Day 1: Overview
Goals for Day 1
- Familiarize yourself with the basics of AI-assisted coding and its workflow.
- Get an intuitive understanding of the key concepts behind the technology, how it works and what its capabilities and limitations are.
- Set up the software environment to be able to practice AI-assisted coding on your own during the self-study period following examples and supplementary materials.
- See AI in action by following along with live demonstrations: understanding data with AI and reproducing, revising, and reusing existing analysis.
Agenda
Day 1: Introduction and common workflows
| Time | Topic |
|---|---|
| 9:00-9:10 | 1. Welcome and course overview (slides) |
| 9:10-9:40 | 2. Setup overview (slides) |
| 9:40-10:20 | 3. Live: Self-Study > Example 0 |
| 10:20-10:30 | ☕ Break (10 min) |
| 10:30-10:50 | 4. Core AI-related concepts |
| 10:50-11:00 | 5. Common workflows and data safeguarding (slides) |
| 11:00-11:30 | 6. Example 2: Understanding data with AI (live) |
Self-study:
Before You Arrive
(Try) completing the setup. For common issues and troubleshooting, refer to the relevant sections in the setup materials:
- Software Setup > Troubleshooting,
- Stata for Positron Extensions > Troubleshooting,
- GitHub & Copilot > Troubleshooting, and
- Positron Assistant > Troubleshooting sections for common issues with the setup.
If you run into a dead end, contact ITHelp@worldbankgroup.org.
Resources
Course data is available in the OneDrive folder: ai4coding-data.
Course examples and materials are available in this repository: ai4coding or on OneDrive: ai4coding-materials.
If, once you open the link to the OneDrive folder, you see a message that you don’t have access, please request access by clicking the “Request Access” button on the page. I will be monitoring and approving access requests as they come in.
Day 1: Materials
1. Welcome and course overview
Motivation for AI-assisted coding in an IDE versus copy-pasting from chat-based tools. Overview of the course structure: two sessions, self-study in between, and the key tools — Positron and GitHub Copilot. Course goals: understand, practice, and take control of AI-assisted coding.
Slides
View slides in a new tab.
2. Setup overview
Important: Set up your working folders properly!
Do not use OneDrive for coding with Positron during this course! (temporary).
- Positron is a new software for the WB.
- Unfortunately we only discovered last week that WB security policies block Positron from writing text files on OneDrive.
- Do not save your code-in-progress in a OneDrive folder.
- New security policy will be implemented Friday, April 24 (Because of the spring meeting freeze).
To follow along with the course and practice on your own, you need to:
- On
C:\WBG\create a folder calledai. - Copy exercise templates into
C:\WBG\ai\ai4coding-practice\using either:- OneDrive (WB only) folder
ai4coding-examples - or forking the GitHub repo worldbank/ai4coding-examples
- OneDrive (WB only) folder
- Work in the relevant exercise subfolder, e.g.
C:\WBG\ai\ai4coding-practice\ex01\. - Access course data:
- Shared through OneDrive (WB only)
ai4coding-data - It is read-only. You can sync it to your OneDrive (where appropriate) and use it from there.
- If you encounter problems, copy the data to
C:\WBG\ai\ai4coding-data\.
- Shared through OneDrive (WB only)
Detailed setup requirements: software (Positron, Python, R, Stata, Quarto), GitHub account and WB organization membership, GitHub Copilot access (WB plan or private), and Positron Assistant configuration. Working folder structure for exercises and data.
Slides
View slides in a new tab.
3. Live: Self-Study > Example 0
Live walkthrough of Self-Study > Example 0 — open the ex00-positron-test folder in Positron, follow the README, and verify that your setup (Positron, Copilot, Stata extension) is working correctly.
4. Core AI-related concepts
Key concepts behind how AI coding assistants work: IDEs, LLMs, context windows, tokens, and embeddings. How context size and quality affect response accuracy. Context saturation and best practices for managing it. Hallucinations — what causes them and how to mitigate them. Data exposure risks when using AI in an IDE. Preview of Day 2 topics: agents, tools, prompt files, and skills.
Slides
View slides in a new tab.
5. Common workflows and data safeguarding
A structured eight-step workflow for AI-assisted data analysis: from setting objectives and defining inputs/outputs, to iterative code development and independent verification. Why structure matters — vague prompts produce generic output; each step narrows the AI’s degrees of freedom.
Data safeguarding practices: key exposure channels (prompts, console output, file reads, shell commands) and five remedies — careful prompting, system instructions, metadata-only approach, data anonymization, and keeping raw data outside the project folder. WB AI Use Policy and responsible use guidelines.
Slides
View slides in a new tab.
6. Live: Self-Study > Example 2
Live walkthrough of Self-Study > Example 2 — using AI to understand an unfamiliar dataset: building a data dictionary, documenting harmonization, generating modular cleaning code, and verifying results. Participants follow along in their own environment.