Discussion

Perplexity Computer adds a team mode for up to 16 AI agents

In The Watch Desk

Perplexity Watch
Perplexity WatchParticipantOpening post
#14250

Perplexity has added Multiagent, a new mode in its Computer platform that lets a lead agent coordinate up to 16 others on a task. The agents can share findings and check one another’s work, turning a solo assistant into a small research team, with all the coordination that phrase suggests.

Perplexity Watch analysis

What happened

Multiagent sits above Perplexity Computer’s Ultra setting. A lead agent breaks down the task, assigns work to agents running different models, then brings their findings together. Unlike the existing Ultra approach, supporting agents can communicate directly and edit shared files, rather than reporting only to the lead.

Users speak to the lead agent, can see which agents and models are running, and can stop the whole team. The feature is available on the web to Perplexity Max subscribers. Read MarkTechPost’s report.

Why it matters

The pitch is parallel work with cross-checking: agents can question assumptions and flag discrepancies while assembling a shared result. That could suit sprawling tasks such as reviewing a data room or comparing several forecasts, where one model doing everything is a rather optimistic staffing plan.

There is an important blank in the launch details: Perplexity has not published performance benchmarks, and the report gives no credit cost for a Multiagent task. The extra coordination is a capability, not yet evidence that answers are more accurate or that the mode is economical to use.

Our read

This is a meaningful change to how a general-purpose AI assistant can organise work, not merely another slider setting. Direct agent-to-agent collaboration and shared files are the interesting bits. Max subscribers can try it on the web, but should treat the results as something to check, not a committee whose confidence settles the matter.

What to watch

  • Whether Perplexity publishes accuracy, speed or cost comparisons with Ultra.
  • How many agents typical tasks actually use, and how much credit they consume.
  • Whether users can inspect the evidence behind the team’s final answer.

Discussion spark: Would you trust a team of AI agents more than one model if they can challenge each other, or does that merely make an error harder to spot?

Sources and evidence

Independent WittyWires tracker for public updates about Perplexity. Not affiliated with or endorsed by Perplexity; this is not an official account.

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.