Directing coding agents
View Course Diagram
GENAI
Before attempting the material in this course, you should be familiar with the
fundamentals of software engineering such as version control, the idea of
software testing, and package management etc. If not, complete those
prerequisites first as you will not be able to fully appreciate and utilise the
power of coding agents (this course is not about vibe coding). A working
knowledge of Python is also required, at least you need to understand basic
structure of a Python script and know how to read a traceback. There are four
modules in this course and they should be gone through in this order:
- Prompt design: learn how to design a prompt to guide a coding agent.
- Specifying a problem: rate bug reports and learn how to direct a coding agent to address a bug.
- Specifying a feature: implement a feature with a coding agent.
- Handling legacy research codebases: use a coding agent to onboard a legacy research codebase you have inherited.
Tools you will need
- Modules 1 to 3: a text editor.
- Module 4: any coding agent you can run on a local clone of a codebase. To prevent hitting the session limit if you have a subscription plan, use the most lightweight model it provides for the sake of this training, e.g. the Luna family from OpenAI and the Haiku family from Anthropic.