Overview¶
Overview¶
This course is about doing faster and more rigorous data analysis with agentic AI. It is practical throughout: we take the problems researchers actually have — loading messy data, settling a hypothesis, plotting it, testing it and writing it up — and work through them with an agent, one step at a time.
This course follows on from Agentic AI for Research. That course covers what agentic AI is, how it differs from a chatbot, and how to set up Claude Code alongside Obsidian. It also carries a first statistical worked example. Start there! And come back here when you are done.
The rest of the course is broken into two parts.
In part 1 a worked example runs the whole way through, from a raw download to a finished report. Along the way we explain enough about how agentic AI works to see where its limits are, because that is what tells you which parts of an analysis you can hand over and which you cannot. The emphasis is on avoiding getting pushed out of the loop by the AI: you should control how the work is carried out, not the other way around. We provide an agentic workflow for going from data to a statistical report.
In part 2 you adapt that workflow to your own field and your own way of working, so that the next analysis starts where this one finished rather than from scratch. We then look at the rules for using AI in published work, and at how far the automation can reasonably be pushed.
What you will be able to do by the end of the course¶
- Work through a complete analysis one step at a time, keeping the decisions yours rather than the agent's — 1.1 Step by step working with AI, and the worked example in 1.2 Importing data, 1.4 Visualisation and 1.6 Statistical analysis.
- Settle a hypothesis before you look at the data, and know what your variables actually measure before you test anything with them — 1.3 Deciding on a hypothesis.
- Judge what the agent proposes rather than accepting it, understanding a test well enough to check it and seeing where this way of working departs from how statistics is usually taught — 1.5 Traditional statistics versus Agentic statistics and 1.8 Best practices for working with agentic AI.
- Turn an analysis into something another person can read and another agent can check, written as the work proceeds rather than at the end — 1.7 Turning your analysis into a report.
- Build a framework for your own field, so the next analysis starts where this one finished — 2.2 Adapting your own framework, with examples in 2.4 A football data framework and 2.5 A statistical machine learning framework.
- Work ethically, and know the limits of what can be automated — 2.6 Ethical aspects of Agentic AI and 2.7 Fully automated research?.