Research agents answer questions that need many searches and sources: market analysis, literature reviews, technical comparisons.
Typical Architecture
- A lead agent plans the research and breaks the question into sub-questions.
- Search and fetch tools gather sources, sometimes via parallel sub-agents.
- The agent synthesises findings into a report with citations.
Keys to Quality
- Source quality: prefer primary sources, official documentation and reputable publications.
- Citations: every claim should link to where it came from.
- Breadth then depth: start with broad searches, then dig into promising leads.
- Knowing when to stop: set budgets; more searching has diminishing returns.
Common Failures
- Relying on low-quality or outdated sources.
- Citations that don't support the claim.
- Missing conflicting evidence.
- Overconfident conclusions.
Verification
Check key claims against their cited sources. Consider a separate verification step or agent that checks citations.
Security
Web content can contain prompt injection. Research agents shouldn't have access to sensitive tools while reading untrusted pages.