6-7 July 2026

Agentic AI is rapidly reshaping how researchers work. This course teaches life scientists how to use agentic AI tools properly: systems that go beyond chatbots to read files, write code, run commands, and execute multi-step research tasks. Despite the growing importance of these tools, there is little structured guidance available for researchers. Most are left to navigate this shift independently.
The course covers the full stack of agentic AI skills, from understanding the tools and structuring effective prompts, through to building reusable research workflows. All exercises use real life sciences data and tasks.

Biologists, bioinformaticians, data scientists, and other life sciences researchers
Basic command line familiarity and one programming language (R or Python). No prior AI experience required.

By the end of the course, participants will be able to:

- Understand the agent-provider-model stack and choose the right tool for a given task
- Apply context engineering to get better, more reliable outputs from AI tools
- Use coding agents for data analysis, figure generation, and automated presentations
- Build interactive data portals from omics data and publish them to GitHub
- Create reusable skills and agent configurations for repeated research tasks
- Connect agents to external knowledge sources through RAG and Model Context Protocol
- Develop critical judgement about when AI helps, when it does not, and how to use it responsibly

Day 1: Foundations

Monday 6 July, 2:00–5:00 PM CET

Part 1: Introduction to Agentic AI (~60 min)

Theory and setup. Understanding what these tools are, how they differ from chatbots, and getting everyone working.

• What LLMs are and how they work in practice

• The agent-provider-model stack: separating the agent, the API provider, and the model

• Chat assistants vs coding agents vs multi-step agentic systems

• Capabilities, limitations, and failure modes

• Data privacy: what stays local and what gets sent externally

• Tool setup: VS Code + GitHub Copilot, Claude Code, or OpenCode with OpenRouter

• First interaction and tool comparison exercise

Part 2: Context Engineering (~60 min)

How to structure prompts, provide context, and manage context windows so the model gives better, cheaper answers.

• Why vague prompts produce vague results

• Structuring prompts with role, task, format, and constraints

• Plan mode: guiding the model before it starts building

• Understanding the context window: what fills it, what it costs, and when it causes hallucinations

• Model pricing comparison and strategies for managing long sessions

Part 3: Data Analysis, Presentations, and Skills (~60 min)

Hands-on practicals applying context engineering to real tasks, and introducing skills as reusable prompt templates that dramatically improve output quality.

• Data analysis with the Palmer Penguins dataset: Plan mode, figures, and export

• Automated journal club presentations: paper to PowerPoint in one prompt

• Introduction to skills: reusable instruction sets that teach the agent how to do a specific task well

• Comparing output with and without a skill loaded: demonstrating the quality difference

• Iterative refinement: being sceptical, reviewing generated code, and redirecting

Day 2: Building Workflows

Tuesday 7 July, 2:00–5:00 PM CET

Part 4: Interactive Data Portals (~60 min)

Using coding agents to scaffold web applications from research data and publish them to GitHub.

• Build an interactive single-cell RNA-seq viewer from CZ CELLxGENE data

• Extend the app with gene search, cell-type filtering, and custom visualisations

• Version control and publishing: push to GitHub using the CLI

Part 5: RAG, Skills, and Agents (~60 min)

How agents access external knowledge, and how to make their behaviour reusable and reproducible.

• Retrieval Augmented Generation and Model Context Protocol (MCP)

• Connecting agents to PubMed, GitHub, and institutional resources

• Rule files (AGENTS.md, CLAUDE.md) for reproducible agent behaviour across sessions

• Building reusable skills: structured prompt templates with defined inputs, outputs, and validation

• Practical: write project rules and a skill for a repeated research task

Part 6: Research Workflows and Second Brain (~60 min)

Combining skills, agents, and knowledge systems into end-to-end research workflows.

• Building custom agents for specific tasks: configure instructions, model, and tool access

• Connecting agents to a personal knowledge base: Obsidian, structured notes, and literature

• End-to-end workflow: from research question to output, pulling together rules, skills, and retrieval

• Open session: participants apply the tools to their own research tasks with support

• Responsible use, reproducibility, and building an AI use policy for your lab

1- AI-Assisted Scientific Writing for Researchers - ONLINE, 3-5 June

2 - AI-Powered Python for Bioinformatics - ONLINE, 1-2 July