Skip to content
The AI Agent Factory

Appendix C. Preparing for PCAO-F and CCAO-F

Approved by
Panaversity ·
25 min read

In everyday life. Before a driving test, you learn the car, watch someone drive, then drive the test route yourself. This page is that route for two certifications, PCAO-F and CCAO-F.

The two names differ by one letter. P is for Panaversity, and C is for Claude.

PCAO-F1CCAO-F2
Full namePanaversity Certified Associate: FoundationsClaude Certified Associate: Foundations
Who gives itPanaversityAnthropic, the company that makes Claude
When you take itFirstAfter PCAO-F, and only if you want to
ExamsTwo: Exam 1 to prepare in steps, then Exam 2One
What it testsThis book, from Claude and ChatGPT 2.0 to Part IIThe 21 skills in Anthropic's exam guide
Kind of questionA short work scenario with choicesThe same

The Roadmaps show where both sit on Panaversity's path. Anthropic's exam guide, version 1.1 lists the 21 skills, in six groups called domains.2 This book teaches all 21, and the map on this page shows where.

The PCAO-F exams

PCAO-F is Panaversity's certification. You earn it by passing two exams, one after the other.1 The Basic Concepts Exam was created only to help candidates prepare for the exam in steps. You can also pass PCAO-F by taking the full exam directly.3

Exam 1: Basic Concepts Exam

  • What it tests. Claude and ChatGPT 2.0, The Five Horizons, the Preface and Part 0.
  • Questions. 35 multiple-choice questions, each about a work scenario.
  • Time. 60 minutes.
  • Format. You take it on a computer.
  • When to take Exam 2. Take Exam 2 when you have mastered the complete syllabus for the certification.

Exam 2: Full Certification Exam

  • What it tests. Everything in Exam 1, and also Part I and Part II.
  • Questions. 70 multiple-choice questions, each about a work scenario.
  • Time. 120 minutes.
  • Format. You take it on a computer.

Exam 2 is the same size as the CCAO-F exam: 70 questions in 120 minutes.2 So is the PCAO-F sample exam, which you can sit as practice for both.

When you have passed Exam 2, you hold PCAO-F, and you have studied everything CCAO-F tests.

The CCAO-F exam

CCAO-F is Anthropic's certification. You earn it by passing one exam. It is for professionals who use Claude as a main productivity tool in a business role.2

  • Format. 70 work-scenario questions in 120 minutes, about 1 minute 43 seconds each. 10 of them are not scored, and you cannot tell which.
  • Pass mark. 720, on a scale of 100 to 1,000. Your score report also shows your percentage in each domain.
  • Fee. $99 for each attempt.
  • Retakes. After 14 days, then 30, then 90, and up to four attempts in 12 months.
  • Renewal. The credential lasts 12 months, and a free online check renews it. If the exam's content changes a lot, Anthropic may ask for the full exam again.

Who can take it. Only people who work at an organization in Anthropic's Claude Partner Network.4,5 To register:

  1. Sign up on the Anthropic Partner Academy with your company email address. A personal email address does not work.6
  2. Book your exam with Pearson. You can take it online or at a test center.2

If your organization is not in the network, there is one other way. Panaversity's FDE Internship Program helps its interns register, as the Roadmaps explain.

How the questions think

The rest of this page helps with PCAO-F and CCAO-F alike, because their exams ask the same kind of question. Anthropic's guide ends with three sample questions, each a short work scene with four choices. Read the three in the guide first. One choice is the careful step, sized to the problem. The others skip a check, trust the wrong thing or go too far.

Five patterns sit behind the right answers.

  • Verify independently (the first sample: a client brief Claude drafted from two research files the manager has not opened). Check each figure and the quote against the files or another trusted source, and flag what you cannot confirm. Asking Claude to confirm its own work is not a check (6.2).
  • Find the layer that changed, and fix that (the second sample: a proposals project that drifted, though its instructions and skill did not change). A retired price list points to stale knowledge, and another client's words to memory from other work (11.2).
  • Protect data before it is shared (the third sample: a staff file with names, addresses and comments, where policy allows AI analysis only on de-identified data). Take out or code everything that identifies a person, free text included, before the upload, and set memory to match. An incognito chat, a project only HR can open, or a request to Claude to clean the file after upload all share it first (10.2).
  • Size the action to the problem (the first and second samples). Deleting every figure throws away what was right (6.2), and a stronger model fixes none of the real causes (11.2).
  • A person stays accountable (the first sample). Claude drafts, and the manager checks the brief before it goes to the client (10.4).

Last, each question says how many answers to choose. Read that number before you choose, as in this book's "Choose two" questions.

Your route to CCAO-F

Follow these five steps in order.

  1. Learn the apps. Read The New Way of Working, the Tour of the Apps and Build Something, and make each artifact yourself. You get the hands-on use the questions expect: projects, artifacts, research and connectors.
  2. Learn the four Ds. Read Chapter 0.3, AI Fluency. The guide says Delegation, Description, Discernment and Diligence underpin every domain.
  3. Watch the work done. Read Case 1 and Case 4, and do Case 4's lab. You see AI work checked by a second reviewer and a person, and a monthly job built from a project, a skill and code.
  4. Learn and practice each skill. Read Chapter 2, then Chapters 5 to 11 in Part II, and do each chapter's lab. Together they teach the 21 skills in the map below. Each lab's exam notes show where the guide, written for earlier apps, expects a different answer from today's.
  5. Check yourself. Do the final rounds and the PCAO-F sample exam, as Check that you are ready explains.

The map: the guide's 21 skills

Each skill has the guide's name, one sentence on what it asks of you, the lesson that teaches it best, with a worked example, and a lab task that practices it.

Domain 1: Task Definition and Prompting (15%)

  • Prompt Construction. Turn a business goal into a prompt with its role, context, task and format, plus what Claude cannot know, so the first answer fits. Read 5.1, then do Lab 5, Task 4.
  • Task Decomposition and Sequencing. Split work that is too big for one request into stages, with a check between them, and shape each prompt to its kind of task. Read 5.3 and 5.2, then do Lab 5, Task 8.
  • Prompt Diagnosis and Iteration. Trace a weak answer to its gap: missing context, an unclear instruction, a missing limit or the wrong strategy. Then iterate, restart or change approach. Read 5.2, then do Lab 5, Task 7.

Domain 2: Output Evaluation and Quality Control (20%)

  • Output Evaluation and Fitness for Purpose. Judge an output against its requirements, sources and professional standards, then call it ready to use, needing revision, or needing human correction. Read 6.2, then do Lab 6, Task 9.
  • Hallucination Detection and Verification. Find invented citations, made-up details, contradictions and bias, and check them against the sources yourself before anyone uses the output. Read 6.2, then do Lab 6, Task 5.
  • Review Depth and Escalation. Review more deeply when an error costs more or the work leaves the company, and decide what a person must check, fix or escalate. Read 6.1 and 10.4, then do Lab 6, Task 1.
  • Output Adaptation and Format Selection. Shape an output for its reader and pick its form: chat reply, artifact, structured data, exact figures from code, or a document a skill makes. Read 5.1 and 6.3, then do Lab 5, Task 5.

Domain 3: Capability, Model, and Context Selection (15%)

  • Capability Selection. Match a task to the right features, such as projects, artifacts, research, code execution, skills, connectors and memory, alone or together. Read 2.4, then do Lab 2, Task 3.
  • Model and Effort Selection. Choose the model and effort level that fit the task's stakes, speed and cost, know when the cheaper choice is right, and defend your choice. Read 2.4, then do Lab 2, Task 5.
  • Context and Memory Management. In long work, decide when to restart, summarize, save to a project or rely on memory. Review, edit and delete memories, and use incognito chats. Read 8.2, then do Lab 8, Task 2.

Domain 4: Workflow Design and Integration (18%)

  • End-to-End Deliverable Workflows. Carry a research, analysis or planning job from its requirements, through cited research and analysis in code, to a plan someone can act on. Read 9.7, then do Lab 9, Task 13.
  • Delegation and Workflow Redesign. Assign each workflow step to Claude, a person, or Claude with a person's review, by stakes, reversibility and judgment, and spot giving Claude too much. Read 9.1, then do Lab 9, Task 2.
  • Repeatable Workflow Design. Turn a task you repeat by hand into a workflow that gives the same result each run and user, and that a colleague can own. Read 9.1 and 11.3, then do Lab 11, Task 11.
  • Capability Communication. Tell people new to AI what Claude can and cannot do for their case, and what a person still checks, without overselling or underselling it. Read 9.4, then do Lab 9, Task 10.

Domain 5: Configuration and Knowledge Management (15%)

  • Project Configuration. Set up a project for one recurring job, put each need in instructions, knowledge or a skill, and keep its memory to that job. Read 8.3, then do Lab 8, Task 13.
  • Connector and Knowledge Management. Choose between an upload and a connector for each source, and keep a project's knowledge current, relevant and free of noise. Read 8.3, then do Lab 8, Task 12.
  • Persistent-Instruction Design. Write the standing instructions that keep answers consistent, and decide what goes in them and what goes in each request. Read 8.3, then do Lab 8, Task 6.
  • Configuration Lifecycle and Degradation. When a setup that worked gives worse answers, find the layer that changed, such as instructions, knowledge, a skill or memory, and fix it. Read 11.2, then do Lab 11, Task 12.

Domain 6: Responsible Use, Data Protection, and Governance (17%)

  • Use-Case Appropriateness. Rate a proposed use as appropriate, appropriate with safeguards, or inappropriate, weighing third-party skills and connectors and Anthropic's Usage Policy, and write down why. Read 10.1, then do Lab 10, Task 2.
  • Data Protection in Practice. Handle sensitive data as policy says: classify it, de-identify it before upload, including for code analysis, and use memory and incognito controls to match. Read 10.2, then do Lab 10, Task 3.
  • Policy, Ethics, and Diligence. Check how people use Claude against policy, find gaps, lapses and bias, and tell a policy gap from a training gap to pick the fix. Read 10.5, then do Lab 10, Task 11.

Practice in Claude, as the guide asks

The guide asks you to use Claude in your own work, and to practice the Associate's tasks.2 This is where the book has you do each one in Claude.

The guide asks you toWhere you do it
Configure a project with instructions, knowledge sources and a skill, for a recurring taskCase 4's lab, Part 2. Lab 8, Task 13 builds a project from the records alone
Connect a workplace data sourceBuild Something: Data, "Going further", on a paid plan
Review, edit and delete memoriesLab 8, Task 3: you sort a project's six memory topics, then edit and delete lines in your own saved memory, and ask a chat with memory off what it knows. Lab 8, Task 13 reads a project's own memory
Diagnose why a prompt produced a generic or off-target outputLab 5, Task 7
Verify the claims and citations in a research output before you use themLab 6, Task 12, and on your own work, Lab 6, Task 13
Choose among projects, skills, research and code-executed analysis for a taskLab 2, Task 3
Decide which steps of a workflow Claude may do, and which a person must keepLab 9, Task 2
Prepare a sensitive dataset for analysis, in line with your organization's policyLab 10, Task 3: you cut a raw vendor export in a spreadsheet, notes included, before anything goes up, then ask Claude what still looks personal

Check that you are ready

Do each chapter's final round, on its Check yourself page. It shows your score in each exam domain, as the CCAO-F score report does.2 If a domain scores low, go back to its skills in the map. When you are ready for the clock, sit the PCAO-F sample exam.

The exam's words and the book's words

Some ideas have one name in the exam and another in the book. Four rows, marked, come from Anthropic's prep course.7 On the exam, think in the exam's words.

The exam's word or ideaThe book saysWhere the book teaches it
PromptBrief5.1
Role, context, task, output formatOutcome, context, format and constraints, with a role line where it helps5.1
Project instructions, persistent instructionsStanding instructions8.3
Scoped memoryA project's memory, kept to that one body of work8.2
Appropriate with safeguardsAppropriate with review10.1
AnonymizationDe-identified, pseudonymized or redacted data10.2
Needs human override (prep course)Requires human correction6.2
Under-specification (prep course)The brief: it never included what the work needed11.2
Stale configuration (prep course)A setup that drifted, such as a source out of date11.2
Least privilege (prep course)Grant what the task needs7.2
Approved knowledge the worker answers fromKSoR, the governed record of what is officially true4.2
The job that a project's instructions describeRole Contract2.6
Human oversight, and what the worker may doAuthority Envelope, the limits of what the worker may doChapter 7

Versions of this page

  • October 2026: written for guide 1.0, with Panaversity's projection of the next guide.
  • November 2026: rewritten for guide 1.1, with a route to the exam that starts in Claude and ChatGPT 2.0. Later that month, rebuilt in the guide's own order and names, with how the exam thinks. Then given Panaversity's two PCAO-F 1.1 exams, and made shorter.
  1. PCAO-F 1.1 exams, Panaversity's announcement to students, 21 November 2026.↑↑2
  2. Claude Certified Associate: Foundations Exam Guide, version 1.1, Anthropic, effective December 7, 2026.↑↑2↑3↑4↑5↑6↑7
  3. Panaversity's editor-in-chief, in a message to the book's editors, 21 November 2026.↑
Owner
Panaversity