Beyond Chatbots: Exploring AI Agents for Research

Imagine pulling key information from a stack of journal articles, compiling the findings into a comparison table, and then having an AI agent repeat the process whenever you need it. Or have an agent monitor new research and alert you to relevant articles as they are published.

These are some of the possibilities explored at AI Agents for Literature Management and Synthesis: ChatGPT and Claude, an hour-long online workshop conducted on 20 August 2026. Organised by NIE Library and the NIE Graduate Student Club (GSC), the workshop was the second event in the Foundations for Academic Success. This multi-part collaborative series runs every semester.

GSC President and final-year PhD student Ni Youhua led the session, drawing on practical examples of how AI agents could help with familiar research challenges.

From Individual Papers to a Bigger Picture

Youhua demonstrated how AI agents such as ChatGPT and Claude can extract information from uploaded PDF journal articles and bring it together into an integrated comparison table.

One particularly useful feature was the ability to link AI-generated claims back to the specific sections of the original papers. Instead of simply accepting an AI-generated summary, researchers can follow the links to check the source themselves. This gives researchers a practical way to keep verification in their hands.

From One-off Prompts to Repeatable Tasks

The workshop then took things a step further. What if a useful sequence of instructions did not have to be recreated from scratch every time?

Youhua showed how a series of steps could become a repeatable task by prompting an AI agent to create a skills markdown file. The file can then be reused by an agent to carry out the same type of task in the future.

He also demonstrated how agents could handle scheduled tasks, such as monitoring RSS feeds for newly published articles relevant to a research topic. Instead of remembering to check for new literature manually, researchers could set up an agent to perform this task on a schedule.

More Automation, More Responsibility

The possibilities are appealing, but handing AI agents a more active role in research also comes with risks.

The workshop highlighted concerns around exposing unpublished manuscripts or confidential information, particularly when agents are given access to research materials or files. An agent may also take actions beyond what the user intended, such as modifying or deleting files on a device.

The session therefore ended with practical ways to limit these risks when working with AI agents.

The key message was not simply that AI agents can do more than answer questions. They can potentially help researchers organise information, repeat routine tasks and automate parts of the research workflow. But as these tools become more capable, careful verification and appropriate safeguards remain essential.