Hands On Full Agentic Workflow

Overview

This topic is the capstone of Module 9 and of the integrated agentic product management curriculum. Everything that has come before — context engineering, discovery automation, requirements generation, prioritization, delivery monitoring, and measurement intelligence — converges here in a single end-to-end exercise. You will run the full agentic PM loop for a real product initiative from discovery through measurement, using the techniques, prompts, and review protocols from this module and the entire course.

The goal of this hands-on session is not to generate outputs quickly. It is to experience the full loop as an integrated workflow, to observe where AI assistance provides the most leverage, and to develop a clear, personal understanding of where your judgment as a PM is most essential. These two observations — where automation wins, and where human judgment is irreplaceable — are the foundation of a sustainable agentic product management practice.

This is a learning exercise as much as a production exercise. You should approach it with genuine curiosity about your own workflow: what changes feel natural and high-value, what feels uncomfortable or uncertain, and what gaps in your current practice the agentic workflow is revealing. The reflection at the end of this topic is not a formality — it is the highest-value output of the entire exercise, because it tells you specifically what to build and practice next.

You will work through four phases: setup (30-45 minutes), execution of the full loop (90-120 minutes), stage-gate decision-making (20-30 minutes), and reflection (30-45 minutes). The total investment is approximately three to four hours. For most PMs, this single session will produce more structured, evidence-connected product planning output than a full day of traditional planning work — which is itself a data point worth reflecting on.

Set Up the Agentic Workflow — Discovery Inputs, Planning Context, and Delivery Integration

The quality of every AI interaction in the agentic workflow depends on the quality of the setup. Like a skilled craftsperson who spends a third of their project time on preparation, the PM who invests in thorough setup consistently produces better agentic workflow outputs than one who rushes to the execution phase. Setup is not overhead — it is leverage.

The setup phase has three components: discovery input preparation, planning context assembly, and delivery integration configuration. Each must be completed before the execution phase begins.

Discovery Input Preparation. Identify the product initiative you will run through the full loop. This should be a real, currently relevant initiative — not a hypothetical. Choose something that is either in early discovery, recently approved as an opportunity, or about to enter sprint planning. The more real and current the initiative, the more useful the exercise will be, because you will be evaluating AI outputs against your own direct knowledge of the situation.

For your chosen initiative, gather or create the following discovery inputs:

Customer Signal Set: a collection of at minimum 10-15 customer signals related to the problem area. These can be support tickets, interview notes, review excerpts, NPS verbatims, or any combination. If you do not have real signals, create representative examples based on your knowledge of the problem area. Quality matters more than volume — 10 well-documented, specific customer signals produce better synthesis than 30 vague one-liners.

Competitive Context: a brief on the competitive landscape relevant to this opportunity. Who is solving a similar problem? How are they solving it? What are the gaps in their approach? Even a one-page competitive context document significantly improves the quality of opportunity framing.

Strategic Context Document: a brief document that contains your current OKRs, your product vision statement, and the high-level roadmap themes for the current quarter. This is the strategic lens through which all AI outputs in the workflow will be interpreted.

Planning Context Assembly. Before moving to the planning phase, prepare the planning context documents that the AI will need:

Current Roadmap Snapshot: a structured list of your current roadmap items with their labels, rationale, quarter assignments, and current priority scores (RICE or equivalent). Format this as a clean table or structured list — not a visual roadmap image.

Team Capacity Profile: a brief document describing your team's typical sprint velocity, current team composition, any known capacity constraints for the upcoming sprint, and your team's Definition of Ready and Definition of Done. This context allows the AI to generate capacity-aware sprint plans rather than abstract lists of stories.

Technical Context Brief: a one-to-two paragraph description of the relevant technical constraints, existing architecture components, and key integrations that affect what can be built and how. Ask your tech lead to contribute to this brief if possible — the more accurate it is, the fewer technical feasibility issues will arise in the plan review.

Delivery Integration Configuration. Before running the delivery phase of the exercise, configure the data sources and document structures that the AI will monitor and generate from:

Sprint Tracking Source: identify where your team tracks sprint work (Jira, Linear, GitHub Issues, or similar). For the exercise, you will either use real data from an active sprint or simulate sprint progress data. If using simulation, create a realistic sprint state document: 5-8 stories with statuses, point estimates, and brief progress notes.

Communication Template Library: gather or create the communication templates you will need during delivery — the standup summary format, the stakeholder update format, and the risk alert format. Having these templates ready before the execution phase means you can focus on output quality rather than format design during the exercise.

Hands-On Steps

1. Select your initiative: write a one-paragraph Initiative Brief that includes the problem statement, the target user segment, the strategic rationale, and the intended outcome. This brief is the north star for the entire exercise — every AI prompt in the workflow will reference it.
2. Gather customer signals: collect or create your Customer Signal Set (minimum 10-15 signals). Organize them in a simple document with one signal per line, including source type (support ticket, interview, review, etc.) and date if available.
3. Build the Strategic Context Document: write it fresh for this exercise rather than copying an existing document. The act of writing it forces you to articulate your strategic context explicitly, which often reveals ambiguities or gaps in your current understanding.
4. Prepare the Current Roadmap Snapshot in structured table format: Item Name | Quarter | One-sentence rationale | Priority score | Key dependencies. Fill in all rows, even for items where the rationale is informal or the priority score is estimated.
5. Complete the setup checklist before beginning execution: [ ] Initiative Brief written [ ] Customer Signal Set collected [ ] Competitive Context brief completed [ ] Strategic Context Document written [ ] Roadmap Snapshot prepared [ ] Team Capacity Profile documented [ ] Technical Context Brief completed [ ] Sprint Tracking Source identified [ ] Communication Templates ready. Do not begin execution until all items are checked.

Prompt Examples

Prompt:

You are a product management workflow advisor. I am about to run the full agentic PM loop for a product initiative. Before I begin the execution phase, I want to verify that my setup is complete and that my context documents are high enough quality to produce useful AI outputs.

I will share my setup documents with you and I need you to assess their quality and completeness.

Document 1 — Initiative Brief:
[Paste your Initiative Brief]

Document 2 — Customer Signal Set:
[Paste your Customer Signal Set]

Document 3 — Strategic Context Document:
[Paste your Strategic Context Document]

Document 4 — Roadmap Snapshot:
[Paste your Roadmap Snapshot table]

Document 5 — Team Capacity Profile:
[Paste your Team Capacity Profile]

For each document, assess:
1. Completeness: are all the required fields and sections present?
2. Specificity: is the information specific enough to guide AI outputs, or is it too vague?
3. Consistency: are there any internal inconsistencies across the documents that might confuse or mislead AI outputs?
4. Missing information: what critical information is absent that could significantly degrade output quality?

After reviewing all documents, give me:
- An overall setup readiness rating: Ready to Execute / Needs Minor Revision / Needs Major Revision
- A prioritized list of the 3 most important improvements to make before beginning execution
- A specific prompt adjustment I should make if I proceed with any missing information (i.e., how to compensate for gaps in the context)

Expected output: A structured quality assessment of each setup document with specific, actionable improvement recommendations and a readiness verdict. This output prevents the common failure of running AI prompts with inadequate context and then wondering why the outputs are generic or misaligned.

Learning Tip: The setup phase is the highest-leverage part of the entire agentic workflow exercise. Every 10 minutes invested in improving context document quality before execution translates into 30+ minutes saved in reviewing and revising poor-quality AI outputs. Experienced agentic PMs spend roughly 30-40% of their total workflow time on setup and context preparation. If you find yourself spending less than 20% on setup, the quality of your outputs is almost certainly lower than it could be.

Execute Each Step of the Agentic PM Loop with AI Assistance

With setup complete, you are ready to run the full loop. This phase moves through all four stages — Discover, Define, Deliver, and Measure — sequentially, using AI assistance at each step. Work through each step in order. Do not skip ahead. The sequential execution is important because each step's output becomes the next step's input, and skipping a step means the downstream steps will lack the specific context they need.

Stage 1: Discover — Customer Signal Synthesis.

Feed your Customer Signal Set into the synthesis prompt from Topic 02. Your goal is to produce a prioritized Opportunity List from the signals. Run the synthesis, review the output, and make the first stage gate decision: do the AI-generated opportunities match your intuition about the most important problems in this space? Which opportunities surprised you? Which are missing that you expected to see?

From the Opportunity List, select the highest-priority opportunity to advance through the full loop. Write your selection rationale in one sentence — this is the first human judgment decision of the exercise.

Stage 2: Discover → Define — Opportunity Statement Generation.

Convert your selected opportunity into a structured Opportunity Statement using the format from Topic 01 (problem, user segment, evidence, impact estimate, strategic alignment, RICE score). Use AI to draft the statement, then review and refine it personally. The refinement process is itself valuable — where you disagree with the AI's framing or scoring reveals your own implicit assumptions about the opportunity.

Stage 3: Define — Requirements Brief Generation.

Feed the approved Opportunity Statement into the Requirements Brief generation prompt from Topic 02. Your goal is to produce a sprint-ready Requirements Brief: epic description, user stories, acceptance criteria, and known unknowns. Run the generation, then apply the INVEST quality check. For each story, assess: Is it Independent? Negotiable? Valuable? Estimable? Small enough? Testable?

Make at minimum one substantive revision to the AI-generated requirements before proceeding. The revision is important — it forces you to engage with the content critically rather than accepting the first draft as authoritative.

Stage 4: Define → Deliver — Sprint Plan Generation.

Feed the approved Requirements Brief into the sprint plan generation prompt from Topic 03. Your goal is to produce a sprint plan: a proposed set of stories from the brief, sequenced, estimated, and matched to your team's capacity. Run the generation, then apply the five-dimension plan review checklist (feasibility, sequencing, dependency coverage, capacity alignment, human judgment inputs).

Issue a Plan Review Verdict: Commit / Revise / Reduce scope / Hold. Write the verdict with a one-sentence justification. This is the second major human judgment decision of the exercise.

Stage 5: Deliver — Alignment and Risk Monitoring.

Simulate Day 5 of the sprint by creating a sprint state document: for each story in your plan, write a brief "current status" note as if you were an engineer updating Jira at the midpoint of the sprint. Include at least one alignment divergence (an implementation detail that deviates from the acceptance criteria), one scope creep signal (a story that has grown), and one risk (a dependency that is delayed).

Feed this sprint state into the alignment check prompt from Topic 04 and the scope creep detection prompt. Review the outputs: did the AI catch the alignment divergence, scope creep, and risk you embedded? What did it miss? What did it surface that you had not thought to flag?

Stage 6: Measure — Hypothesis Validation and Intelligence Routing.

Create a set of "Week 2 post-launch metrics" for the initiative: make up plausible metric data that includes one positive surprise, one negative surprise, and one neutral result. Feed this into the measurement chaining prompt from Topic 05.

Review the AI's output: what Opportunity Hypotheses did it generate from the anomalies? What Roadmap Adjustment Triggers did it propose? Do these outputs accurately reflect the strategic implications of the simulated metric data?

Hands-On Steps

1. Run each stage sequentially with a 10-minute time limit for reviewing and revising AI outputs at each stage. The time box forces you to prioritize your review effort on the most important decisions rather than perfecting every detail.
2. Document each AI output in a single running document — your "Agentic Workflow Transcript." Paste each prompt, AI output, and your review decision (with reasoning) in sequence. This transcript becomes a valuable reference for future workflow runs.
3. At the end of each stage, write two sentences in your transcript: what the AI did well in this stage, and what required the most human correction. These observations accumulate into the reflection input for Phase 4.
4. Resist the temptation to go back and re-run earlier prompts when a later stage reveals a gap. Instead, note the gap in your transcript and keep moving forward. The gaps are learning data — they tell you what context was missing from the earlier setup documents.
5. At the end of all six stages, read the full transcript from beginning to end. Trace one specific piece of information — a specific customer pain from the Signal Set — through all six stages. Where did it appear? Where was it lost? Where did it drive a specific decision? This tracing exercise reveals the coherence (or lack thereof) of the end-to-end workflow.

Prompt Examples

Prompt:

[This is the Discovery-to-Opportunity-Statement chaining prompt — step 2 of the execution sequence]

You are a product discovery analyst. I have just completed a customer signal synthesis and produced a prioritized Opportunity List. I now need to convert my selected highest-priority opportunity into a structured Opportunity Statement ready for the planning phase.

My selected opportunity from the synthesis output:
Theme name: [paste from synthesis output]
Description: [paste from synthesis output]
Frequency: [paste]
Representative quotes: [paste 2]
Affected user segments: [paste]
Initial RICE score: [paste]

My product context for scoring and framing:
- Product: [1-sentence description]
- Target persona: [2-3 sentences]
- Current OKRs: [list]
- Current quarter roadmap themes: [list]

Generate a complete Opportunity Statement with the following fields:
1. Problem Statement: a precise, user-centered description of the problem (2-3 sentences, no solution language)
2. User Segment: which specific users experience this problem and with what frequency
3. Evidence Summary: what the synthesis found, with frequency counts and representative quotes
4. Impact Estimate: if this problem were solved, what would change? Expressed as behavioral outcomes, not just satisfaction
5. Strategic Alignment: which OKR or roadmap theme does solving this opportunity advance, and how directly?
6. RICE Score: calculate Reach, Impact, Confidence, and Effort scores using the following rubric: [paste your scoring rubric]
7. Confidence Level: High / Medium / Low — with a one-sentence justification
8. Recommended next step: should this opportunity proceed to requirements definition, require additional research first, or be held for a future quarter? Why?

Format this as a single structured document I can add directly to my opportunity backlog.

Expected output: A complete, formatted Opportunity Statement that is ready to enter the planning phase as the primary input for Requirements Brief generation. The recommended next step field is particularly important — it provides a PM decision prompt that prevents good opportunities from stalling between stages.

Learning Tip: The most revealing moment in the execution phase is not when the AI produces a good output — it is when it produces an output that is confidently wrong. Pay close attention to moments where the AI's output sounds plausible but contradicts something you know from your own product context. These moments tell you what critical context was missing from your setup documents. Every time you catch a confidently wrong output, add the missing context to your context document library so it is available for future runs.

Review Outputs at Each Stage and Make Go/No-Go Decisions

The go/no-go decision at each stage gate is the most consequential act in the agentic PM workflow. It is the moment when the PM's judgment integrates with AI-generated analysis to produce a commitment — an agreement to invest organizational resources and team effort in a specific direction. Done well, this decision is explicit, documented, and based on clear criteria. Done poorly, it is implicit, undocumented, and driven more by momentum than evidence.

There are four stage gates in the full agentic PM loop, each with specific decision criteria:

Gate 1: Discovery → Definition (Go/No-Go on Opportunity).

The question at this gate is: is this opportunity compelling enough, and well-enough evidenced, to invest the time required to define it fully? The decision criteria are:

• Evidence quality: is the evidence for this opportunity from multiple independent sources, or from a single channel?
• Scope clarity: can we articulate the problem clearly enough that we could recognize a solution when we see one?
• Strategic fit: does this opportunity clearly advance a current OKR, or is it interesting but strategically peripheral?
• Timing: is this the right time to address this opportunity, or is there a dependency or market condition that suggests waiting?
• Confidence: given what we know, how confident are we that solving this problem will produce a measurable positive outcome for users?

A Go decision at Gate 1 means: "We have sufficient evidence, strategic alignment, and problem clarity to invest in defining this opportunity fully." A No-Go means: "We need more evidence, clearer strategic framing, or a different timing before we invest in requirements definition." A Conditional Go means: "Proceed to definition, but address [specific gap] within [specific time] or revisit this gate."

Gate 2: Definition → Delivery (Go/No-Go on Sprint Plan).

The question at this gate is: are we confident enough in this sprint plan to commit the team's execution effort for the next two weeks? The decision criteria are the five dimensions from the plan review checklist: feasibility, sequencing, dependency coverage, capacity alignment, and human judgment inputs. A Go decision means all five dimensions have been reviewed and no blocking issues remain. A No-Go means blocking issues (typically dependency failures or feasibility concerns) are unresolved. A Conditional Go means: "Proceed to sprint planning, but resolve [specific item] before Day 3 of the sprint."

Gate 3: Delivery Midpoint (Continue/Re-plan).

The question at this gate is: given the sprint's current state at the midpoint, are we on track to achieve the sprint goal, or does the plan need to change? The decision criteria are the risk detection outputs from Topic 04: scope creep assessment, delivery pace, dependency status, and quality signal. A Continue decision means: "Sprint is on track. Continue with current plan." A Re-plan decision means: "Context has changed sufficiently to warrant a revised sprint commitment. Generate re-planning recommendation and communicate to stakeholders."

Gate 4: Measurement → Next Discovery (Iterate/Pivot/Scale).

The question at this gate is: based on measurement data from the delivered feature, what should the next action be? There are three options:

• Iterate: the feature is producing positive signals but below target performance. The direction is right; the execution needs refinement. Generate a set of iteration hypotheses and add them to the discovery queue.
• Pivot: the feature is producing signals that indicate the problem statement or solution approach was wrong. The direction needs to change. Generate a Hypothesis Revision and re-enter the discovery stage with updated framing.
• Scale: the feature is performing at or above target. The next question is: how do we extend this success to additional user segments, use cases, or market areas? Generate scaling opportunity hypotheses and add them to the discovery queue.

Hands-On Steps

1. Before the exercise begins, write your Go/No-Go criteria for each of the four gates in your specific context. Customize the generic criteria above to match your product, team, and current strategic priorities. These criteria are your commitment to yourself about what "good enough" looks like at each gate.
2. At each gate during the execution phase, make the decision explicitly and in writing — not mentally. Write "Go / No-Go / Conditional Go" and one sentence of reasoning. Practice this decision discipline even when the answer seems obvious.
3. For any Conditional Go decision, write the specific condition that must be met and the deadline by which it must be met. Vague conditions ("this should be clarified before we start") are not conditions — they are deferrals. A specific condition ("engineering lead must confirm the API rate limits will not affect story #3 by the end of Day 1 of the sprint") is actionable.
4. After all four gate decisions are made, read them in sequence: do the decisions tell a coherent story? Is the evidence at Gate 4 consistent with the assumptions you made at Gate 1? Where did your early-stage assumptions hold up, and where were they revised by subsequent evidence?
5. Identify the gate decision that required the most deliberation during the exercise. This is the gate where your judgment was most exercised — and likely the gate where your criteria were least clear. Revise your criteria for that gate based on what you learned during the deliberation.

Prompt Examples

Prompt:

You are a product strategy advisor helping me make a stage-gate decision in the agentic PM loop. I am at Gate 1 (Discovery → Definition) and I need to decide whether to invest in fully defining this opportunity.

Opportunity Statement:
[Paste your full Opportunity Statement from Stage 2 of the execution phase]

My gate decision criteria:
- Evidence quality: [describe the evidence quality standard I require before proceeding]
- Scope clarity: [describe the scope clarity standard]
- Strategic fit: [describe the strategic alignment standard]
- Timing considerations: [list any timing factors relevant to this decision]

Please assess this opportunity against each criterion and:
1. Provide a criterion-by-criterion assessment (Meets / Partially Meets / Does Not Meet) with one-sentence justification for each
2. Identify the single weakest criterion — the one that most argues against proceeding
3. Provide a Go / No-Go / Conditional Go recommendation with a clear, specific rationale
4. If Conditional Go: specify the exact condition that must be met before proceeding, and suggest how to meet it
5. If No-Go: specify what additional evidence or analysis would change this to a Go decision

Do not just tell me "this looks good." Give me a genuinely critical assessment that surfaces the weakest parts of this opportunity's case. My team is about to invest significant time if I say Go — the value of your input is in finding the reasons to be cautious, not in validating my enthusiasm.

Expected output: A critical, criterion-by-criterion assessment of the opportunity with a specific Go/No-Go/Conditional Go recommendation and clear rationale. The instruction to be genuinely critical (rather than validating) is essential — without it, AI tends to produce affirming outputs that do not serve the PM's decision-making needs at a high-stakes gate.

Learning Tip: Gate decisions are where PMs most often let AI outputs substitute for judgment rather than inform it. The warning sign is a gate decision that is made primarily because "the AI said the output was good." AI can assess whether the output meets structural and content quality standards; it cannot assess the strategic context, organizational dynamics, or relational factors that often determine whether a product initiative should proceed. Always have at least one non-AI reason for a Go decision — one piece of evidence or context that comes from your own judgment, not from the AI's assessment.

Reflect on What the Agentic Workflow Automated vs. Where Human Judgment Was Essential

The reflection phase is the most important 30 minutes of the entire exercise. It is where individual observations become transferable learning, and where the experience of running the full loop becomes the foundation of a personal agentic PM practice. This is not a casual debrief — approach it as a structured analysis of your own performance and the workflow's performance.

The reflection framework has four components:

Component 1: Automation Value Assessment. Review your Agentic Workflow Transcript and identify every task that AI performed that you would have done manually in a traditional workflow. For each, estimate: how long would this have taken you manually? How does the AI output quality compare to what you would have produced? Is the AI output good enough to use directly (with review), or does it consistently require significant revision?

Summarize your findings as an Automation Value Statement: "AI provided high value in [specific tasks], medium value in [specific tasks], and low value or required more human effort than it saved in [specific tasks]." This statement is the empirical foundation of your personal automation strategy.

Component 2: Human Judgment Point Inventory. Review your Transcript again, this time identifying every moment where you made a judgment call that AI could not have made — decisions where your knowledge of strategic context, organizational dynamics, customer relationships, or technical history was essential. These are your Human Judgment Points.

List each Judgment Point with a one-sentence description of the context knowledge it required: "Decided to deprioritize Opportunity #2 despite its high RICE score because [strategic context the AI did not have]." The list of Judgment Points reveals what is irreplaceable about the PM role in an agentic workflow. These are the competencies to invest in and protect, even as AI handles more of the mechanical work.

Component 3: Workflow Improvement Opportunities. Identify three to five specific improvements to make to the workflow before the next run. For each, write: the specific problem that occurred in this run (what went wrong, what was missing, what produced poor output), the specific improvement to make (revise the context document, add a new template, change the prompt, add a review step), and the expected impact of the improvement (what output quality improvement should result?).

Common improvement opportunities that emerge from first-run exercises: context documents that were too vague, missing technical context that caused feasibility issues in the sprint plan, prompts that did not specify output format precisely enough, or review time estimates that were too optimistic.

Component 4: Personal Agentic Practice Commitments. Based on everything observed in the exercise, write three specific, time-bound commitments about changes to your regular product management practice:

• One change to implement this week: something small that will immediately improve your daily AI-assisted PM work.
• One change to implement in the next sprint: a workflow adjustment that requires building a new prompt, template, or review protocol.
• One change to implement this quarter: a larger structural change to how your team approaches discovery, planning, delivery monitoring, or measurement.

These commitments are your personal agentic PM development plan. Write them with enough specificity that you can evaluate whether you followed through at the end of each time horizon.

Hands-On Steps

1. Immediately after completing the execution phase, spend 10 minutes writing unfiltered observations in your Transcript before any reflection analysis. What surprised you? What frustrated you? What exceeded your expectations? Raw observations made immediately after an experience are more accurate than polished reflections made hours later.
2. Run the Automation Value Assessment: go through your Transcript task by task and estimate the time the AI saved. Total the time savings across the full loop. This is your measured ROI for one initiative run — extrapolate it to a full quarter to understand the strategic value at scale.
3. Run the Human Judgment Point Inventory: list every moment you made a decision that AI could not. For each, identify the type of knowledge it required: strategic context, organizational context, customer relationship, technical history, or political/relational dynamics. This categorization reveals the nature of PM judgment work in an agentic workflow.
4. Write the Workflow Improvement Opportunities in a structured format: Problem | Improvement | Expected Impact | Owner (you) | Deadline. This converts observations into commitments and prevents the common failure of completing a reflection without changing anything.
5. Share your reflection with at least one other PM or colleague who also runs this exercise. Compare your Human Judgment Point inventories: do you identify the same judgment types as essential, or do your inventories differ based on your different product contexts? Cross-comparing reflections often surfaces blind spots that solo reflection misses.

Prompt Examples

Prompt:

You are a product management learning coach. I have just completed a full run of the agentic PM loop for a real product initiative. I am going to share my Agentic Workflow Transcript with you and I need you to help me extract structured learnings.

My Workflow Transcript:
[Paste your full transcript — prompts, AI outputs, review decisions, and personal observations from all six stages]

Please help me analyze this transcript across four dimensions:

1. Automation Value Assessment:
   - For each stage of the loop, identify the tasks where AI provided the most value (time saved + quality maintained or improved)
   - For each stage, identify the tasks where AI output required the most human revision
   - What is the estimated total time saved by AI assistance in this run, compared to doing the same work manually?

2. Human Judgment Point Inventory:
   - Identify every moment in the transcript where my decisions went beyond or corrected the AI output
   - For each, classify the type of knowledge required: Strategic Context / Organizational Context / Customer Relationship / Technical History / Other
   - Which judgment type appears most frequently in this run?

3. Workflow Quality Assessment:
   - Where did AI produce outputs that were confidently wrong or significantly misaligned with the initiative context?
   - What was the root cause of each misalignment — missing context, vague prompt, or AI limitation?
   - Which context documents (setup phase) showed the most impact on output quality, and which showed the least?

4. Improvement Recommendations:
   - Based on the transcript, identify the 3 highest-impact improvements to the workflow for the next run
   - For each, specify: what problem it addresses, what specific change to make, and what quality improvement to expect

Present your findings as a structured Learning Summary Report that I can reference for my next workflow run.

Expected output: A structured Learning Summary Report with an automation value assessment, human judgment inventory, workflow quality analysis, and specific improvement recommendations. This report is the primary output of the entire capstone exercise — more valuable than any single deliverable produced during execution, because it is the foundation of continuous improvement.

Learning Tip: The most common mistake in the reflection phase is being too generic: "AI was helpful for writing stories but less helpful for strategic decisions." This level of generality does not improve your next run. Force yourself to be specific: "AI was helpful for writing the acceptance criteria in Stories #2, #3, and #5, but produced acceptance criteria that were implementation-oriented rather than behavior-oriented in Stories #1 and #4 — likely because the acceptance criteria prompt did not include the behavioral framing instruction from the requirements module." Specific observations produce specific improvements. Generic observations produce generic results.

Key Takeaways

• The full agentic PM loop exercise produces its highest value not from the deliverables generated but from the learning about where AI assistance provides leverage and where human judgment is essential.
• Setup quality is the primary determinant of execution quality. Every 10 minutes invested in context document preparation during setup prevents 30+ minutes of revision work during execution.
• The four stage gates — Discovery→Definition, Definition→Delivery, Delivery Midpoint, and Measurement→Next Discovery — are the PM's most consequential acts in the agentic workflow. Each requires explicit, documented decisions with clear criteria.
• Gate decisions must include at least one non-AI reason for a Go. AI can assess structural quality; only the PM can assess strategic context, organizational dynamics, and relational factors.
• The Automation Value Assessment and Human Judgment Point Inventory are the two most important outputs of the reflection phase. They tell the PM specifically where to invest in building AI capabilities and where to invest in building human judgment.
• The reflection phase should produce three specific, time-bound practice commitments: one change this week, one change this sprint, one change this quarter. Without commitments, reflection produces insight but not improvement.
• The full agentic PM loop, once practiced and calibrated, consistently compresses the time from insight to delivered value by 40-60% for experienced practitioners — not by removing human judgment, but by eliminating the mechanical synthesis and generation work that previously consumed a disproportionate share of the PM's cognitive energy.
• An agentic PM workflow is never finished. Each run produces observations that improve the next run. The compounding value of this continuous improvement is the most important long-term benefit of agentic product management practice.