Use an agent to trace papers from a seed set
An agent can start with a few papers, follow citations both ways, and keep narrowing a review until the results stop changing.
- What they did
- A poster created a small literature-review agent because general chatbots were overlooking relevant studies and citation chains. The workflow begins with seed papers, then moves backward through their reference lists and forward through later papers that cite them. It checks each candidate against set filters, revises the review along the way, and is designed to run in Claude Code, Codex, Gemini CLI, or OpenClaw.
- How it went
- Their tests showed a stronger final review than a plain AI-generated draft, especially for finding the right papers and covering the field more completely. The setup is not fully autonomous, since part of it is controlled by scripted steps rather than open-ended agent behavior.
- Tips
- Use the agent for structured searching and screening, not as a replacement for your own judgment. Have it stop for approval before any irreversible action, including submitting or deleting anything.
Try it yourself with Claude Code
Start a literature review on [topic] from these seed papers: [titles, DOIs, or links]. Look first at each paper’s references, then at later papers that cite it, and keep only items that fit these rules: [inclusion criteria]. After each pass, update the search based on what is still missing, and keep going until another round adds nothing useful. Before you finalize the review, show me the included papers, a list of exclusions with reasons, and a draft summary; wait for my approval before saving, submitting, or otherwise making any irreversible change. Then deliver a table of included studies plus a short synthesis of the main themes.
Fill in the [brackets]. Agents can make mistakes, so review anything before it is sent, paid or deleted.
Get 3 agent uses like this every morning
With the day's top AI news. Free, and one click to unsubscribe.
Related agent uses
Check how long open roles have been live at target companies
An agent sampled jobs on five company boards and measured how long each posting had been open.
Use a research agent for systematic literature reviews
A Reddit thread points to tools that can help with systematic reviews, paper notes, and keeping a project organized over time.
Build a reusable stock-analysis workspace
A custom agent kept one investment project alive, saving models and notes so later sessions could pick up where the last run stopped.
Use a browser agent to find and save papers
A commenter uses a custom browser agent for a narrow set of research and data-entry jobs, including finding scientific papers and downloading bank CSVs.
Build a stock research agent for updated company reports
They made a stock-research tool that refreshes reports on tracked companies using live market data instead of stale chatbot answers.
Build a dashboard from a semantic data layer
Use a coding agent with Graphene to define metrics, query your warehouse, and publish a report with charts and screenshots.