Perplexity Launches AI Agents for Work: What 'Teammates' Means for Founders

Perplexity has just launched AI agents that don't wait for your next question, they keep working on a project until it's done, check back in when priorities shift, and show you what they've done before you sign off. It's called 'Teammates,' and it's a meaningful shift from the chatbot model most of us are used to.

On 4 November 2026 (updated 6 November), Perplexity announced Teammates, a system of personal AI agents built for 'complex, ongoing work.' The pitch is simple on paper: instead of a chat window that forgets everything the moment you close the tab, a teammate plans a project, adapts when your priorities change midway, presents its results for you to review, and lets you inspect or redirect its work at any point.

It's currently available on web to Perplexity Max subscribers, with an iOS release planned. If you're on Free or Pro, Perplexity says you can upgrade through Settings → Account → Subscription. Notably, the announcement doesn't mention a separate price tag just for Teammates, it rides on your existing Max subscription.

A normal AI search or chat is one-shot: you ask, it answers, the thread ends. A teammate is designed to run continuously across sessions. Perplexity says a teammate can coordinate Computer sessions (their agent that operates a virtual computer) and even coordinate other teammates, so multiple agents contribute to one ongoing project rather than each handling an isolated task.

This lines up with something Perplexity published the very next day, 5 November 2026, on AI agent orchestration, describing how independent tasks can run in parallel and get coordinated through an orchestrator. In plain terms: instead of one AI doing everything sequentially, you get several narrow agents each doing their bit, with something sitting above them making sure the pieces fit together. That orchestration layer is the real engineering problem, and it's also where most DIY attempts at 'AI agents for my business' quietly fall apart.

Perplexity's own 3 November guide draws a clear line here, and it's worth knowing before you pick a tool:

So it's a ladder: quick task, then a longer single session, then an ongoing agent that manages the whole workflow and reports back. Founders evaluating this shouldn't assume 'AI agent' means one thing, the capability and the risk level both rise as you move up that ladder.

Perplexity's own pricing page (13 October 2026) lists Max at US$200 per month. At roughly ₹83.5 to the dollar, that works out to about ₹16,700 a month per person, before any taxes or foreign exchange charges your card issuer adds. That's a currency conversion, not an India-specific price, Perplexity has not published INR pricing or GST handling.

For teams, Perplexity lists Enterprise Pro at US$40 per user per month (US$400 annually) and Enterprise Max at US$325 per user per month (US$3,250 annually). None of these figures are confirmed as final or exclusive for Teammates access specifically, and Perplexity hasn't disclosed any Indian companies using it, Indian-language support, or local data residency. If you're evaluating this for an Indian SMB, budget in dollars and expect the real monthly cost to land somewhere above the raw conversion once fees are added.

This is the question that should worry you more than the price. The moment an agent browses the web or operates software on your behalf, it's exposed to prompt injection, where hidden instructions buried in a webpage or document trick the agent into doing something you never asked for. Perplexity clearly takes this seriously, on 2 November 2026 it published 'BrowseSafe++,' an open benchmark plus two lightweight detectors built specifically to catch prompt injection in browser agents.

A day after the Teammates launch, Perplexity also improved its Agent API effort levels. The new 'xhigh' level can execute code in a sandbox and sustain long tool-use loops across many sources, and a new 'ultra' level runs three agents in parallel, each researching the task independently before results get combined. This tells you something important: giving an agent real autonomy over a longer task isn't a flip of a switch, it needs sandboxing, parallel checking, and injection detection built in. That's engineering work, not a settings toggle.

The real question isn't 'can an AI do this task,' it's 'what happens the one time it does the task wrong, and who catches it before a customer does.'

Don't start by letting an agent run your whole marketing function or manage customer refunds unsupervised. Start with ongoing work that's valuable but low-risk if it's slightly wrong, and easy for a human to spot-check in two minutes.

Notice the pattern: research, drafting, monitoring, and summarising. These are jobs where the agent saves hours of grunt work, but a person still reviews before anything reaches a customer, a vendor, or a public channel. That's exactly the 'inspect and guide' model Perplexity is describing with Teammates, and it's the right instinct whether you use Perplexity or build your own setup.

We see three recurring mistakes when founders try to wire up AI agents themselves. First, no orchestration layer, so multiple AI tools end up stepping on each other or duplicating work because nothing is coordinating them. Second, agents connected to nothing real, they produce a nice output in a chat window but it never actually touches your CRM, your inbox, or your inventory system, so someone still has to copy-paste it across manually. Third, no review checkpoint, an agent gets given too much autonomy too early, and the first time it gets something wrong, it's already gone out to a customer.

None of these are reasons to avoid agents, they're reasons to build them properly, with the right tool connections and a human in the loop at the right moments.

This is exactly the gap ODIV's multi-agent-systems service exists to close. Rather than you experimenting with a general-purpose tool and hoping it fits your workflow, our team builds you a custom set of agents designed around your actual business, one agent that researches, one that drafts, one that monitors, coordinated properly and wired into the systems you already run your business on, your CRM, your inbox, your spreadsheets, and where relevant, your WhatsApp line through ODIV Engage so your team gets notified the moment an agent finishes a task or needs a human decision.

Our engineers work hands-on in modern AI build environments like Lovable and Claude Code, alongside conventional engineering practice, which is exactly why this comes together faster and for a fraction of the time and cost of a traditional, hand-coded custom development project. The AI tools get you to a working build quickly, our engineers make sure it's correct, secure, properly integrated with your existing tools, and maintainable after launch, not a fragile prototype that breaks the week after it ships.

If you've read this and have a specific ongoing task in mind, competitor tracking, lead research, report generation, support triage, start a chat with us on WhatsApp and tell us what you're trying to automate. We'll tell you plainly whether a multi-agent build makes sense for it, and what we'd actually build.