PSM-AI — Scrum.org Professional Scrum Master - AI Essentials Exam Blueprint
Practical PSM-AI exam blueprint for Scrum Masters reviewing Scrum, AI use, responsible practices, scenarios, and final readiness.
How to Use This Exam Blueprint
Use this checklist as a practical study map for the Scrum.org Professional Scrum Master - AI Essentials (PSM-AI) exam. It is not an official Scrum.org outline and does not claim exact exam weights. The goal is to help you verify whether you can apply Scrum Master judgment when artificial intelligence is used by a Scrum Team, embedded in a product, or discussed with stakeholders.
Work through the sections in three passes:
- Coverage pass: Mark every topic as reviewed or not reviewed.
- Scenario pass: Practice deciding what a Scrum Master should do next, what to make transparent, and what not to control.
- Final-readiness pass: Confirm you can explain your choices using Scrum, empiricism, ethics, value, and risk.
Exam identity
| Item | Checklist use |
|---|---|
| Vendor/provider | Scrum.org |
| Official exam title | Scrum.org Professional Scrum Master - AI Essentials (PSM-AI) |
| Official exam code | PSM-AI |
| Professional vertical | Project Management / Scrum |
| Readiness focus | Scrum Master thinking in AI-enabled contexts |
| What to avoid | Memorizing AI buzzwords without connecting them to Scrum accountability, transparency, inspection, adaptation, and responsible delivery |
Topic-area readiness table
| Readiness area | What to review | Ready means you can… |
|---|---|---|
| Scrum fundamentals | Empiricism, Scrum values, accountabilities, events, artifacts, commitments | Apply Scrum correctly even when the scenario includes AI tools, automation, data, or uncertainty |
| Scrum Master accountability | Coaching, facilitation, impediment removal, servant leadership, organizational change | Choose actions that improve transparency and self-management rather than taking over decisions |
| AI essentials for Scrum contexts | Common AI capabilities, limitations, hallucination, bias, prompt quality, output validation | Explain how AI can assist work while preserving human accountability |
| Responsible AI use | Privacy, confidentiality, security, fairness, transparency, explainability, compliance awareness | Identify risks before using sensitive data or acting on generated outputs |
| AI-assisted Scrum events | Sprint Planning, Daily Scrum, Sprint Review, Retrospective, refinement support | Use AI to support preparation, synthesis, or insight without replacing collaboration |
| Product Backlog and value | Product Goal, Product Backlog ordering, hypotheses, stakeholder feedback, value assumptions | Distinguish AI-generated suggestions from validated product decisions |
| AI product delivery | AI features, data quality, acceptance criteria, Definition of Done, monitoring, model behavior | Recognize what must be transparent before calling an AI-enabled Increment usable |
| Risk and uncertainty | Complex work, empirical learning, experiments, spikes, unknowns, model drift | Recommend inspection/adaptation rather than false certainty |
| Stakeholders and transparency | Expectations, tradeoffs, review evidence, AI limitations, ethical concerns | Communicate uncertainty and evidence clearly without overpromising AI outcomes |
| Governance and organizational constraints | Policies, approved tools, data handling rules, audit needs, escalation paths | Know when the Scrum Team can proceed, when to ask for guidance, and what to make visible |
| Prompting and AI interaction | Clear context, constraints, examples, review criteria, iteration | Build prompts that support useful outputs and know how to evaluate those outputs |
| Exam scenario judgment | “What should the Scrum Master do?” “What should be updated?” “What is the risk?” | Select responses consistent with Scrum, empiricism, accountability, and responsible AI |
Scrum foundation checklist for PSM-AI
AI does not change Scrum’s core accountabilities and commitments. Be ready to apply Scrum first, then reason about how AI affects transparency, risk, and learning.
Empiricism and complexity
Check that you can explain and apply:
- Transparency: Work, progress, risks, assumptions, data quality, and AI limitations must be visible.
- Inspection: The team and stakeholders inspect real evidence, not just AI-generated summaries or optimistic forecasts.
- Adaptation: The team changes plans, backlog items, experiments, or practices based on what is learned.
- Complex work: AI work often includes uncertainty around data, behavior, user trust, ethics, model performance, and integration.
- Empirical control: Use short feedback loops instead of large upfront predictions when uncertainty is high.
Scrum values in AI-enabled work
| Scrum value | AI-context readiness question |
|---|---|
| Commitment | Can the team commit to a Sprint Goal without pretending AI uncertainty is removed? |
| Focus | Can AI support focus without flooding the team with low-value analysis or generated backlog noise? |
| Openness | Are limitations, risks, and assumptions about AI visible to stakeholders? |
| Respect | Are people treated as accountable professionals, not replaced by tool-generated judgments? |
| Courage | Can the Scrum Master raise privacy, bias, safety, or quality concerns even when AI is popular with stakeholders? |
Scrum accountabilities
| Accountability | What remains true in AI contexts | Common exam trap |
|---|---|---|
| Scrum Master | Accountable for establishing Scrum as defined and helping everyone understand theory and practice | Acting as the AI tool administrator who assigns work or approves technical outputs |
| Product Owner | Accountable for maximizing product value and managing the Product Backlog | Letting AI order the backlog without Product Owner judgment and stakeholder learning |
| Developers | Accountable for creating a usable Increment each Sprint | Treating AI-generated code, tests, content, or analysis as done without verification |
| Scrum Team | Collaborates to deliver value and improve effectiveness | Allowing AI outputs to replace conversation, shared understanding, or empirical feedback |
Scrum artifacts, commitments, and AI readiness
| Artifact or commitment | AI-related review focus | Ready check |
|---|---|---|
| Product Goal | Whether AI capabilities support a clear product direction | Can you identify when AI is a solution looking for a problem? |
| Product Backlog | Ordering by value, risk, dependencies, learning, and stakeholder needs | Can you tell whether AI-generated backlog items need Product Owner review? |
| Sprint Goal | A coherent objective for the Sprint | Can you detect when AI uncertainty threatens the Sprint Goal and should be made transparent? |
| Sprint Backlog | Developers’ plan for meeting the Sprint Goal | Can you explain why AI-generated task plans do not replace Developer ownership? |
| Increment | Usable, integrated work that meets the Definition of Done | Can you identify whether an AI-enabled feature is actually usable and inspectable? |
| Definition of Done | Shared quality standard | Can you include validation, security, data handling, and review expectations where relevant? |
Scrum event readiness in AI scenarios
| Event | What AI can support | What AI must not replace | Scenario cues to watch |
|---|---|---|---|
| Sprint Planning | Summarizing backlog context, surfacing risks, suggesting tasks, analyzing dependencies | Product Owner value decisions, Developers’ forecast, team collaboration | “The tool created a Sprint plan; should the team accept it?” |
| Daily Scrum | Highlighting patterns, blockers, aging work, dependency signals | Developers inspecting progress and adapting their plan | “The AI dashboard says everything is on track, but team members disagree.” |
| Sprint Review | Summarizing feedback, showing data, preparing evidence | Stakeholder inspection of the Increment and adaptation of the Product Backlog | “Stakeholders rely only on an AI-generated report instead of inspecting the Increment.” |
| Sprint Retrospective | Clustering feedback, identifying improvement themes, drafting experiments | Psychological safety, team ownership, honest conversation | “The AI identifies who caused delays.” |
| Product Backlog refinement | Breaking down items, identifying acceptance criteria, clarifying assumptions | Product Owner accountability, stakeholder input, shared understanding | “AI generated many user stories; what should happen next?” |
AI essentials checklist for Scrum Masters
Core AI concepts to understand at exam level
You do not need to be a machine learning engineer to reason about AI in Scrum scenarios. You should be able to explain practical implications.
- AI systems generate or recommend outputs based on patterns, data, instructions, and model behavior.
- AI outputs can be useful, incomplete, biased, outdated, fabricated, or contextually wrong.
- A confident answer is not the same as a correct answer.
- Prompt quality affects output quality, but prompt engineering does not remove the need for review.
- AI can accelerate discovery, synthesis, drafting, coding, testing, analysis, and facilitation support.
- AI can also introduce privacy, security, ethical, legal, quality, and transparency risks.
- Human accountability remains with people and the Scrum Team accountabilities, not the AI system.
- Empirical evidence is needed before assuming an AI-enabled change improves value or quality.
AI use cases a Scrum Master should be able to evaluate
| Use case | Helpful when… | Risk if misused | Readiness prompt |
|---|---|---|---|
| Backlog item drafting | It helps create starting points for conversation | Generated items become “requirements” without validation | Who reviews value, user need, and ordering? |
| Acceptance criteria suggestions | It exposes edge cases and test ideas | Criteria are accepted without stakeholder and Developer understanding | Are criteria testable and aligned to value? |
| Meeting summaries | It captures decisions, actions, and themes | Sensitive discussion is uploaded or context is lost | Was consent, policy, and accuracy addressed? |
| Risk identification | It suggests possible failure modes | The team treats generated risk lists as complete | What evidence supports the risk? What was missed? |
| Forecasting and planning | It analyzes historical patterns | It creates false certainty in complex work | Is the forecast transparent and empirically inspected? |
| Code or test generation | It accelerates technical work | Generated work bypasses review, security, or DoD | Does it meet the Definition of Done? |
| Stakeholder communication | It drafts concise explanations | It overstates benefits or hides uncertainty | Are limitations and assumptions visible? |
| Retrospective synthesis | It groups feedback themes | It attributes blame or exposes sensitive input | Does it support safety and team ownership? |
Responsible AI and governance readiness
For PSM-AI preparation, be ready to choose actions that make risks visible, respect organizational policy, and preserve professional accountability.
Responsible AI checklist
- Can you identify when sensitive, confidential, personal, or regulated data should not be entered into an AI tool?
- Can you explain why the team should understand approved tools and organizational policies before using AI?
- Can you detect when AI-generated output needs human review before stakeholder use?
- Can you spot bias, unfairness, exclusion, or representational gaps in AI-supported decisions?
- Can you distinguish explainability needs from simple convenience?
- Can you explain why transparency about AI use may matter to stakeholders and users?
- Can you identify when legal, compliance, security, or privacy experts should be consulted?
- Can you avoid presenting AI-generated analysis as verified fact?
- Can you connect responsible AI concerns to the Definition of Done, Product Backlog refinement, and Sprint Review evidence?
Data and privacy decision checks
| Scenario cue | Better Scrum Master response | Weak response |
|---|---|---|
| Team wants to paste customer data into a public AI tool | Ask about policy, sensitivity, consent, anonymization, and approved tools; make risk visible | “Try it and see if the output is useful.” |
| AI summary includes incorrect stakeholder commitments | Verify with participants and correct the transparent record | Assume the AI transcript is authoritative |
| Developers use AI-generated code | Ensure review, testing, security checks, and DoD expectations are met | Treat generated code as automatically production-ready |
| Product Owner wants AI to rank the Product Backlog | Use AI as input, then apply Product Owner judgment, value, risk, and stakeholder feedback | Let the AI determine priority |
| Stakeholder asks for AI feature promises | Explain uncertainty, assumptions, validation needs, and empirical learning path | Promise performance before evidence exists |
Prompting and output evaluation checklist
Be ready to reason about prompts as a tool for clarity, not as a replacement for professional judgment.
Prompt quality factors
| Factor | What good looks like | Exam-style warning sign |
|---|---|---|
| Context | The prompt states product, users, constraints, and purpose | “Generate user stories” with no product context |
| Role and task | The request defines the desired perspective and output | The AI is asked to “decide” accountability questions |
| Constraints | Security, privacy, format, scope, and assumptions are included | Sensitive data is provided casually |
| Examples | Useful examples guide structure and quality | AI output is accepted without calibration |
| Review criteria | Output is checked against Scrum, DoD, policy, and evidence | The team assumes fluency equals correctness |
| Iteration | The team refines prompts and validates outputs | The first answer becomes the plan |
Can you evaluate an AI output?
Before using AI output in Scrum work, ask:
- Is the output aligned with the Sprint Goal, Product Goal, or stakeholder need?
- Is it based on reliable context, or did the model infer missing details?
- What assumptions are hidden?
- What facts require verification?
- Could the output reveal or misuse sensitive information?
- Could the output be biased, exclusionary, unsafe, or misleading?
- Does the Scrum Team understand it well enough to own it?
- Does it meet the Definition of Done if it becomes part of the Increment?
- Should it be inspected with stakeholders before further investment?
AI-enabled product delivery checklist
Some PSM-AI scenarios may involve a product that uses AI, not just a Scrum Team using AI tools. In those cases, think about value, risk, and empirical validation.
Product Backlog checks for AI-enabled features
- Is the user problem clear, or is the item only “add AI”?
- Is the expected value stated in inspectable terms?
- Are assumptions about data availability, quality, and permissions visible?
- Are risks related to fairness, safety, explainability, misuse, and trust visible?
- Are acceptance criteria testable?
- Is the Product Owner ordering the item using value, risk, learning, and dependencies?
- Does the item require stakeholder feedback before full implementation?
- Is there a learning experiment, prototype, spike, or validation step when uncertainty is high?
Definition of Done considerations for AI-enabled work
The Definition of Done remains a shared quality commitment. For AI-enabled work, readiness means you can identify when the team may need explicit checks such as:
| DoD consideration | Why it matters |
|---|---|
| Functional behavior | The feature does what users need in expected scenarios |
| Edge cases | AI behavior may vary or fail in unusual inputs |
| Data handling | Data sources, permissions, retention, and sensitivity affect risk |
| Security review | AI features may introduce new attack surfaces or data exposure |
| Bias and fairness review | Outputs may affect users differently |
| Explainability or transparency | Users and stakeholders may need to understand AI involvement |
| Monitoring or feedback | AI behavior may need inspection after release |
| Human override or review | Some decisions should not be fully automated |
| Documentation of limitations | Stakeholders need realistic expectations |
| Compliance or policy check | Organizational constraints may apply before release |
Scenario judgment checklist
Use this section to practice “what should happen next?” reasoning. The best answer usually protects Scrum principles, makes important information transparent, and supports the Scrum Team in adapting.
Scrum Master action patterns
| If the scenario says… | Think… | Likely good action |
|---|---|---|
| The team adopted an AI tool without discussing risks | Transparency and policy | Facilitate discussion of risks, data handling, and working agreements |
| AI generated a complete Sprint plan | Developer ownership | Developers inspect and adapt the plan; AI is input, not authority |
| AI predicts the team will miss the Sprint Goal | Empirical inspection | Discuss evidence with Developers and adapt the Sprint Backlog if needed |
| Stakeholders want a fixed commitment for uncertain AI work | Complexity and expectations | Make uncertainty visible and propose empirical validation |
| AI summaries conflict with team memory | Accuracy and transparency | Verify with people and artifacts before acting |
| A manager uses AI analytics to rank individual performance | Scrum values and self-management | Coach away from individual blame and toward system-level improvement |
| Product Owner wants to ship an AI feature despite unknown risks | Increment and DoD | Inspect quality, risk, DoD, and stakeholder impact before release decisions |
| Developers are relying heavily on AI-generated tests | Quality | Review coverage, false confidence, and alignment with acceptance criteria |
| Team members disagree on whether AI use is acceptable | Facilitation | Bring policy, risk, ethics, and team working agreements into a transparent conversation |
| AI produces many backlog ideas | Focus and value | Help the Product Owner refine, validate, and order based on value and evidence |
Artifact update decisions
| Situation | Artifact or practice to inspect |
|---|---|
| New stakeholder feedback changes the value of an AI feature | Product Backlog and Product Goal alignment |
| AI risk threatens Sprint work | Sprint Backlog and Sprint Goal conversation |
| AI quality expectations are unclear | Definition of Done and acceptance criteria |
| Generated backlog items lack user value | Product Backlog refinement |
| The team learned a prompt or tool is unreliable | Working agreements, DoD if quality-related, and retrospective improvement actions |
| A released AI feature behaves unexpectedly | Product Backlog, monitoring feedback, and stakeholder inspection |
| Stakeholders misunderstand AI limitations | Sprint Review evidence and product communication |
“Can you do this?” readiness checklist
Mark each item only if you can answer out loud without notes.
Scrum and AI application
- Explain why AI does not change Scrum accountabilities.
- Explain how empiricism helps manage uncertainty in AI work.
- Identify when AI improves transparency and when it creates false transparency.
- Distinguish AI-assisted collaboration from AI-replaced collaboration.
- Choose a Scrum Master action that coaches instead of commands.
- Identify which Scrum artifact or event should surface a specific AI-related issue.
- Explain why the Product Owner remains accountable for Product Backlog ordering.
- Explain why Developers remain accountable for the quality of AI-assisted work.
- Connect AI risks to Definition of Done, acceptance criteria, and Sprint Review evidence.
- Recognize when organizational policy or expert consultation is needed.
Responsible AI
- Identify privacy and confidentiality risks in AI tool usage.
- Explain why bias and fairness matter to product value and stakeholder trust.
- Recognize hallucinated or unsupported AI output.
- Ask verification questions before using AI-generated content.
- Explain why human review is needed for important decisions.
- Identify when transparency about AI use is appropriate.
- Distinguish experimentation from release readiness.
- Recognize when AI recommendations could undermine Scrum values.
- Identify risks from over-automation.
- Explain why evidence matters more than AI confidence.
Facilitation and coaching
- Facilitate a conversation about team AI working agreements.
- Coach stakeholders on uncertainty without dismissing their goals.
- Help the team inspect AI use in a retrospective.
- Help the Product Owner refine AI-generated backlog ideas.
- Encourage Developers to validate AI-generated work against DoD.
- Make risks visible without becoming the decision owner.
- Support psychological safety when AI analytics or summaries are used.
- Help the organization understand how AI affects Scrum adoption.
- Identify impediments related to approved tools, data access, or policy ambiguity.
- Keep the team focused on value rather than novelty.
Common weak areas and traps
| Trap | Why it is weak | Better exam-ready thinking |
|---|---|---|
| “AI said it, so it is probably right” | AI output can be wrong, incomplete, or fabricated | Verify against evidence, context, and human expertise |
| “The Scrum Master should approve AI use” | Scrum Master is not a command-and-control gatekeeper | Facilitate transparency, policy awareness, and team accountability |
| “AI can replace refinement” | Refinement builds shared understanding | Use AI to support refinement, not replace collaboration |
| “Generated backlog items are ready by default” | Items need value, clarity, ordering, and shared understanding | Product Owner and Scrum Team inspect and refine |
| “AI estimates remove uncertainty” | Complex work remains uncertain | Use forecasts as inputs, inspect progress empirically |
| “More AI output means more value” | Output volume is not product value | Focus on validated outcomes and stakeholder needs |
| “A polished summary is transparent” | It may omit assumptions or errors | Make sources, uncertainty, and decisions inspectable |
| “Retrospective AI analysis should identify underperformers” | This harms safety and misuses Scrum | Focus on team learning and system improvement |
| “AI features are done when they work once” | AI behavior may vary and create new risks | Apply DoD, testing, monitoring, and stakeholder inspection |
| “Policy questions slow the team down, so ignore them” | Unmanaged risk can create larger delays and harm | Make constraints visible early and seek guidance when needed |
Final-week review checklist
Use this as a last-pass readiness filter before practice questions or the real assessment.
Scrum review
- Re-read and summarize Scrum accountabilities, events, artifacts, and commitments.
- Practice explaining transparency, inspection, and adaptation in AI-related scenarios.
- Review the difference between Sprint Review and Sprint Retrospective.
- Review Product Owner vs Developers vs Scrum Master decision boundaries.
- Review how the Definition of Done supports transparency and quality.
- Review how Product Backlog ordering differs from task assignment.
AI essentials review
- Review AI limitations: hallucination, bias, incompleteness, overconfidence, and context gaps.
- Review privacy, confidentiality, and approved-tool concerns.
- Review prompt-quality factors and output validation.
- Review how AI can support but not replace Scrum events.
- Review AI-enabled product risks: data, explainability, monitoring, user trust, safety.
- Review when to escalate to policy, security, legal, compliance, or domain experts.
Scenario practice
- For each scenario, identify the Scrum principle involved.
- Ask: What information is not transparent?
- Ask: Who is accountable for the decision?
- Ask: What artifact, event, or conversation should be inspected?
- Ask: Is the issue about value, quality, risk, policy, or team effectiveness?
- Ask: Does the answer preserve self-management?
- Reject answers that rely on AI as the final authority.
- Reject answers that make the Scrum Master a task manager or approver.
- Prefer answers that enable inspection, adaptation, and responsible action.
Quick readiness scorecard
| Area | Green | Yellow | Red |
|---|---|---|---|
| Scrum basics | You can apply Scrum without notes | You know terms but hesitate in scenarios | You confuse accountabilities, events, or artifacts |
| AI limitations | You routinely question and verify AI output | You know risks but miss them in examples | You treat AI output as factual by default |
| Responsible use | You identify privacy, bias, and policy concerns | You identify obvious risks only | You ignore data and ethical implications |
| Scrum Master judgment | You coach, facilitate, and make issues transparent | You sometimes choose directive actions | You take over decisions from the team |
| Product value | You connect AI work to Product Goal and stakeholder outcomes | You focus on features more than outcomes | You assume “AI” equals value |
| Quality and DoD | You connect AI work to usable Increment standards | You mention testing but not broader quality | You treat generated work as done |
| Scenario reasoning | You can explain why an answer fits Scrum | You rely on memorized phrases | You cannot justify choices clearly |
Practical next step
Take a small set of PSM-AI-style practice scenarios and force yourself to write a one-sentence reason for each answer. If your reason does not mention Scrum accountability, transparency, empirical learning, value, risk, or responsible AI use, revisit the matching checklist section before continuing.