PSM-AI — Scrum.org Professional Scrum Master - AI Essentials Quick Review
Quick Review for Scrum.org Professional Scrum Master - AI Essentials (PSM-AI): Scrum, AI use cases, risks, traps, and practice focus.
Quick Review purpose
This Quick Review is for candidates preparing for the Scrum.org Professional Scrum Master - AI Essentials (PSM-AI) exam, official exam code PSM-AI. It is designed as a fast, practical review before you move into topic drills, mock exams, original practice questions, and detailed explanations.
Use it to reinforce:
- How Scrum principles apply when AI is introduced into product delivery.
- Where AI can support Scrum Teams without replacing accountability.
- How to spot unsafe, low-transparency, or anti-Scrum uses of AI.
- Common exam traps around automation, decision-making, facilitation, quality, and empirical control.
This page is PM Mastery review support and is not affiliated with Scrum.org.
Core exam mindset
The most important idea: AI may assist people, but it does not replace Scrum accountabilities, empiricism, transparency, or professional judgment.
A strong PSM-AI answer usually favors:
| Prefer | Be careful with |
|---|---|
| Transparency, inspection, and adaptation | Hidden AI-generated work, unverified outputs, or opaque decisions |
| Human accountability | “The AI decided,” “the tool prioritized,” or “the model approved” |
| Evidence-based decisions | Confident AI claims without validation |
| Scrum values | AI uses that reduce openness, respect, courage, focus, or commitment |
| Incremental improvement | Big-bang AI adoption without learning loops |
| Clear Definition of Done | AI-created work that bypasses quality standards |
| Collaboration with stakeholders | Replacing conversation with generated summaries only |
| Responsible use of data | Sharing sensitive, proprietary, or personal data carelessly |
Scrum foundations to keep active
PSM-AI is still grounded in Scrum. AI questions often test whether you preserve Scrum while evaluating AI support.
| Scrum concept | What to remember | AI-related trap |
|---|---|---|
| Empiricism | Decisions are based on observation and evidence. | Treating AI output as fact without inspection. |
| Transparency | Work, progress, quality, and risks must be visible. | Using AI in ways the team or stakeholders cannot see or understand. |
| Inspection | Scrum Teams inspect artifacts and progress frequently. | Assuming generated plans, estimates, or summaries are correct. |
| Adaptation | Teams adjust when evidence shows a better path. | Locking into an AI-generated plan despite new learning. |
| Lean thinking | Reduce waste and focus on value. | Using AI to generate more documents, backlog items, or reports that do not help outcomes. |
| Scrum Values | Commitment, focus, openness, respect, courage. | Using AI to avoid difficult conversations, hide uncertainty, or blame a tool. |
Scrum accountabilities and AI boundaries
AI can support an accountability. It cannot hold one.
| Accountability | Owns / is accountable for | AI may help with | AI must not replace |
|---|---|---|---|
| Product Owner | Maximizing product value and managing the Product Backlog. | Drafting backlog item options, summarizing stakeholder feedback, exploring value hypotheses, identifying possible metrics. | Product decisions, ordering decisions, stakeholder accountability, Product Goal ownership. |
| Scrum Master | Establishing Scrum as defined in the Scrum Guide and helping the team and organization improve. | Preparing facilitation options, identifying coaching patterns, drafting retrospective prompts, summarizing impediment themes. | Coaching judgment, servant leadership, conflict navigation, organizational change leadership. |
| Developers | Creating a usable Increment each Sprint. | Code assistance, test ideas, design alternatives, documentation drafts, defect analysis. | Quality responsibility, engineering judgment, Definition of Done compliance, technical ownership. |
| Scrum Team | Delivering valuable, usable Increments and adapting through Scrum. | Learning faster, reducing repetitive work, generating options. | Collaboration, accountability, shared understanding, empirical decision-making. |
Fast rule
If an answer says AI supports, drafts, suggests, summarizes, or helps inspect, it may be reasonable.
If an answer says AI decides, owns, guarantees, replaces, approves, commits, or makes accountability unnecessary, be skeptical.
Artifacts, commitments, and AI transparency
Scrum artifacts exist to maximize transparency. AI should improve transparency, not create a false sense of certainty.
| Artifact | Commitment | AI can support | Common mistake |
|---|---|---|---|
| Product Backlog | Product Goal | Generate candidate backlog items, identify missing stakeholder perspectives, classify feedback themes. | Creating a large AI-generated backlog without validation or ordering by value. |
| Sprint Backlog | Sprint Goal | Help split work, identify dependencies, draft task options, highlight risks. | Letting AI produce a Sprint plan that the Developers do not understand or own. |
| Increment | Definition of Done | Suggest test cases, review code, check documentation gaps, identify quality risks. | Treating AI-generated code or analysis as “done” without meeting the Definition of Done. |
Scrum events: AI opportunities and traps
| Event | Purpose to protect | Helpful AI use | Trap answer to avoid |
|---|---|---|---|
| Sprint | Container for all Scrum events; creates cadence for inspection and adaptation. | Track patterns, summarize risks, support learning. | Changing the Sprint Goal because AI generated a better idea without Scrum Team inspection. |
| Sprint Planning | Establish why the Sprint is valuable, what can be done, and how it will be done. | Draft options for Sprint Goal wording, identify risks, suggest work breakdowns. | AI selects the Sprint Goal or forecasts capacity without Developers’ judgment. |
| Daily Scrum | Developers inspect progress toward the Sprint Goal and adapt the plan. | Surface blockers, summarize board changes, highlight risk signals. | Turning the Daily Scrum into a status report to an AI tool or manager. |
| Sprint Review | Inspect the Increment and adapt the Product Backlog. | Summarize stakeholder feedback, cluster themes, compare outcomes to goals. | Replacing stakeholder collaboration with AI-generated feedback analysis. |
| Sprint Retrospective | Improve effectiveness and quality. | Suggest retro formats, analyze team survey themes, draft improvement experiments. | Using AI-generated conclusions to judge individuals or avoid open discussion. |
AI essentials: high-yield concepts
| Concept | Exam-relevant meaning | Candidate trap |
|---|---|---|
| Generative AI | Produces text, code, images, summaries, or other outputs from patterns in data. | Assuming output is verified knowledge. |
| Large language model | Predicts likely language based on context and training. | Assuming it “understands” the product, user, or domain like a human expert. |
| Hallucination | Plausible but false or unsupported output. | Accepting confident answers without checking sources or evidence. |
| Bias | Output may reflect biased data, prompts, or assumptions. | Using AI recommendations without examining fairness or impact. |
| Prompt | Input or instruction provided to the AI system. | Writing vague prompts and blaming the tool for poor results. |
| Context | Information supplied to guide the output. | Providing too little context or exposing sensitive data unnecessarily. |
| Model limitation | AI may be outdated, incomplete, inconsistent, or non-deterministic. | Expecting the same reliability as a deterministic business rule. |
| Human-in-the-loop | People review, validate, and remain accountable. | Treating review as optional because the AI seems accurate. |
| Data privacy and confidentiality | Sensitive information must be handled responsibly. | Pasting proprietary, personal, or regulated data into tools without approval. |
| Explainability | Ability to understand why an output or recommendation was produced. | Acting on black-box recommendations that cannot be challenged. |
Responsible AI in Scrum settings
A good Scrum Master helps the team use AI in ways that improve outcomes while protecting transparency, ethics, quality, and trust.
Use this review checklist
Before using AI for Scrum work, ask:
- Purpose: What problem are we solving?
- Value: Does AI improve value delivery or just create more output?
- Transparency: Will the Scrum Team and stakeholders know AI was used where relevant?
- Data safety: Are we avoiding sensitive, proprietary, or personal data exposure?
- Validation: Who will inspect the output and against what evidence?
- Accountability: Which human accountability remains responsible?
- Bias and impact: Could the output disadvantage users, stakeholders, or team members?
- Quality: Does the result meet the Definition of Done or other quality standards?
- Learning: What will we inspect and adapt after trying it?
- Organizational policy: Are we following applicable internal rules for AI tools and data?
Decision path for AI use in Scrum
flowchart TD
A[Consider using AI] --> B{Does it support a clear Scrum or product outcome?}
B -- No --> X[Do not use it just to create more output]
B -- Yes --> C{Can data be shared safely?}
C -- No --> Y[Remove sensitive data or use approved alternatives]
C -- Yes --> D{Is a Scrum accountability still clearly responsible?}
D -- No --> Z[Redesign: AI cannot own accountability]
D -- Yes --> E{Can the output be inspected and validated?}
E -- No --> W[Use only for low-risk exploration or avoid]
E -- Yes --> F[Use AI as support]
F --> G[Inspect results with people]
G --> H[Adapt process, backlog, plan, or practice based on evidence]
High-yield AI use cases by Scrum area
| Area | Strong AI-assisted use | Weak or risky use |
|---|---|---|
| Product discovery | Generate interview questions, summarize discovery notes, identify assumptions. | Let AI define customer needs without real user evidence. |
| Product Backlog refinement | Draft item wording, suggest acceptance criteria, identify dependencies. | Creating many backlog items that the Product Owner has not validated or ordered. |
| Sprint Planning | Explore implementation approaches, risks, and test ideas. | AI commits the team to scope or decides what Developers can complete. |
| Engineering | Generate code snippets, test cases, documentation drafts, refactoring suggestions. | Merging AI-generated code without review, tests, security checks, or DoD compliance. |
| Quality | Suggest edge cases, regression risks, and test data patterns. | Assuming generated tests are complete. |
| Facilitation | Draft agendas, prompts, Liberating Structures ideas, retro exercises. | Replacing live facilitation and listening with scripted AI output. |
| Metrics | Summarize trends, detect anomalies, visualize flow data. | Optimizing for vanity metrics or using AI metrics to control individuals. |
| Stakeholder communication | Draft release notes, summarize feedback, prepare options. | Sending unreviewed AI-generated statements as official commitments. |
| Organizational change | Analyze impediment themes and generate experiment options. | Treating AI diagnosis as a substitute for observing the organization. |
Product Owner review points
AI can help the Product Owner see options, but value decisions remain human and empirical.
| Product Owner topic | Quick review |
|---|---|
| Product Goal | AI may help draft possible wording, but the Product Owner remains accountable for communicating and managing toward the Product Goal. |
| Product Backlog ordering | AI can provide inputs such as risk, dependency, user segment, or effort signals. It does not maximize value by itself. |
| Stakeholder feedback | AI can cluster feedback themes, but stakeholders still need collaboration and clarification. |
| Acceptance criteria | AI can draft examples, edge cases, and ambiguity checks. The team must inspect them. |
| Market or user insight | AI can accelerate research synthesis, but generated claims need evidence. |
| Value measurement | AI can suggest metrics, but the Scrum Team must inspect whether those metrics reflect real outcomes. |
Common Product Owner trap
An exam option may sound efficient: “Use AI to automatically order the Product Backlog based on predicted business value.”
Better thinking: AI may suggest ordering factors, but the Product Owner remains accountable and should use transparency, stakeholder input, evidence, and empirical learning.
Scrum Master review points
A Scrum Master should help the team adopt AI in a way that strengthens Scrum rather than bypassing it.
| Scrum Master concern | Good response |
|---|---|
| Team wants AI to write all backlog items | Encourage transparency, validation, and Product Owner accountability. |
| Developers use AI-generated code secretly | Promote openness, quality standards, and shared working agreements. |
| Organization wants AI status reports instead of Scrum events | Explain the purpose of Scrum events and protect inspection/adaptation. |
| Stakeholders trust AI forecasts too much | Reframe forecasts as uncertain and inspect empirical evidence. |
| Team fears AI will replace collaboration | Facilitate discussion, working agreements, learning, and safe experimentation. |
| AI outputs cause conflict | Bring the conversation back to evidence, values, and shared goals. |
Scrum Master decision rule
The Scrum Master does not need to be the “AI police,” but should coach the team to make AI use:
- Transparent.
- Ethical.
- Empirical.
- Aligned with Scrum.
- Safe for data and people.
- Useful for value delivery.
- Subject to inspection and adaptation.
Developers review points
Developers remain accountable for creating a usable Increment. AI-generated work is still work the Developers own.
| Developer activity | AI support | Required human responsibility |
|---|---|---|
| Coding | Generate examples, boilerplate, refactoring ideas. | Review, test, secure, integrate, and understand the code. |
| Testing | Suggest test cases and edge conditions. | Decide coverage, automate appropriately, verify results. |
| Architecture/design | Explore alternatives and tradeoffs. | Choose fit-for-purpose designs based on context. |
| Documentation | Draft user notes or technical summaries. | Ensure accuracy, clarity, and maintainability. |
| Debugging | Suggest root causes and fixes. | Validate with evidence and avoid speculative changes. |
| Security | Identify possible vulnerabilities. | Apply secure engineering practices and approved checks. |
Definition of Done trap
AI output does not lower the Definition of Done. If code, tests, documentation, or analysis are generated by AI, they still must meet the same quality expectations as any other work.
Prompting essentials for exam scenarios
You do not need to memorize a single prompt formula, but you should understand what makes AI assistance more useful and safer.
Strong prompts usually include
- Role or perspective: “Act as a facilitator,” “review as a tester,” “analyze as a Product Owner.”
- Context: product, users, goal, constraints, current Sprint situation.
- Task: what output is needed.
- Format: table, checklist, questions, risks, options.
- Boundaries: avoid assumptions, identify uncertainty, do not invent facts.
- Validation request: ask for risks, gaps, assumptions, or evidence needed.
Weak prompts often
- Ask for final decisions without context.
- Invite the model to invent facts.
- Include confidential data unnecessarily.
- Fail to ask for uncertainty or assumptions.
- Produce outputs too broad to inspect.
- Generate more content than the team can use.
Example review pattern
A useful AI prompt for Scrum work often asks for options, not final answers:
“Given this Sprint Goal and these known risks, suggest five facilitation questions the Scrum Master could use to help Developers inspect whether the Sprint plan still supports the Sprint Goal. List assumptions and risks separately.”
That kind of prompt supports human inspection. It does not transfer accountability to the tool.
Transparency and evidence
When AI is used, transparency matters at multiple levels.
| Transparency question | Why it matters |
|---|---|
| Was AI used? | The team may need to inspect reliability, ownership, and risks. |
| What input was used? | Output quality depends heavily on context and data. |
| What assumptions were made? | Hidden assumptions can create false certainty. |
| What was validated? | AI output is not evidence by itself. |
| What remains uncertain? | Uncertainty should guide inspection and adaptation. |
| Who is accountable? | Scrum accountabilities remain with people. |
Metrics, forecasting, and AI
AI can help analyze trends, but Scrum teams should avoid turning predictions into promises.
| Topic | Good use | Trap |
|---|---|---|
| Velocity or throughput trends | Identify patterns for discussion. | Treat AI forecast as a commitment. |
| Cycle time | Surface flow issues and bottlenecks. | Blame individuals based on automated analysis. |
| Quality metrics | Detect defect patterns or test gaps. | Declare quality acceptable without inspection. |
| Stakeholder sentiment | Summarize themes from feedback. | Treat sentiment summary as complete market truth. |
| Risk analysis | Generate possible risks to inspect. | Accept a risk ranking without team context. |
Forecasting rule
A forecast is a forecast. AI does not remove uncertainty. Scrum manages uncertainty through empiricism, not prediction alone.
Common PSM-AI traps
Trap 1: “AI improves Scrum by replacing events”
Wrong direction. Scrum events are opportunities for inspection, adaptation, alignment, and collaboration. AI may support preparation or summarization, but it should not replace the purpose of the event.
Trap 2: “AI-generated output is objective”
AI output can be biased, incomplete, outdated, or fabricated. Treat it as input for inspection, not as objective truth.
Trap 3: “The Scrum Master should make AI decisions for the team”
The Scrum Master coaches, facilitates, and helps remove impediments. The Scrum Master should not become a command-and-control gatekeeper for every AI use.
Trap 4: “The Product Owner can delegate value maximization to AI”
AI can provide analysis and suggestions. The Product Owner remains accountable for maximizing value and managing the Product Backlog.
Trap 5: “Developers can rely on AI to meet quality standards”
Developers remain accountable for the Increment. AI assistance does not bypass reviews, tests, security practices, or the Definition of Done.
Trap 6: “More generated backlog items means better refinement”
A larger backlog is not automatically better. Refinement should improve shared understanding, value, readiness, and transparency.
Trap 7: “AI reduces the need for stakeholder collaboration”
AI can summarize stakeholder input, but it cannot replace collaboration, feedback, negotiation, or shared understanding.
Trap 8: “If the AI tool is approved, all outputs are safe”
Tool approval does not guarantee every use is appropriate. Context, data, validation, and accountability still matter.
Scenario decision table
Use this table to answer scenario-style questions quickly.
| If the question says… | Strong answer direction |
|---|---|
| AI produced a recommendation the team does not understand. | Inspect assumptions, ask for explanation, validate with evidence, do not blindly follow it. |
| AI suggests changing the Sprint Goal mid-Sprint. | Discuss within the Scrum Team; preserve focus; adapt only through Scrum understanding, not tool authority. |
| AI generates code that appears to work. | Review, test, secure, and ensure it meets the Definition of Done. |
| Stakeholders want AI-generated progress reports instead of Sprint Reviews. | Explain the Sprint Review’s collaborative inspection purpose; AI summaries may support but not replace it. |
| Product Backlog is automatically reordered by AI. | Product Owner may use suggestions, but remains accountable for ordering and value decisions. |
| Team members hide AI use due to fear. | Encourage openness, working agreements, psychological safety, and responsible transparency. |
| AI summary conflicts with stakeholder comments. | Inspect the source data, talk to stakeholders, and resolve ambiguity with evidence. |
| AI creates many “ready” items. | Validate value, clarity, dependencies, and shared understanding before relying on them. |
| A manager wants AI metrics to compare individual Developers. | Avoid misuse of metrics; focus on team outcomes, flow, and improvement. |
| AI provides a confident answer with no evidence. | Treat it as unvalidated; seek supporting evidence or expert review. |
Scrum values with AI
| Scrum Value | AI-aligned behavior | Anti-pattern |
|---|---|---|
| Commitment | Use AI to help meet goals responsibly. | Commit to AI-generated scope without team ownership. |
| Focus | Use AI to reduce noise and clarify priorities. | Generate excessive options that distract from the Sprint Goal. |
| Openness | Be transparent about AI use, risks, and uncertainty. | Hide AI-generated work or limitations. |
| Respect | Use AI to support people, not judge or replace them. | Use AI analysis to shame individuals. |
| Courage | Challenge AI outputs and unsafe practices. | Accept tool recommendations because questioning them is uncomfortable. |
Quality and Definition of Done review
AI can accelerate production, but quality still depends on disciplined inspection.
| Quality question | What to look for |
|---|---|
| Is the work usable? | A usable Increment must meet the Definition of Done. |
| Is the output understood? | Developers should understand and be able to maintain what they deliver. |
| Has it been tested? | AI-generated tests may help, but test adequacy still needs judgment. |
| Are security and compliance concerns considered? | Do not assume AI-generated code or documents are safe. |
| Is documentation accurate? | AI may produce plausible but incorrect documentation. |
| Are assumptions visible? | Hidden assumptions reduce transparency. |
AI adoption as an empirical experiment
A Scrum-consistent approach to AI adoption is incremental.
| Step | Practical action |
|---|---|
| Identify a problem | Example: slow refinement, weak test coverage, repetitive documentation. |
| Select a small experiment | Try AI for a narrow, low-risk use case. |
| Define expected benefit | Faster preparation, better test ideas, clearer stakeholder summaries. |
| Establish safeguards | Data rules, review expectations, transparency agreements. |
| Inspect results | Did value, quality, or flow improve? What risks appeared? |
| Adapt | Expand, change, or stop the AI use based on evidence. |
Avoid answers that introduce AI as a broad mandate without inspection, learning, or team involvement.
Quick comparison: good vs. poor AI integration
| Good integration | Poor integration |
|---|---|
| AI is used to generate options. | AI is treated as the decision-maker. |
| Outputs are reviewed by accountable people. | Outputs are accepted because they sound confident. |
| Use is transparent to the team. | Use is hidden or not discussed. |
| Data handling is deliberate. | Sensitive data is pasted into tools casually. |
| Quality standards remain unchanged. | AI work gets a quality shortcut. |
| Scrum events remain meaningful. | Events are replaced with automated reports. |
| Learning is empirical. | Adoption is based on hype or fear. |
| Stakeholders remain engaged. | Stakeholder collaboration is replaced by summaries. |
Rapid review questions to ask yourself
Before moving to practice questions, make sure you can answer these without hesitation:
- Why does AI not replace Scrum accountabilities?
- How can AI support Product Backlog refinement without creating waste?
- What should Developers do before accepting AI-generated code?
- How can a Scrum Master encourage responsible AI use without becoming command-and-control?
- Why is transparency important when AI is used?
- What is the risk of using AI-generated forecasts as commitments?
- How can AI help a Sprint Retrospective without replacing team conversation?
- What makes an AI output unsafe or unreliable?
- How does the Definition of Done apply to AI-assisted work?
- What should a Product Owner do with AI-generated ordering suggestions?
Practice focus for the question bank
When you move into PM Mastery practice, prioritize original practice questions that force you to choose between “efficient but anti-Scrum” and “empirical, accountable, transparent” answers.
High-value topic drills should cover:
- Scrum accountabilities and AI boundaries.
- AI use in Scrum events.
- Product Backlog and Sprint Backlog scenarios.
- Definition of Done and AI-generated work.
- Transparency, inspection, and adaptation.
- Data privacy, confidentiality, and responsible use.
- Prompt quality and validation.
- AI hallucination, bias, and uncertainty.
- Metrics, forecasts, and stakeholder communication.
- Scrum Master coaching responses to AI adoption problems.
Final quick-review rule set
Use these rules when answering PSM-AI scenario questions:
- AI assists; people remain accountable.
- Generated output is not automatically true, valuable, safe, or done.
- Scrum events should not be replaced by automated summaries.
- The Product Owner owns value decisions.
- Developers own quality and the Increment.
- The Scrum Master coaches responsible, transparent, empirical use.
- Do not expose sensitive data casually.
- Use AI to improve inspection and adaptation, not to avoid them.
- Prefer small experiments over broad mandates.
- When uncertain, inspect evidence with the Scrum Team and adapt.
Next step: use topic drills and mock exams with detailed explanations to practice applying these rules under exam-style wording, especially scenarios where AI appears efficient but weakens Scrum accountability, transparency, or empirical control.
Continue in PM Mastery
Use this Quick Review as a final concept map, then move into PM Mastery for focused topic drills, mixed practice sets, timed mock exams, and detailed explanations. The practice questions are original PM Mastery practice items; they are not official Scrum.org questions, copied live-exam content, or exam dumps.