22 Products, Models & Configuration Practice Questions & Answers

1. 1. A marketing analyst wants to work through several pricing scenarios for a proposal over an afternoon, asking follow-up questions and revising the reasoning as she goes. Which kind of Claude surface fits this work?

   • A.A hosted agent service that runs unattended on a schedule
   • B.A programmatic interface called by a script so the scenarios run automatically
   • C.An agentic coding environment that edits files and runs commands in a repository
   • D.A conversational assistant interface, where a person drives each turn and iterates on the answer✓ Answer

   The decisive question is whether a human is in the loop on every turn. This is one person thinking interactively, with no filesystem access, no volume, and no schedule involved, so the conversational assistant is the right category; automating it would add engineering overhead without removing any repeated manual work.

   Source: Anthropic Claude certification (Associate – Foundations), Domain 3 — Product and Model Selection: select appropriate Claude product featuresReport a problem with this question

2. 2. A developer needs to rename a widely used function across roughly 40 source files, run the test suite, and fix whatever breaks. Which surface category is built for this?

   • A.A hosted agent service that the provider operates on the team's behalf
   • B.A conversational assistant, pasting each file in and copying the edits back by hand
   • C.A programmatic interface embedded in the company's customer-facing web application
   • D.An agentic coding environment that reads and edits files, runs commands, and iterates on real test output✓ Answer

   The work requires direct access to a filesystem and a command line plus a feedback loop against actual test results, which is exactly what the agentic coding category provides. Copy-pasting files into a chat reproduces the same task by hand and loses the ability to verify the change by running it.

   Source: Anthropic Claude certification (Associate – Foundations), Domain 4 — Workflow Integration and Solution Design: integrate Claude into existing workflowsReport a problem with this question

3. 3. A support operations team must categorize about 80,000 incoming tickets every night and write the resulting labels back into its ticketing system. Which surface category fits?

   • A.A programmatic interface inside an automated pipeline that reads tickets and writes results back✓ Answer
   • B.A conversational assistant shared with the whole team so each person handles a portion of the tickets
   • C.An agentic coding environment opened on the ticketing system's source repository
   • D.A conversational assistant, with a staff member pasting batches of tickets in each night

   Three signals point to the programmatic category at once: high volume, a repeatable rule applied the same way every time, and output that feeds another system rather than a reader. Doing the same job by hand in a chat is the classic under-powering error.

   Source: Anthropic Claude certification (Associate – Foundations), Domain 4 — Workflow Integration and Solution Design: integrate Claude to augment or redesign existing workflowsReport a problem with this question

4. 4. A small team wants an agent to carry out multi-step tasks on its behalf, but has nobody to operate servers, sandboxes, or the agent's execution loop. Which surface category matches that constraint?

   • A.A programmatic interface, with the team building and hosting the agent loop themselves
   • B.A hosted agent service in which the provider runs the agent loop and the execution environment✓ Answer
   • C.A conversational assistant used separately by each team member
   • D.An agentic coding environment running on a developer's laptop

   The defining difference between the programmatic category and a hosted agent service is who operates the loop and the execution environment. When the customer cannot run that infrastructure, the hosted option is the category that removes exactly that burden.

   Source: Anthropic Claude certification (Associate – Foundations), Domain 3 — Product and Model Selection; Domain 4 — solution design and iterationReport a problem with this question

5. 5. Which question most directly decides whether a task belongs in an agentic coding environment rather than a conversational assistant?

   • A.Whether the work must read and modify files and run commands, then iterate on the results it gets back✓ Answer
   • B.Whether the answer should be delivered as a formatted document
   • C.Whether the person making the request is technical
   • D.Whether the request is long or short

   Surface categories are separated by the level of system access and the interaction model, not by topic or by who is asking. A non-engineer discussing code still belongs in a conversational assistant; an engineer who needs files changed and commands run needs the coding environment.

   Source: Anthropic Claude certification (Associate – Foundations), Domain 3 — Product and Model Selection: align selection with task requirementsReport a problem with this question

6. 6. A procurement manager will compare six vendor contracts once, ahead of a single meeting next week. An engineer offers to build an automated pipeline that ingests contracts and produces comparisons. What is the better call?

   • A.Do the comparison interactively in a conversational assistant with the six contracts attached✓ Answer
   • B.Commission a hosted agent service to monitor the contract folder continuously
   • C.Build the pipeline, because automation is always more reliable than manual work
   • D.Skip Claude entirely and have a junior analyst read all six contracts

   Automation earns its build cost through repetition; a task performed once by one person never repays an integration. Attaching the source documents to a single conversation grounds the analysis in the right material without creating anything to maintain.

   Source: Anthropic Claude certification (Associate – Foundations), Domain 4 — Workflow Integration and Solution Design: analyze requirements and use casesReport a problem with this question

7. 7. An operations associate spends 30 minutes every morning pasting the same 400-row export into a chat, applying the same instructions, and copying the results into a spreadsheet. Which change is most appropriate?

   • A.Ask for a more capable model so the daily pasting finishes faster
   • B.Keep the manual process but shorten the instructions
   • C.Split the export among several teammates so each person pastes fewer rows
   • D.Move the recurring job to a programmatic interface so the pipeline runs automatically each morning✓ Answer

   A fixed procedure repeated on a fixed schedule over a fixed data shape is the textbook profile for automation, and the manual step adds no judgment. Choosing a different model or spreading the same manual labor across more people leaves the repeated handwork in place.

   Source: Anthropic Claude certification (Associate – Foundations), Domain 7 — Troubleshooting and Optimization: optimize workflows for efficiency and effectivenessReport a problem with this question

8. 8. A finance team needs a summary report built from the previous day's transactions and waiting in their inbox by 7:00 a.m. Nobody watches it run. Which characterization should drive the design?

   • A.It requires an agentic coding environment because it touches financial data
   • B.It requires a person in the loop to approve each step while it runs
   • C.It is latency-sensitive, so responses must stream back token by token
   • D.It is an asynchronous batch job, so throughput and cost matter more than response time✓ Answer

   Latency only matters when someone is waiting on the answer. With a deadline hours away and no observer, the job is asynchronous, so the design should optimize for reliable completion, throughput, and cost rather than immediate response.

   Source: Anthropic Claude certification (Associate – Foundations), Domain 4 — Workflow Integration and Solution Design: support solution design and iterationReport a problem with this question

9. 9. A software company wants an in-product assistant that answers customer questions inside its application, with replies appearing while the customer waits. Which surface category fits?

   • A.A conversational assistant interface, with support staff relaying answers to customers
   • B.An agentic coding environment left open on the product's codebase
   • C.A scheduled batch job that answers all the questions collected during the day
   • D.A programmatic interface embedded in the application, returning responses to end users in real time✓ Answer

   When the model has to live inside another piece of software and serve end users, the surface must be programmatic — a person cannot sit between the customer and the answer at product scale. Here latency genuinely matters because a user is waiting.

   Source: Anthropic Claude certification (Associate – Foundations), Domain 4 — Workflow Integration and Solution Design; escalation to Claude Developers for API-based buildsReport a problem with this question

10. 10. Whenever a new applicant record appears in an internal system, a short standardized summary must be produced and attached to that record — dozens of times a day, with nobody waiting on any single one. Which approach fits?

    • A.A person opening a conversational assistant each time a record appears
    • B.An agentic coding environment triggered by hand for each record
    • C.An event-triggered automated integration through a programmatic interface✓ Answer
    • D.One long-running chat into which records are pasted as they arrive

    Work that starts in response to a system event, runs unattended, and writes back into the same system is by definition an integration rather than a conversation. A single long chat also degrades as unrelated records accumulate in the same context.

    Source: Anthropic Claude certification (Associate – Foundations), Domain 4 — Workflow Integration and Solution Design: integrate Claude into existing workflowsReport a problem with this question

11. 11. A legal team re-pastes the same 60-page policy manual and the same standing instructions into nearly every new conversation. What is the right fix?

    • A.Keep one conversation open indefinitely so the manual stays in context
    • B.Ask for a more capable model so it remembers the manual on its own
    • C.Have engineering build a custom application around the manual
    • D.Attach the manual and the standing instructions to a persistent shared workspace so every new conversation starts grounded in them✓ Answer

    Repeatedly supplying the same reference material is the signal that the context is durable and should be persisted rather than re-pasted. Persisting it also makes every future conversation, by anyone on the team, start from the same grounded material.

    Source: Anthropic Claude certification (Associate – Foundations), Domain 5 — Configuration and Knowledge Management: configure projects with instructions and knowledge sourcesReport a problem with this question

12. 12. An analyst needs a one-time read of a single 20-page report that nobody else will ever use again. How should the material be supplied?

    • A.Create a persistent shared workspace and load the report as team knowledge
    • B.Ask the model to recall the report from what it already knows
    • C.Build a live connection to the document repository so the report stays current
    • D.Attach the report to that one conversation and work from it there✓ Answer

    Grounding should match the lifespan of the need: a document used once belongs in the conversation that uses it. Standing up a shared workspace or a live connection for a single throwaway task is over-engineering that leaves stale material behind.

    Source: Anthropic Claude certification (Associate – Foundations), Domain 5 — Configuration and Knowledge Management: manage uploaded knowledgeReport a problem with this question

13. 13. A team is considering connecting Claude to an internal system so it can read and act on live records. Which statement best captures the trade-off?

    • A.The connection removes the need to review outputs, since the source data is authoritative
    • B.It provides current data instead of a stale snapshot, but obligates the team to scope permissions narrowly and review what actions it is allowed to take✓ Answer
    • C.It affects only speed; permissions and review responsibilities are unchanged
    • D.It converts an interactive task into a coding task

    Connecting to an external system changes both what is possible and what is owed: it grants live reads and, often, the ability to act, which is exactly why access must be scoped to the minimum needed and consequential actions must stay reviewable.

    Source: Anthropic Claude certification (Associate – Foundations), Domain 5 — manage knowledge sources and connectors; Domain 6 — governance and responsible useReport a problem with this question

14. 14. Reference pricing changes several times a week, and answers must always reflect the current figures. Which approach fits?

    • A.Have each person re-paste the price list at the start of every conversation
    • B.Connect to the source system so the current values are read at the time of the request✓ Answer
    • C.Upload the price list once, since a single upload is simpler to manage
    • D.Instruct the model to estimate a price whenever it is unsure

    A one-time upload is a snapshot frozen at the moment it was made, so it goes stale as soon as the source changes; a live connection reads the source at request time. The rule is to choose by whether the underlying material changes.

    Source: Anthropic Claude certification (Associate – Foundations), Domain 5 — Configuration and Knowledge Management: maintain and update knowledge sources over timeReport a problem with this question

15. 15. The same nine-step quarterly review analysis is re-explained from scratch each quarter by whoever is on duty, and the results vary noticeably. What is the most durable fix?

    • A.Have each person write a longer prompt
    • B.Ask engineering to build a bespoke internal application for the analysis
    • C.Capture the workflow once as a reusable, shared set of instructions the team invokes each quarter✓ Answer
    • D.Use the most capable model available so the missing steps are inferred

    Variation comes from the procedure being re-invented each time, so the remedy is to make the procedure itself an artifact the team reuses. A stronger model cannot infer steps that were never written down, and custom software is unnecessary overhead for a documented routine.

    Source: Anthropic Claude certification (Associate – Foundations), Domain 5 — create effective system-level instructions; Domain 7 — optimize workflowsReport a problem with this question

16. 16. Six people on a team each keep private conversations with their own version of the department's tone guide, and the output they produce is inconsistent. What is the best fix?

    • A.Ask everyone to be more careful when writing prompts
    • B.Have one person do all of the work so it stays consistent
    • C.Have each person start a fresh conversation for every task
    • D.Move the shared context and instructions into a workspace the team can share, so everyone works from the same material✓ Answer

    Inconsistency traced to six divergent private copies of the same reference is a collaboration problem, and the structural fix is one shared source of context and instructions rather than individual discipline. Funnelling the work through one person removes the inconsistency by creating a bottleneck.

    Source: Anthropic Claude certification (Associate – Foundations), Domain 5 — Configuration and Knowledge Management; Domain 4 — workflow integrationReport a problem with this question

17. 17. A department head asks a certified associate to set up an autonomous system that receives customer refund requests, decides them against policy, and issues refunds without human review. What is the appropriate response?

    • A.Set it up in a conversational assistant and paste refund requests in as they arrive
    • B.Promise the capability, since an assistant can already read policy documents
    • C.Escalate to the technical implementation team, since this is an agentic system with real-world side effects that requires developer-level design✓ Answer
    • D.Decline the project outright as impossible

    An unattended system that takes consequential real-world actions is an engineered agentic build, not a task an associate performs in an assistant interface; recognizing that boundary and escalating is the expected behavior. The work is feasible — it simply belongs to the people who design such systems.

    Source: Anthropic Claude certification (Associate – Foundations), Intended Audience and Scope: recognize limitations and escalate complex or technical implementationsReport a problem with this question

18. 18. A team asks engineering to build a custom internal tool because they want consistent, document-grounded answers about their own employee handbook. What should be recommended first?

    • A.Configure a shared workspace with the handbook and standing instructions, since that meets the need without custom development✓ Answer
    • B.Have each member paste the handbook into their own conversations
    • C.Build the custom tool, because internal documents always require engineering
    • D.Tell the team the need cannot be met without a full integration

    Escalating to development is correct only when the need genuinely exceeds what configuration can do; here a shared, document-grounded workspace with standing instructions already delivers consistency. Over-escalation wastes engineering capacity exactly as over-engineering wastes it.

    Source: Anthropic Claude certification (Associate – Foundations), Domain 5 — Configuration and Knowledge Management; Intended Audience: escalation boundaryReport a problem with this question

19. 19. The output of a repeated task must be consumed directly by another program, with no person reading it in between. What does that imply about the surface choice?

    • A.It implies an agentic coding environment, because another program is involved
    • B.It makes no difference; where the output goes is irrelevant to surface choice
    • C.It implies an interactive assistant, since a person can copy the results into the other program
    • D.It points to a programmatic interface returning machine-readable output the receiving system can parse✓ Answer

    Who or what consumes the output is one of the core selection questions: output destined for a person can be prose in a conversation, while output destined for software must be structured and delivered programmatically so it can be parsed reliably.

    Source: Anthropic Claude certification (Associate – Foundations), Domain 2 — select appropriate output formats; Domain 4 — workflow integrationReport a problem with this question

20. 20. A team has an idea for a repeated document-triage task but has never tested whether the output quality is acceptable. What is the sensible sequence?

    • A.Skip testing and hand the task to an unattended agent service straight away
    • B.Prototype interactively on real examples first, then move to an automated interface once the approach is proven and the volume justifies it✓ Answer
    • C.Decide based on whichever surface the team already has access to
    • D.Build the automated pipeline first and evaluate quality once it is running in production

    Interactive use is the cheapest place to discover that an approach does not work, because a human sees every result. Automating an unvalidated approach multiplies an unknown error rate across every run before anyone has confirmed the output is usable.

    Source: Anthropic Claude certification (Associate – Foundations), Domain 4 — support solution design, development, and iterationReport a problem with this question

21. 21. A data scientist wants Claude to load a large local dataset, run analysis scripts, look at the errors and plots they produce, and correct its own code until the analysis completes. Which surface category is required?

    • A.A scheduled batch job that returns the finished analysis in the morning
    • B.A programmatic interface embedded in a reporting dashboard
    • C.An environment where the model can execute code, read the actual results, and iterate on the files✓ Answer
    • D.A conversational assistant with the dataset described to it in words

    The requirement is a closed loop of execute, observe, and correct against real data, which only a surface with code execution and file access can provide. Describing the data in words gives the model no way to see what actually happened when the script ran.

    Source: Anthropic Claude certification (Associate – Foundations), Section 7 — How to Prepare: code execution; Domain 3 — align selection with task requirementsReport a problem with this question

22. 22. A team is moving a task from a supervised conversation to an unattended automated run. Which consequence should most shape the design?

    • A.Nobody reviews each output any more, so the design needs logging, error handling, and sampled human review✓ Answer
    • B.Unattended runs require a more capable model by definition
    • C.Unattended runs remove the need to define what a good output looks like
    • D.Nothing changes, since the same task produces the same results either way

    In an interactive session the human is the error-detection mechanism; removing the human removes that safeguard, so it has to be replaced by logging, explicit failure handling, and periodic sampling. The model tier is a separate decision driven by task difficulty, not by supervision.

    Source: Anthropic Claude certification (Associate – Foundations), Domain 2 — determine when human review or additional verification is requiredReport a problem with this question

Practice questions based on the official Claude Certified Associate – Foundations (CCAO-F) exam guide and Anthropic's public documentation. This is an independent study tool, not affiliated with or endorsed by Anthropic, and does not grant certification. The real exam is 60 questions, 120 minutes, passing at a scaled 720/1000, delivered via Pearson VUE ($99). Official certification page →