Introduction to AI Agents for Researchers (April 2026)

English summary

The presentation introduces the use of large language model based AI agents in academic research. It contrasts agents with traditional chatbots and describes agent configuration, research workflows, and the division of responsibilities between humans and AI. An AI agent can use a high level goal to plan multiple steps, execute code, work with files and external tools, and repeat tasks in response to execution results.

Key points

• A traditional chatbot proceeds through prompt and response exchanges, leaving the user to move data, run code, interpret documents, and direct each step. An AI agent can plan multistep work, call tools, manipulate files, use external APIs, correct errors, and coordinate subagents.
• The basic agent loop is presented as goal setting, reasoning and planning, tool execution, observation of results, and either completion or another reasoning cycle. Short term memory consists of conversation context, whereas long term memory includes project files, state logs, and intermediate summary documents.
• AGENTS.md is described as a project configuration file containing background, directories, data locations, analysis procedures, naming rules, formatting requirements, and operational constraints. Skills are reusable standard operating procedures for repeated tasks, and the Model Context Protocol is presented as an interface linking agents with external tools, APIs, and data resources.

Audience

Academic researchers, graduate students, laboratory investigators, and especially early career researchers integrating AI into research

Scope and limitations

This summary is based only on the supplied evidence notes from the presentation. The original sources and full methods underlying the cited examples of automated research, autonomous laboratory work, data integration, document extraction, and parallel literature screening were not independently verified here. The reported performance, cost reduction, peer review outcome, and other case figures should therefore be understood only as statements described in the presentation notes. The notes also identify risks including missing data, field name mismatches, outdated APIs, inappropriate statistical assumptions, data leakage, and unverifiable conclusions.

Research context

For medical AI and data science, the topic is thematically connected to applying AI agents to nonstandardized file integration, research data inventory, document based data extraction, analysis automation, and result auditing. The presentation places research direction, ethical judgment, and scientific validation with humans, while assigning agents responsibility for project context and execution and distinguishing computational work from physical or subjective tasks.