Perplexity Deep Research
Perplexity Deep Research builds a research search workflow using parallel retrieval and model-generated code. Task: Suggested implementation: A lead agent plans the research and saves the plan to memory. It spawns parallel sub-agents that search different angles, while the model writes code to assemble its own searches. A separate citation agent matches every claim to its source at the end. Human role: Define the research scope and review the final report for accuracy.; tools: Tool described in the source. Human role: The source summary does not fully detail human review and operational roles; actual deployment should include human oversight based on task requirements.; limitations: This is a token-hungry setup, so use it only for questions worth the cost. The source is a secondary summary of public posts, so verify against original vendor documentation. Human approval is required before publishing or acting on the research findings..
Tools
Tool described in the source
Published on Agent Work Atlas:
Overview
Perplexity Deep Research builds a research search workflow using parallel retrieval and model-generated code.
Workflow
Suggested implementation: A lead agent plans the research and saves the plan to memory. It spawns parallel sub-agents that search different angles, while the model writes code to assemble its own searches. A separate citation agent matches every claim to its source at the end. Human role: Define the research scope and review the final report for accuracy.
Reported result
The source reports that Perplexity's Deep Research improved its BrowseComp score from 40.7% to 83.8%. This is a benchmark metric, not a verified business outcome. Actual deployment results, cost, and time savings are unknown and require independent testing.
Human role
The source summary does not fully detail human review and operational roles; actual deployment should include human oversight based on task requirements.
Limitations
This is a token-hungry setup, so use it only for questions worth the cost. The source is a secondary summary of public posts, so verify against original vendor documentation. Human approval is required before publishing or acting on the research findings.
Try it
I want to test a parallel research workflow for a specific business question. My topic is [replaceable topic], and my key questions are [replaceable questions]. First, perform a dry run on a small subset of sources to verify the citation accuracy and logical flow. If the dry run is successful, proceed with the full research. Use a lead agent to plan the search, spawn parallel sub-agents for different angles, and have the model write code to assemble the searches. Finally, use a citation agent to match every claim to its source. Output a structured report with clear citations. Measure success by the percentage of claims with verified sources and the time taken compared to manual research. Do not publish or use the results without my explicit confirmation.
Source reports are not independently verified. Publication date is the catalog publication date, not the original article date.