts on its own, keeps access to Slack, and is paid for out of expenses is shadow AI in a harder form than the chatbots of 2023, because it keeps running after the person who set it up has moved to another team. It will arrive through an expense claim, not a security review.

The pattern itself is sound and the pricing is honest. Perplexity is charging $200 a month for something that does work while you are not watching (the Max plan help page lists $200 monthly or $2,000 annually, and says the tier is sold to consumers rather than as an enterprise seat: https://www.perplexity.ai/help-center/en/articles/11680686-perplexity-max), and it has proved that people will pay for it. What is missing for an enterprise is everything around the agent: identity, scope, budget, retention and an approval point.

How Azrty would approach it: adopt the operating model, not the consumer subscription. We would run an always-on agent on your own infrastructure. Each responsibility gets its own runtime with a memory service behind it, so the agent keeps what matters between conversations and nothing leaks across projects (Novamem, /software/novamem). Model calls are forced through one OpenAI-compatible gateway that holds routing, per-agent budgets and access control in a single place, so a runaway agent hits a spend cap rather than a credit card (FastLLM Proxy, /software/fastllm-proxy). Work on code lands in a self-hosted Git platform where an agent picks up jobs in isolated runs and quality gates plus an independent review decide what merges, not the agent itself (NovaForge, /software/novaforge). Anything that sends an external message or touches production passes through a workflow with a named human approver (Dhole, /software/dhole). Connectors are scoped to a service account, to named repositories and to named channels, read-only where the job allows it. Our team in the UAE does the design, the deployment and the support.

Where teams will get it wrong: treating the subscription as the whole cost and never measuring work per outcome, then discovering that autonomous agents need an owner, a scope and a log before an incident forces the point. Perplexity has shown the market exists (its own launch post for teammates is dated 4 November 2026: https://www.perplexity.ai/hub/blog/teammates-are-agents-for-complex-ongoing-work). The differentiator for an enterprise here is not the model. It is who holds the runtime, the data and the merge button.

What to do now

  1. Audit Slack and GitHub this week for Perplexity or Comet app installs and OAuth grants, and revoke anything that is not on an approved list. Where you allow agents, scope GitHub access to named repositories and Slack access to named channels.
  2. Set the policy before procurement has to: no company data to a consumer AI seat, and a governed alternative for anyone who genuinely needs a persistent agent.
  3. Run any pilot behind a gateway. Put FastLLM Proxy (/software/fastllm-proxy) in front of your model endpoints, give each agent its own budget and scope, and require a named approver for external messages or production changes.
  4. Keep agent memory and code in your own estate from day one (Novamem (/software/novamem) for memory, NovaForge (/software/novaforge) for repositories), so context and source stay inside your perimeter and stay auditable.
  5. Instrument the pilot: log tasks per agent, spend per task and each approval decision, so you can compare the cost of running the pattern yourself with $200 per seat per month before you buy more seats.
Run agents on infrastructure you controlAzrty designs, builds and runs the platform your AI sits on: your GPUs and models, one gateway for routing, budgets and access, with approvals where they matter.
PerplexityAI agentsshadow AIgovernanceGCCself-hosted

More from the Brief

Perplexity brings always-on Teammates agents to the Mac, on the $200 Max plan: the Azrty take | Azrty Brief