How to Use AI for Coding

Value of Coding in the AI Era

• AI is not most valuable as a random Q&A tool or an essay/email draft tool.
• AI is most powerful in coding, because AI itself is built with code.
• This changes how coding skills such as Python, R, and Stata are evaluated on your CV.

The key is:

• Your capability to control AI so it can code for your task.
• The idea must be yours (a calculator does not know what should be computed).
• You must give AI logical instructions (because AI builds on logical coding).

Learning Objective (Today)

• write clear and logical prompts for coding tasks,
• check whether AI output is trustworthy,
• and summarize results in plain language.

Start By Creating A Dedicated Folder

• Create one dedicated folder for this project.
• In VS Code, use Open Folder and move into that folder first.
• This should be the starting point for all files, code, figures, and reports.
• It helps keep your work organized and reproducible.

How To Start The Chat

Example starter prompt:

• I am a student of ECON10 and a beginner in Python. Help me collect, clean, visualize, and report data in Quarto. Explain each step in plain language so I can stay accountable.

Context Information To Include

Context to include:

• Role: ECON10 student
• Coding skill level: Beginner in Python
• Objective: Visualization of macroeconomic data
• Output: Python + Quarto report in PDF
• Constraints: Source citation + cleaning log

Brainstorm Before Asking for Data Sources

• If the data source is unclear, AI can help brainstorm.
• Give your context first.

Example brainstorming prompt:

• I want to analyze labor market conditions in the United States.
• We discussed unemployment rate and different types of unemployment.
• Suggest several feasible directions.

This Helps You

• clarify your research question and research context
• narrow candidate datasets
• get better AI recommendations

Step-by-step: Data Source → Download → Save

Task 1: Choose the source

• Suggest 3-5 credible data sources for this task.
• Recommend one and show the exact URL and file name.

Task 2: Plan the folder structure

• Suggest a simple folder structure for this project.
• Save raw data under data/ and figures under output/.

Task 3: Keep a report-ready log

• In the report, include the source URL, retrieval date, local file path, and filter rules.

Decide Which Visualization To Build

• Decide the visualization before asking for code.

Example prompt:

• Suggest several effective visualization options for the dataset.
• I would like to try different ways of visualization, such as line chart, bar chart, scatter plot. I am also interested in visualizing in maps.

This Helps You Decide

• which chart or map to build first
• what comparison to emphasize
• which outputs you need for PDF versus interactive use

Ask AI To Write Python Visualization Code

• After planning and saving the data, ask AI for runnable code.

Example prompt:

• Using the saved file in data/ and the selected plan, write Python code to clean the data and create the required charts/maps.
• Save figures in output/ and add brief comments for a beginner.

If You Do Not Understand The Code, Ask AI

• Ask follow-up questions.
• Ask until you can explain each key step.
• Even when using AI for coding, you are fully responsible for what the code does and what results you report.

First, Try Running Code In Interactive Window

• First, try running your code in the Interactive Window.
• Run code block by block.
• Inspect outputs and error tracebacks.
• Revise the prompt and rerun.
• Open a .py file and press Shift+Enter on a line or selection.
• Right-click in the editor and choose Run Current File in Interactive Window.
• Command Palette (Ctrl+Shift+P) → Python: Run Current File in Interactive Window.

Interactive Window Screenshot

If You See An Error, Paste It As It Is

• If an error appears, paste the traceback as it is.
• AI can often explain the cause and suggest the next fix.
• Do not paraphrase the error too much.

Example prompt:

• Here is my traceback. Explain the cause simply.
• Give the smallest fix first.
• Then give one check to confirm it worked.

Before Plotting, Ask For A Table

• Before plotting, it is fine to ask AI to show the data in a table first.
• A table helps you check whether the variables and values make sense.
• Then move to visualization.

Example prompt:

• Show the data in a simple table first.
• Display the relevant columns and a small number of rows.
• Then suggest which figure we should create next.

Start With Basic Plots First

• Start with the national trend.
• Start with a simple line plot or bar plot.
• Complex visuals can hide the data pattern.
• If the first chart is unclear, do not trust later charts.

Example: Does This Plot Look Right?

Notice When Something Looks Off

• A strange spike appears in one period.
• The trend looks unrealistic.
• Stop and check data definitions, missing values, and filters before interpretation.

AI Suggested A Fix. Check It Anyway.

• I noticed that the plot looked wrong.
• I could re-check the data myself, or ask AI to help inspect it again.
• AI suggested that “only series ending in 003 should be plotted,” and then rewrote the figure. <!– - That is still not enough.
• A corrected figure is not proof.
• You should ask why AI judged this version to be correct. –>

AI Rewrote The National Trend

• This version looks more plausible.
• But you should not trust it only because AI redrew it.
• Ask why AI determined this was the correct series.

When AI Says “003 Is Correct,” Ask For Evidence

• Why do you think the series ending in 003 is the correct measure?
• Show the official documentation and explain how you mapped code to meaning.
• Show why other series codes are not the target variable for this analysis.
• Do not accept “AI says so.” Ask for a credible source and reasoning chain.

After The National Trend, Look At Regional Patterns

• Once the national trend looks reasonable, move to more detailed views.
• For example, compare states or regions.
• An interactive map is one useful example.

Example: Interactive Map

• Use an interactive map to inspect patterns by state and year.
• This helps you notice regional differences that a national trend cannot show.
• Use it for exploration before deciding what to include in the final report.

From Interactive To Static

• An interactive map is useful for exploration.
• But you cannot place that interactive version in a PDF report.
• So AI can also create a static map for Quarto.

Example: Static Map For The Report

Ask AI To Organize The Quarto Report

• Once you have the visualizations you want to include, ask AI to organize them in Quarto.
• Create a simple Quarto report with sections for data, cleaning, figures, and findings.
• Include source citation, retrieval date, and file path.
• Show me where I should write my own observations and interpretation.

Let AI Build The Skeleton, Not Your Thinking

• Ask AI to suggest headings and structure.
• Ask AI where your observation, interpretation, and limitation should go.
• Write your own main finding, supporting evidence, and caution.
• Write the actual interpretation yourself.

Your brain is still required here.

Quarto Can Export To PDF Or Word

• PDF output is possible.
• Word output is also possible.
• Quarto rendering requires a Terminal command.

Prompt example:

• Revise this Quarto file so it can render to PDF.
• Also add Word output (docx).
• Show me the exact render command I should run in the Terminal.

Render command examples:

• quarto render your_file.qmd --to pdf
• quarto render your_file.qmd --to docx
• quarto render your_file.qmd

Ask AI To Save A Markdown Work Log

• Before stopping, ask AI to summarize the work in Markdown.
• This makes it easier to restart later.
• Include data used, files created, figures made, problems found, and next steps.

Prompt example:

• Summarize today’s work in Markdown.
• Use sections for data, code, figures, decisions, errors, and next steps.
• Write it so we can restart this project later.

Important Cautions

• Permission safety
• Academic integrity
• Responsibility

Permission Safety

• Allow grants permission, such as file access or command execution.
• Keep keeps or downloads a file your browser or OS flagged.
• Approve only what you understand and need.
• If you are unsure, stop and ask first.

Academic Integrity: Think Of A Calculator

• Using a calculator is not considered cheating.
• You decide what should be calculated.
• You give the instruction.

Academic Integrity: The Same Logic Applies To AI

• The same logic applies to AI use.
• Even a very polished analysis should get 0 if it does not answer the assignment for this course.
• AI output does not replace your academic judgment.

Responsibility: You Must Be Able To Explain It

• You must be able to explain every step in your own words.
• That includes data collection, series choice, and what each figure shows.
• It is fine to let AI handle basic mechanical tasks.
• For example: drawing a line chart or coloring a map from low unemployment in green to high unemployment in red.
• Do not include anything complicated that you cannot explain yourself.

Final Takeaway

• Start simple, and check the data step by step.
• Use AI for structure and mechanical coding, not for unexamined conclusions.
• Verify evidence, and explain every result in your own words.