/pattern/exploration-discovery/

03 · Multi-AgentDeep ResearchResearch AgentDeep Research Agent

Exploration & Discovery.
Map the space. Cluster, select, deep-dive, synthesize.

A research agent maps a knowledge space, clusters findings, selects leads by novelty, impact, and feasibility, deep-dives into the promising ones, and synthesizes a cited report — coordinated as an orchestrator over parallel search and subtopic agents.

When to reach for it

  • The task is open-ended research, competitive analysis, or literature / R&D synthesis across many sources.
  • Breadth and novelty matter more than a single lookup.
  • Findings must be synthesized and cited, not just retrieved.

When it backfires

  • The answer is a bounded lookup or a single retrieval.
  • Cost and latency budgets are tight — this pattern is resource-intensive.
  • The knowledge space is small enough for one pass.

The tradeoff

Breadth, novelty, and synthesis quality are gained at a high token and latency cost; it needs strong stopping criteria to avoid unbounded exploration.

The mental model

A shape you can draw on a napkin.

A research orchestrator maps the space into disjoint subtopics, dispatches parallel search agents, then folds their findings into a cited synthesis.

Research OrchestratorSearch AgentSearch AgentSynthesizer
Pitfalls

Three ways this pattern will hurt you.

Unbounded exploration

Without explicit stopping criteria the agent keeps spawning searches and subtopics, burning tokens and time with diminishing returns.

Fix · Set a research budget (depth, breadth, token / time caps) and an explicit selection rubric (novelty, impact, feasibility) that prunes leads before deep investigation.

Synthesis without provenance

Merging parallel findings into one report without tracking which source produced which claim yields confident but unverifiable output.

Fix · Carry provenance through the merge — every synthesized claim links back to its source and the agent that retrieved it; surface citations in the final report.

Redundant parallel searches

Independently spawned search agents chase the same leads, wasting budget and over-weighting duplicated findings in the final synthesis.

Fix · Deduplicate the lead set and assign disjoint subtopics before fan-out; have the orchestrator track coverage so agents explore complementary, not overlapping, ground.

Framework support

Where Exploration & Discovery is native.

LangGraphas a custom orchestrator/worker research graphAdaptable
OpenAI Deep Researchproductized research agentAdaptable
Perplexity / Gemini Deep Researchproductized research agentsAdaptable

Search

Search patterns, frameworks, and pages.