The Agentic UX Research Repository: Building with AI Agents
Learn to build a customized repository with AI agents that conduct analysis, create cross-project insights, and so much more.
- Format
- 100% hands-on
- Dates
- To be announced
- Instructor
- Leo Hoar, PhD
Course dates coming soon. Sign up to get notified.
What is an agentic UX research repository?
Most research repositories are static, library-like compilations that get stale quickly. Moreover, the actual analytical tools built into them are often limited and not customizable.
Leveraging AI, particularly agentic AI, changes all of that.
An agentic research repository is one that:
- Leverages AI to assist with research tasks using precisely the workflow you specify;
- Uses autonomous and semi-autonomous agents to actively update insights, draw connections between projects, and leverage past research to make planned projects more targeted and efficient.
Learn to leverage the full spectrum of agentic AI
Skills are a great start in the journey toward leveraging AI. But they are only the beginning. A skill is little more than a pre-defined workflow that requires your intervention. There are many more exciting possibilities in the world of agents.
In this course, we'll begin by learning the nuts and bolts of agents and establishing clear principles for agent usage in UX research.
Then we'll decompose our research work and apply a simple framework to determine what's eligible for agentic application, and what needs to be performed strictly by a human research expert.
We'll step beyond thinking about applying AI only to a "workflow," and instead consider opportunities to leverage agentic AI to do laborious tasks we often neglect, like updating insights based on new evidence.
You'll learn how to leverage AI agents to:
- 01
Support robust thematic analysis
- 02
Evaluate and compare human and AI output to assess reliability
- 03
Flag insights for review or retirement in order to ensure the validity of a repository
- 04
Actively interrogate existing insights in light of new data and flag for review
- 05
Connect context, like open questions and assumptions, to ongoing research
- 06
Easily understand how past research relates to a new request
Value of building an agentic research repository
Increased trust
Ever had a stakeholder ask "how do you know"? In an agentic repository, not only does every insight trace back to the exact lines of data that support it, but agents can conduct regular checks to see if insights are superseded or complicated by new data.
Guaranteed relevance
Every entity and relationship in the repository is transparently mapped. Agents can then do things like flag insights for review when the business question behind them is decided or retired, or alert you when data from one project answers the research questions in another.
Solves the context window problem
Out-of-the-box repos deploy AI across your data, but almost never have the context window to handle all of it. Agentic repositories do more targeted searches, which means they'll never skip or ignore chunks of your data.
Build in checks where AI is unreliable
AI can't guarantee the fidelity of quotes, but simple bits of Python script can. The course provides these and shows you how to incorporate them.
Your repo can now play a role in planning and prioritization
With all recent, valid insights readily accessible, new projects start from what the repository already knows.
Stakeholders can query it directly and get insights with their lineage.
No stakeholder will ever be shown an expired or superseded insight.
It is plain files on your machine.
You own it and you can change it.
What it is
A course in which you build an agentic AI research repository that executes customized analytical workflows, enables cross-project insights, and enforces a robust connection between insights and raw data. Working in the CLI tool of your choice (Claude Code, Gemini CLI, or Chat-GPT CLI), you will build everything you need, and will leave with a working folder on your own machine: a protocol written as files, subagents that code each transcript blind, a second coder that reconciles them, scripts that reject any quote that is not verbatim, and gates that stop the run until you decide. By the final session, twelve transcripts have been coded twice, reconciled, verified, calibrated against your own coding, themed, and turned into insights where every claim traces to a line number.
This course picks up where Beyond Prompting leaves off. That course helps you make all the decisions about what context, memory, and other crucial elements need to contain for accurate, reliable performance.
Who it is for
Researchers who already do qualitative work and are familiar with the function of a repository. No command-line or coding experience is assumed; the course teaches the few commands it needs and keeps every instruction in plain language. Beyond Prompting is recommended, not required.
What you build
An agentic UX research repository, filled in over the course from annotated templates. Every file below is something you will write, run, and take home.
Instructions
CLAUDE.md: the house rules every session and every subagent reads. Quote verbatim. Never invent a code. A method change is not in effect until it is written into the codebook.- A skill: your analysis protocol as seven step documents (codebook, calibration, coding, verification, themes, negative case, insights), each with inputs, a procedure, outputs, and a pass test.
- Format files: the exact layout of a coding file, a reconciliation table, a themes file, an insight file.
Subagents
transcript-coder: codes one transcript against one codebook, sees nothing else, writes one file.second-coder: compares two codings of the same transcript and classifies every disagreement.negative-case-hunter: given a theme it did not build, finds every passage that cuts against it.
Scripts
verify_quotes.py: every quoted passage is an exact substring of the transcript at the cited lines.agreement.py: how closely the subagent's coding matches yours on a calibration set, before coding begins for real.check_codebook.pyandcheck_insights.py: the codebook and insight files are structurally complete.traceability_check.py: every insight's excerpts appear in its theme, every theme's excerpts appear in a coding file, every coding row verifies against the transcript.
The study folder
- A versioned codebook with include and exclude rules, boundary cases, and a change log.
- Alignment notes: reading rules written from calibration disagreements.
- A state file the coordinator updates before every run ends, so any session can resume.
- Calibration records, coding files, a reconciliation table, themes, and insights as files with IDs and relations.
The business context folder
- Open questions with IDs, assumptions with IDs, current bets, a glossary. An insight is not an insight until it points at one of these.
Every file comes blank with field-by-field guidance, and filled in for the course's running example, a fictional twelve-interview study of how small business owners choose payroll software.
How it is taught
This course is 100% hands-on. You'll start by building the essential core of the repository, and then we'll continue to build up in layers. The course will present just enough theory to enable you to understand the why and how behind each layer.
If you are currently in a role with a need for a repository, you are welcome and encouraged to build the real thing in the course, while observing all confidentiality requirements for your employer.
If you're between roles, the course will still have value, as it's an opportunity to embody how you think about research in a functioning repository.
What it is not
This is not an engineering course. We won't be writing any code, though we will learn some key concepts so that we understand how to review AI output.
Leo Hoar, PhD
Leo founded the UXR Institute after years of working at the intersection of qualitative research and the kind of evidence skepticism that researchers routinely face in product and business settings. Full bio →
Create a repository that works for you.
Course dates coming soon. Sign up to get notified.
Get notified →Loading the question form…

