Agentic Coding for Survey Research - Online

A three-day online course, delivered over three weeks (sessions in the afternoons), on using AI agents to improve survey research workflows while preserving privacy, reproducibility, review discipline, and institutional data-governance boundaries.
The course adapts agentic coding methods to the survey-research lifecycle: designing survey instruments, analysing survey data, documenting metadata and derived variables, checking outputs, and preparing reproducible or release-ready packages. Participants work hands-on with Cursor, GitHub, and role-specific repository workflows, producing auditable artifacts they can reuse in their own organisations.

The course covers:

• Agentic coding foundations and Cursor workflows for survey research
• GitHub issues, branches, pull requests, review logs, and acceptance criteria
• Project guardrails: rules, AGENTS.md, .cursorignore, .gitignore, and documented data-use boundaries
• Restricted-data decision rules for cloud AI, local models, metadata-only work, synthetic data, and TRE/sandboxed environments
• Spec-driven development for survey workflows
• Applied researcher track: literature/context, safe data exploration, cleaning, descriptives, regression or standard analysis, tables/figures, reproducible write-up support
• Survey practitioner track: measure search, questionnaire specification, respondent materials, piloting, ethics documentation, testing notes, fieldwork monitoring
• Survey data manager track: metadata, derived-variable specifications, longitudinal consistency, cleaning checks, disclosure checks, deposit/user documentation
• Reusable skills and subagents for measure mapping, questionnaire review, metadata checks, derived-variable checks, disclosure risk review, and user-guide review
• Role-specific capstone packaging: reproducible analysis, questionnaire/specification workflow, or data-release documentation
• 30-day adoption planning for post-course implementation

By the end of the course participants will:

• Set up a practical agentic workflow using Cursor, GitHub, and a role-specific repository structure
• Use issues, branches, pull requests, review logs, and acceptance criteria to make survey workflow tasks auditable
• Apply restricted-data decision rules for cloud AI, local models, metadata-only work, synthetic data, and TRE/sandboxed environments
• Configure project guardrails with rules, AGENTS.md, .cursorignore, .gitignore, and documented data-use boundaries
• Turn a survey workflow into a spec-driven plan with a safe input map, role-specific deliverables, and verification checks
• Build or adapt reusable skills and subagents for survey-specific work such as measure mapping, questionnaire review, metadata checks, derived-variable checks, disclosure risk review, and user-guide review
• Package a reviewed output for their role track: reproducible analysis, questionnaire/specification workflow, or data-release documentation

Target Audience

The course is designed for a mixed audience across the survey research lifecycle:
• Applied researchers using existing survey data for empirical social science (economics, sociology, political science, and related disciplines)
• Survey practitioners designing questionnaires, respondent materials, pilot tests, and fieldwork processes
• Survey data managers preparing, documenting, checking, and releasing survey datasets
• Government researchers, academic researchers, and staff at survey agencies and research institutions
• Mixed seniority: junior and senior researchers, analysts, and operational staff

Technical levels range from participants comfortable in R/Stata/Python to participants who mainly work through specifications, documents, and operational workflows. Recommended cohort cap: 20 participants unless additional facilitators support breakouts and troubleshooting.

The course will run over 3 live online sessions on 5, 12 and 19 November (13:00-17:00) and there will be asynchronous assignments between sessions. 

Course Description

Pre-requisites

No specialist prior knowledge of agentic coding or AI tools is required. Participants should:
• Be comfortable using a computer, web browser, and video conferencing
• Have interest in survey research workflows (design, analysis, or data management)
• Be willing to create GitHub and Cursor accounts before Day 1
• Complete the pre-course setup checklist or attend the setup clinic
• Choose one role track before or during prework: applied researcher, survey practitioner, or survey data manager
• Bring a non-sensitive workflow description, public dataset, synthetic example, questionnaire specification, or metadata extract

Software prerequisites:
• Cursor and GitHub accounts (required)
• Python 3.10+ recommended for examples; not required for participants using R/Stata/SPSS patterns only
• Participants must avoid bringing real restricted microdata unless explicit permission exists for the exact tool workflow

Preparatory reading

Required before Day 1:
• Create GitHub and Cursor accounts
• Complete the course setup guide and prework checklist (course repository)
• Choose one role track: applied researcher, survey practitioner, or survey data manager
• Select a non-sensitive workflow or use the provided synthetic survey example
• Read the restricted-data safety guidance

Desirable:
• Attend the pre-course setup clinic if you have limited admin rights or limited Git/GitHub experience
• Review basic GitHub concepts: repositories, issues, branches, and pull requests
• Install minimum viable toolset: Cursor, browser access to GitHub, ability to download/unzip files
• Optional recommended setup: Git, Python 3.10+, Node.js 18+, and any role-specific tools (R, Stata, SPSS)

Course materials

Participants will receive course slides and access to course repository materials.

Participants should have/bring:
• A computer with internet access and ability to join live video sessions and share screen
• GitHub and Cursor accounts (set up before Day 1)
• A non-sensitive workflow description, public dataset, synthetic example, questionnaire specification, or metadata extract for their capstone work
• Headphones/microphone recommended for breakout-room work

Participants should NOT bring real restricted microdata unless explicit permission exists for the exact tool workflow.