The Agentic Paradigm Shift
The "Why" Behind the Course
Traditional Product Owners spend 70% of their time writing user stories, grooming backlogs, and explaining acceptance criteria to developers. In the AI Agile Operating Model, these mechanical tasks are fully automated.
The AI-Enabled PO reclaims that time to focus entirely on synthetic user testing, market strategy, and outcome prediction. You are moving from a "Backlog Administrator" to a "Market Predictor."
- Deploy Synthetic Personas: Use LLMs to simulate hundreds of customer interactions to validate feature hypotheses before engineering begins.
- Automate Strategic Alignment: Build custom AI scoring engines to calculate WSJF (Weighted Shortest Job First) and remove cognitive bias.
- Execute Zero-Touch Backlogs: Prompt AI to autonomously slice massive Epics into INVEST-compliant User Stories with perfect BDD criteria.
- Predict Delivery Risk: Analyze raw sprint telemetry to identify hidden bottlenecks.
Facilitation Strategy
Throughout the class, trainers must constantly reinforce the Agile Visa philosophy:
- Rule 1: AI does not replace the Product Owner. AI replaces the administration. The human becomes the editor and strategist.
- Rule 2: Never teach "Prompt Engineering" in isolation. Always teach "Agentic Workflows" (systems where AI performs tasks sequentially with human oversight).
- Rule 3: The Trap/Critique methodology is vital. Let them fail with basic prompts first so they feel the value of the Master Prompts.
You will have traditional Agile purists in the room who say, "Agile is about individuals and interactions over tools."
Your response: "Exactly. By offloading ticket-writing and data-crunching to AI, the Product Owner finally has time to actually interact with individuals (customers and stakeholders) instead of staring at Jira all day."
The Synthetic Persona
Context & Theory
Traditional Product Owners wait 3 weeks to run a focus group or launch an A/B test to see if a feature is valuable. By the time you get the data, the sprint is over. Today, we kill guesswork by building a Synthetic Customer Persona. We feed an LLM real, messy support tickets, tell it to "become" our most frustrated user, and pitch our roadmap to the AI before we write a single line of code.
Student Action: Copy the dummy data from your industry (in the Data Repo) and paste it into ChatGPT with this prompt.
Read this customer feedback data and summarize what our users want.
Trainer Review: Look at the output. It gave you a bulleted executive summary. An executive summary hides the bleeding. It smoothed out the anger. If you build a product based on this summary, you build a mediocre product for an average user.
Student Action: Open a new chat. Paste your industry data along with this constrained, role-playing escalation prompt to build your Synthetic Customer.
You are a Forensic Data Analyst and Behavioral Psychologist. I am providing you with unstructured customer feedback. Do not summarize it. Step 1: Cluster the data by "Severity" and "Churn Risk". Extract the top 3 critical bugs or UX failures. Step 2: Transition into a "Synthetic Persona" representing the angriest, most frustrated user from that top cluster. From now on, you will act strictly in character as this Persona. When I pitch a new Epic or Feature to you, you must: 1. Provide a "Value Score" (1-10) based on how much it solves YOUR specific pain. 2. Provide a brutal, cynical critique of why my feature might fail in the real world. 3. Suggest one alternative feature you would prefer I build. Here is the data: [PASTE INDUSTRY DATA HERE]
Algorithmic Prioritization
Context & Theory
How many of you have sat in a 2-hour backlog refinement meeting where stakeholders argue over what gets built first based on their emotions or rank? In the AI Agile Operating Model, we use cold math. We will use an LLM as an unbiased Lean Portfolio Manager to automatically calculate WSJF (Weighted Shortest Job First).
Student Action: Paste the 5 Epics from the Data Repo into the LLM.
Here are 5 feature Epics. Tell me which one our engineering team should build first and why.
Trainer Review: The AI just hallucinated a roadmap based on its training data bias, not your company's reality. It didn't ask you about your budget, your timeline, or your strategic goals. Never let an AI make an unconstrained strategic decision.
Student Action: Force the AI to use cold math. Use this prompt to build a WSJF scoring engine.
Act as an unbiased Lean Portfolio Manager. I will provide you with a list of proposed Epics. Do not prioritize them immediately. Instead, you must force me to provide you with a score (1-10) for each Epic across the following parameters: - User & Business Value - Time Criticality - Risk Reduction / Opportunity Enablement - Job Size / Engineering Effort Once I provide those numbers, calculate the WSJF score using the SAFe formula: (Value + Time + Risk) / Job Size. Output the final result in a perfectly formatted Markdown table sorted by the highest WSJF score to lowest. Here are the Epics: [PASTE 5 EPICS HERE]
Zero-Touch Backlog Engineering
Context & Theory
A Product Owner is an orchestrator of value, not a Jira typist. Today, we automate execution. We are going to take the winning Epic from yesterday's WSJF exercise and feed it into our Execution Agent. The AI will slice the Epic, write the stories using the INVEST framework, and generate Gherkin acceptance criteria instantly.
Student Action: Take the top-ranked Epic from your WSJF table and ask the AI to slice it.
Break this Epic down into user stories for my dev team: [INSERT EPIC]
Trainer Review: Look closely at the stories. "Design the Database," "Build the API," "Create the UI." The AI just sliced horizontally. This is anti-Agile. If you give this to your team, they will build technical silos and deliver zero usable value until the end of the sprint.
Student Action: We must force the AI to adhere to strict Agile frameworks (INVEST) and generate ironclad Acceptance Criteria (Gherkin).
I am providing an Epic. Act as an elite Enterprise Product Owner. Break this Epic down into exactly 6 logically sequenced User Stories. CRITICAL CONSTRAINTS: 1. Slice vertically. Every story MUST deliver end-to-end user value. 2. Every story must strictly adhere to the INVEST principle (Independent, Negotiable, Valuable, Estimable, Small, Testable). 3. For each story, provide exactly 3 Acceptance Criteria written strictly in BDD Gherkin format (Given / When / Then). Ensure at least one is a negative path/edge case. Format the output cleanly using Markdown headers so I can easily copy/paste it into Jira. The Epic: [INSERT EPIC]
Quality Control & Telemetry
Context & Theory
LLMs are inherently sycophantic. They want to please you and tell you your work is great. To ensure high-quality software delivery, Product Owners must learn to force AI into an adversarial, critical posture. We call this Reflective QA Prompting.
Student Action: Ask the AI to review the stories it just generated.
Are these user stories ready for the dev team, or did you miss anything?
Trainer Review: The AI likely replied, "Yes, they look great! They follow the INVEST model perfectly." Never trust an LLM to review its own unconstrained output without a persona switch.
Student Action: Execute "Reflective Prompting" to force the AI to tear down its own work.
Forget your previous persona. Act as the world's most pedantic, aggressive Lead QA Engineer. Interrogate the Gherkin acceptance criteria you just wrote. Your goal is to find 3 massive loopholes where a junior developer might make a wrong assumption and break the system. List the 3 loopholes, explain the business risk of each, and rewrite the acceptance criteria to close them.
Data Repository (BYOC)
Select your industry below. Use this data for Module 1 and Module 2.