Design 7: Research agent
Words made simple
A few useful meanings before you read the explanation.
- Agent
- An application that uses a model to choose its next action within defined limits. It can use tools and continue through several steps.
Overview
A bounded system gathers external evidence and produces a cited research draft. Its output is decision support, not verified universal truth.
Why this matters
Questions may require several public sources and comparisons.
An everyday example
A researcher collects references and explains their differences.
A technical example
Search approved public sites, fetch bounded pages, and compare dated product information.
How it works
- Define scope
- Search
- Validate destinations
- Collect evidence
- Synthesize
- Verify citations/conflicts
- Stop.
Architecture and flow
Question → bounded search/fetch → evidence store → cited synthesis → review.
The insurance assistant example
Staff compare public claim-submission processes across product documentation.
When to use it
Evidence gathering with clear time/source budgets.
When to avoid it
Do not expose private records to arbitrary sites or use unsourced research for claim adjudication.
Interview answer
“Research agents need bounded search, safe fetching, dated provenance, citation checks, and explicit uncertainty.”
Benefits, trade-offs, and production considerations
Prevent SSRF, unrestricted downloads, indirect injection, endless browsing, and stale citations. Save permitted source snapshots/versions for reproducibility.
Can you explain it without the notes?
Design 7: Research agent
Use your own words. Explain an example and one limitation before checking the explanation.
Check your explanation
A bounded system gathers external evidence and produces a cited research draft. Its output is decision support, not verified universal truth.
Turn this concept into a clear answer.
Practice this conceptAdapted from 13_AI.MD · “Design 7: Research agent”. Examples are study fragments; framework behavior can depend on your version.