Skip to main content
Practice set 1 is free for every certification. Unlock sets 2-5 for $10.
Practice Exam

GH-700 — GitHub Certified: Agentic AI Business Professional

Test your knowledge with official exam-style questions

Questions25Passing700Exam time100 min

Questions and options are shuffled each attempt

Email me this practice set

Get a link sent to your inbox so you can pick this back up anytime.

▶GitHub Certified: Agentic AI Business Professional — Practice Set 1: All Questions & Explanations

Full question text, answer options, and explanations for this practice set — a spoiler-free alternative is the interactive quiz above for scored, shuffled practice.

  1. 1. A program manager helps roll out GitHub Copilot to 400 employees. Finance is worried that usage of premium AI models could push monthly spend well beyond the approved amount. Which action best addresses this concern?

    • A. Ask each employee to estimate and self-report their own AI usage at the end of the quarter
    • B. Set a budget with a spending limit that stops usage once reached, plus alerts at set thresholds(correct)
    • C. Remove Copilot access for everyone until the next fiscal year starts
    • D. Wait for the annual invoice and then review whether the spend was acceptable

    Explanation: Budget guardrails give finance proactive control and visibility over AI-related spend, which is exactly the risk described. For metered usage such as premium model usage, a budget can be set to stop usage once the threshold is reached, while alerts only notify recipients as spending approaches the budget; they do not cap spend on their own. Evaluating these guardrails is part of applying GitHub management best practices. Why the other options are wrong - A: Self-reporting is manual, inconsistent and after the fact, so it provides no real control. C: Removing access eliminates the business value of the rollout instead of managing the cost risk. D: An annual invoice review is purely reactive; overspend would already have happened.

  2. 2. A business analyst is drafting instructions for an agent that must call an internal billing service. To make the agent work quickly, she considers pasting the service API key directly into the instruction text. What is the recommended approach?

    • A. Paste the key into the instruction file, because the repository is private and only colleagues can ever read it
    • B. Put the key in a wiki page so every agent can look it up when needed
    • C. Type the key into each chat prompt so it is never stored in a file
    • D. Keep the credential in a protected secret and reference it by name rather than pasting its value anywhere(correct)

    Explanation: Credentials should live in a managed secrets store and be referenced indirectly. Anything written into prompts, instructions, issues or wikis can be read back, logged, copied or surfaced in model output, so it must never contain the secret value. Why the other options are wrong - A: Private repositories still have many readers, and instruction files are versioned, copied and forked, so the key would be exposed permanently in history. B: A wiki is a documentation space readable by anyone with repository access; it is not a secure store. C: Pasting it into prompts still sends the secret to the model and can leave it in chat history and logs.

  3. 3. A product manager must track quarterly feature delivery. Executives want a timeline across teams, while the delivery team wants drag-and-drop status columns for the same work items. What is the best way to set this up in GitHub?

    • A. Create one project from a template and add a roadmap view for the timeline and a board view for status(correct)
    • B. Create two separate projects and copy every item into both so each audience has its own separate list to maintain
    • C. Track the work in a wiki page and update a table by hand each week
    • D. Create one table view only and ask executives to read it from top to bottom

    Explanation: A single project can have several saved views (table, board, roadmap) over the same items, so each audience gets the layout it needs without duplicating data. Choosing a template and customizing views are core "plan and manage work" skills. Why the other options are wrong - B: Duplicating items across two projects creates conflicting sources of truth and double the maintenance. C: A hand-maintained wiki table is not linked to live issues and goes stale quickly. D: A single table view cannot give executives a timeline or the team a board.

  4. 4. A business analyst triages 30 incoming customer requests that were logged as GitHub issues. Leadership wants every request to have a clear owner and to be filterable by type of request. What should the analyst do?

    • A. Add all requests as comments inside one long issue titled "Requests"
    • B. Close the issues that do not have an obvious owner
    • C. Assign each issue to an owner and apply labels that describe the request type(correct)
    • D. Rename each issue with the requester's name and leave it unassigned

    Explanation: Assigning gives accountability and labels (tags) make the issues filterable by category; both are core issue-management actions. Linking related issues can then add traceability. Why the other options are wrong - A: One giant issue cannot be assigned, filtered or closed per request. B: Closing issues removes work from the backlog without resolving it. D: Names in titles do not provide ownership or categorization, and unassigned issues have no accountable person.

  5. 5. A technical project manager wants a pull request that implements a customer-requested change to be traceable to the original request and visible to stakeholders. Which TWO actions achieve this?

    • A. Link the pull request to the issue that captured the customer request(correct)
    • B. Rename the pull request branch after the customer and leave everything else unchanged
    • C. Copy the pull request number into an email to the stakeholders
    • D. Add the pull request to the project so it appears alongside the related work items(correct)
    • E. Paste the pull request text into a wiki page and delete the pull request

    Explanation: Linking a pull request to its issue creates a durable, navigable trail from request to change, and adding the pull request to the project makes it visible in the project views stakeholders already use. Both are part of tracking and capturing work. Why the other options are wrong - B: A branch name is not a link and does not connect the work to the request or the project. C: An email is not part of the work record, cannot be queried later and does not update any views. E: Deleting the pull request destroys the very record being tracked; a wiki copy does not link back to anything.

  6. 6. A team uses Copilot for two types of work: deep analysis of long, ambiguous requirement documents, and quick reformatting of meeting notes. Budget is limited. Which model strategy is most appropriate?

    • A. Always use the largest available model for every single task, because quality is the only factor that matters to the business
    • B. Always use the cheapest model for every task, because output differences are cosmetic
    • C. Choose the model per task: a stronger reasoning model for complex analysis, a lighter and faster one for routine formatting(correct)
    • D. Let each person pick randomly, because the choice of model does not affect output or cost

    Explanation: Different models have different strengths, speeds and cost profiles. Matching the model to the task balances quality, speed and spend, and results should still be reviewed. Why the other options are wrong - A: Using the largest model everywhere ignores cost and speed and is wasteful for simple tasks. B: The cheapest model may be inadequate for complex analysis; output quality and explainability can differ materially between models. D: Model choice does affect output quality, latency and cost, so random selection is poor governance.

  7. 7. A product manager has a long-running Copilot chat about the Q3 roadmap that she returns to every week, plus a dozen finished chats cluttering her list. How should she organize them?

    • A. Pin the roadmap session and archive the completed ones(correct)
    • B. Delete the roadmap chat after each use and recreate it from memory next week
    • C. Rename every finished chat to "Roadmap" so Copilot merges them automatically
    • D. Keep everything in a single never-ending session so nothing is ever separated

    Explanation: Sessions can be started, resumed, renamed, pinned and archived. Pinning keeps an active conversation within reach, and archiving declutters the list without losing history. Why the other options are wrong - B: Deleting and recreating loses the accumulated context the manager wants to resume. C: Renaming only changes a label; Copilot does not merge sessions based on names. D: One endless session mixes unrelated topics and makes context noisy and hard to navigate.

  8. 8. A program manager wants Copilot to draft release notes based on one specific issue and a requirements file, rather than guessing from general knowledge. What should she do?

    • A. Type a long description from memory and hope Copilot finds the right documents
    • B. Ask Copilot to read every issue in the organization and decide which are relevant
    • C. Rename the issue to include the words "release notes" and wait for Copilot to notice
    • D. Use @ mentions in the chat to attach the issue and the file as context for the request(correct)

    Explanation: Using @ mentions attaches precise content (tools, files and issues) to the request so the response is grounded in the intended sources. Why the other options are wrong - A: Describing from memory invites inaccuracies and does not give Copilot the real content. B: Pointing Copilot at everything adds noise and increases the chance of using the wrong material. C: Renaming an item does not attach it to the conversation.

  9. 9. A business analyst is handed a script written by a former colleague and needs a plain-language walkthrough of what it does. Which slash command fits best?

    • A. /fix, which proposes a correction for a problem in the selected content
    • B. /explain, which describes what the selected content does in plain language(correct)
    • C. /tests, which generates tests for the selected content
    • D. /clear, which clears the current conversation

    Explanation: /explain is the command that walks through what selected content does, which is exactly what the analyst needs. Why the other options are wrong - A: /fix is for resolving problems; nothing here is broken and the analyst wants understanding, not changes. C: /tests generates test cases, which does not explain the purpose of the script. D: /clear resets the conversation and would remove, not provide, an explanation.

  10. 10. A product manager asks Copilot, "Summarize the feedback," and gets differently shaped answers every time. She wants consistent output for a weekly leadership review. Which revised prompt is best?

    • A. Summarize the feedback better this time and be creative about how you present it
    • B. Summarize all feedback in as much detail as possible, however long it needs to be, so that nothing is left out
    • C. Summarize the 40 attached responses in a table of Theme, Mentions and Sentiment, under 150 words(correct)
    • D. Summarize the feedback. Make it good and make sure leadership likes it

    Explanation: A reliable prompt identifies the data to act on (the attached 40 responses) and specifies the response structure (a table with defined columns and a length limit), which makes outputs consistent and reviewable. Why the other options are wrong - A: "Better" and "creative" are vague and invite even more variation between runs. B: Unbounded length gives no structure and produces output that is hard to compare week to week. D: Subjective goals such as "make it good" give Copilot no data, structure or criteria to work with.

  11. 11. During a long Copilot session, an agent made a series of edits to a set of planning documents. The last three steps degraded the content, but the first part of the session was good. What is the best way to recover?

    • A. Undo each of the changes by hand from memory, one document at a time
    • B. Return to the checkpoint taken before the problem steps(correct)
    • C. Delete everything and restart the entire project from scratch
    • D. Ask Copilot to forget the last three steps and continue as if nothing happened

    Explanation: Checkpoints let you revert to an earlier known-good state of the session, discarding only the unwanted changes while preserving earlier progress. Why the other options are wrong - A: Manual undo is slow and error-prone, and it is easy to miss changes. C: Starting from scratch throws away the good early work unnecessarily. D: Asking the model to "forget" does not revert the actual file changes that were made.

  12. 12. A team builds a custom agent to draft customer communications. The business requires that it never commit to pricing or legal terms. Which design step best supports this requirement?

    • A. Define guardrails for the agent: scope, prohibited actions, permitted tools and human review before sending(correct)
    • B. Rely on each user remembering the rules whenever they use the agent
    • C. Give the agent every available tool so that it can always find a way to finish the task
    • D. Allow unrestricted web search so the agent can look up what competitors promise customers in similar messages

    Explanation: Guardrails encoded in the agent (scope, prohibited actions, limited tools and review requirements) enforce the business rule consistently rather than depending on individual users. Why the other options are wrong - B: Relying on memory is inconsistent and untraceable. C: Maximum tooling expands what the agent can do, which increases rather than reduces risk. D: Unrestricted search adds uncontrolled information sources and does nothing to prevent commitments.

  13. 13. Which of the following business tasks is the best candidate for an agentic workflow?

    • A. Negotiating a one-time strategic partnership where every decision is subjective and unprecedented
    • B. A task performed once every few years with no repeatable pattern or defined outcome
    • C. Triaging newly opened issues every week: repetitive, rule-based and measurable(correct)
    • D. A decision that legally requires a named executive to personally exercise judgment

    Explanation: Agentic workflows suit repetitive, multi-step processes with clear criteria and measurable outcomes, such as recurring triage. One-off, highly subjective or legally personal decisions are poor candidates. Why the other options are wrong - A: A one-time subjective negotiation has no repeatable pattern to automate. B: Rare, undefined tasks do not repay the design effort of a workflow. D: Decisions that must be made by a named person should stay with that person, with the agent at most assisting.

  14. 14. A program manager wants to automate the "weekly status report" process. What is the best first design step?

    • A. Write a single very large prompt that says "produce the weekly report" and iterate until it works
    • B. Choose the tools and models first and let the process shape itself afterwards
    • C. Skip analysis and automate only the final formatting step, ignoring everything before it
    • D. Break the process into discrete tasks such as gather updates, summarize per workstream, flag risks and format the report(correct)

    Explanation: Decomposing a use case into processes and tasks clarifies inputs, outputs and decision points for each step, which makes the workflow easier to build, test and supervise. Why the other options are wrong - A: One monolithic prompt hides the steps, making failures hard to diagnose and review. B: Selecting tools before understanding the tasks risks a poor fit. C: Automating only the end leaves the hard, time-consuming steps manual and delivers little value.

  15. 15. An agentic workflow assigns a priority to each new support issue. Which instruction gives the most reliable and auditable behavior?

    • A. Define explicit criteria, e.g. High if it mentions an outage or affects over 50 customers(correct)
    • B. Use your best judgment to decide how important each issue feels to you on the day, without written rules
    • C. Assign priorities at random so that the workload is spread evenly across everyone on the team
    • D. Assign High to anything written in capital letters

    Explanation: Explicit, business-defined decision criteria make the agent's choices consistent, explainable and easy to audit. Vague or arbitrary rules produce unpredictable outcomes. Why the other options are wrong - B: "Best judgment" is undefined and leads to inconsistent, unexplainable decisions. C: Random assignment ignores business impact entirely. D: Capital letters are an unreliable proxy for priority and could be gamed or misleading.

  16. 16. A team schedules an agentic workflow to compile a Monday summary of open issues. The security reviewer wants the lowest practical risk. Which configuration is most appropriate?

    • A. Run it on a weekly schedule with broad write access to every repository so that it never gets blocked
    • B. Run it weekly, grant only the read access needed for issues, and define a clear single summary output(correct)
    • C. Run it only when someone remembers to start it, with no defined output
    • D. Grant it administrator-level permissions so it can fix anything it finds

    Explanation: A scheduled run with least-privilege permissions and a well-defined output delivers the value while limiting what the workflow can change, which is the intent of configuring schedule, permissions and output deliberately. Why the other options are wrong - A: Broad write access violates least privilege and increases the damage a mistake could cause. C: Manual starts defeat the point of a schedule, and an undefined output cannot be reviewed. D: Administrator rights are far beyond what a read-and-summarize task needs.

  17. 17. A workflow drafts customer-facing summaries. The team wants the agent to improve its own draft before a person sees it. Which TWO design choices are appropriate for a self-critique loop?

    • A. Repeat revisions without limit until the agent decides it is perfect
    • B. Instruct the agent to check its draft against a written list of acceptance criteria before finalizing(correct)
    • C. Remove human review, since the agent already critiques itself
    • D. Use no criteria so the agent is free to critique however it likes
    • E. Limit the number of revision passes so the loop always ends(correct)

    Explanation: A self-critique loop works best when the agent compares output against explicit criteria and the number of passes is bounded. It improves quality but does not replace human review. Why the other options are wrong - A: An unbounded loop can run indefinitely, waste resources and never converge. C: Self-critique reduces errors but does not remove the need for human oversight of customer-facing content. D: Without criteria the critique is arbitrary and cannot be evaluated.

  18. 18. A team wants an automation to react every time a new issue labeled "customer-escalation" is opened. Which trigger approach is most suitable?

    • A. Run the automation every minute on a timer and check whether anything is new
    • B. Run it manually on the first of each month
    • C. Trigger it only when a pull request is merged
    • D. Use an event-based trigger that fires when a labeled issue is opened(correct)

    Explanation: An event-based trigger reacts exactly when the relevant event happens, so escalations are handled promptly without wasteful polling. Why the other options are wrong - A: Per-minute polling is wasteful and still delays or duplicates handling compared with an event. B: Monthly manual runs would leave escalations waiting for weeks. C: Merged pull requests are unrelated to new escalation issues.

  19. 19. An agentic workflow must reference live ticket data from the company's service desk system, which is not stored in the repository. How should the team give the workflow this context?

    • A. Paste a weekly export of the tickets into the instruction file by hand
    • B. Tell the agent to guess the ticket status based on its general training
    • C. Connect the workflow to an MCP server for the service desk data(correct)
    • D. Move the whole service desk system into the repository

    Explanation: Model Context Protocol (MCP) servers give agents standard, governed access to external systems and data, improving the context the workflow reasons over. Why the other options are wrong - A: Hand-pasted exports go stale immediately and bloat the instructions. B: Guessing guarantees inaccurate, unverifiable outputs. D: Relocating an entire system is disproportionate and unnecessary when a connection can supply the data.

  20. 20. An agentic workflow gathers order details, drafts a response and proposes refunds. Which step most clearly requires human judgment?

    • A. Approving refunds above the policy threshold(correct)
    • B. Gathering order numbers from the text of each incoming issue
    • C. Sorting the drafts alphabetically by customer
    • D. Formatting the response with the standard company header and footer

    Explanation: Financial commitments, customer-impacting and hard-to-reverse decisions are the critical decision points where a human should decide. Routine retrieval and formatting steps carry little risk. Why the other options are wrong - B: Collecting data is low risk and easily verified. C: Sorting has no business consequence. D: Formatting is cosmetic and easily corrected.

  21. 21. A team is updating a customer-visible policy document with an agent. They want to review and approve the agent's step-by-step approach up front, before any content is changed. Which mode fits?

    • A. Autopilot, where the agent completes the task independently without further input
    • B. Plan mode, where the agent proposes its approach for review first(correct)
    • C. None; mode selection only applies to professional software developers
    • D. A scheduled run with write permissions to every document

    Explanation: Plan mode has the agent lay out what it intends to do so people can approve or adjust it first. Autopilot is for tasks that can safely run end to end without stopping. Why the other options are wrong - A: Autopilot removes the up-front review the team wants. C: The study guide expects business professionals to choose how agents perform work, so this is within scope. D: A scheduled, write-enabled run increases risk and provides no pre-approval of the approach.

  22. 22. A business owner wants an agentic workflow to read data freely but to pause for a person whenever it tries to take an action that changes something outside the workflow, such as posting to customers. How should the escalation be designed?

    • A. Require approval for every single read operation, even for completely harmless lookups of data that nobody considers sensitive
    • B. Never require approval so the workflow is never slowed down
    • C. Escalate only at a fixed time of day regardless of what the agent does
    • D. Require approval for tools that change external state; read-only tools run uninterrupted(correct)

    Explanation: Escalation based on tool usage ties human review to risk: low-risk read tools run freely, and tools with side effects stop for approval. Why the other options are wrong - A: Approving every read adds friction and approval fatigue without reducing meaningful risk. B: No approval leaves high-impact actions uncontrolled. C: Time-based escalation is unrelated to what the agent is actually doing.

  23. 23. A workflow pauses and asks for approval to relabel 200 issues. The proposed action list unexpectedly includes closing some of them, which was never part of the requirement. Which TWO responses are appropriate?

    • A. Approve it immediately to stay on schedule and clean up afterwards if needed
    • B. Leave the request unanswered so the agent continues on its own assumptions
    • C. Reject or request a revision, with feedback that closing issues is out of scope(correct)
    • D. Turn off review for future runs so the interruption does not happen again
    • E. Compare the proposed actions to the requirements before deciding(correct)

    Explanation: Reviewers should check proposed actions against requirements and, where they deviate, reject or request a revision with clear feedback. Approval is a decision, not a formality. Why the other options are wrong - A: Approving first and cleaning up later risks irreversible changes to the issue backlog. B: Ignoring the request does not give the agent permission and leaves the workflow stalled or acting without oversight. D: Removing review in response to a legitimate catch removes the control that just worked.

  24. 24. While monitoring a running workflow, a manager sees it posting duplicate comments on customer-facing issues. What should she do first?

    • A. Stop the run right away, assess what was posted, then correct the instructions(correct)
    • B. Let it finish, because stopping a run is never allowed once it has started, and the damage is unavoidable
    • C. Delete the repository so the workflow can no longer run
    • D. Wait for the next scheduled run to see whether the problem repeats

    Explanation: Humans must be able to halt a running workflow when it misbehaves. Stopping contains the damage, then the team can review and fix the cause. Why the other options are wrong - B: Runs can be stopped; waiting lets the problem reach more customers. C: Deleting the repository is destructive and out of proportion to the issue. D: Waiting for another run delays containment and risks repeating the error.

  25. 25. A manager wants to confirm that last night's scheduled workflow ran and produced what the business expected. What is the best approach?

    • A. Assume success unless a customer complains
    • B. Ask the agent to say whether it believes it succeeded and accept its answer
    • C. Review the run history and outputs against the stated requirements(correct)
    • D. Check only whether the schedule setting is still switched on

    Explanation: Monitoring means looking at actual run records and outputs and verifying them against requirements, rather than relying on assumptions or the agent's self-assessment alone. Why the other options are wrong - A: Waiting for complaints is reactive and misses silent failures. B: An agent's self-report is not independent verification. D: A schedule being enabled does not show that a run succeeded or that outputs were correct.