CCAO-F: Associate – Foundations Exam

Format

Questions: 60 (multiple choice and multi-response).
Length: 120 minutes.
Fee: $99 (the cheapest Anthropic exam; partner discounts in certifications-overview).
Passing: 720 on the 100–1,000 scaled score.
Delivery: Pearson VUE (OnVUE or test center).

Who it's for

Professionals who use Claude as a productivity tool — analysts, operators, managers, consultants — not API developers. This boundary is itself tested: a hallmark CCAO-F answer pattern is knowing when to escalate technical work to Developers/Architects rather than attempt it in the chat surface. If a scenario needs custom code, an API integration, or an MCP server built, the "correct" associate move is to scope and hand off.

The seven domains

#	Domain	Weight
1	Prompting and Task Execution	14%
2	Output Evaluation and Validation	21%
3	Product and Model Selection	12%
4	Workflow Integration and Solution Design	16%
5	Configuration and Knowledge Management	12%
6	Governance, Risk, and Responsible Use	15%
7	Troubleshooting and Optimization	10%

Domain 2 is the heavyweight — evaluation beats prompting (21% vs 14%). The exam cares more that you can catch bad output than write clever prompts.

Tested concepts by domain

Prompting (D1) — prompt structuring (context, task, format, constraints), iteration (refining via the Description–Discernment loop from ai-fluency-4d), and decomposition — splitting a big request into sequenced smaller ones.
Evaluation (D2) — hallucination identification (spotting confident fabrication), bias identification, fact-checking workflows (verify against sources, not against the model itself), and output format selection — choosing among artifacts (standalone, iterable deliverables), inline responses (quick answers), and structured output (tables, lists for downstream use).
Product/model selection (D3) — Haiku vs Sonnet vs Opus on the cost/speed/quality triangle (model-family), plus choosing the right claude.ai feature: Projects, research mode (multi-source investigation), artifacts.
Workflow integration (D4) — designing repeatable Claude-in-the-loop processes; knowing context limitations and the three responses to hitting them: restart (fresh conversation), summarize (compress and continue), persist (move durable knowledge into a Project or document).
Configuration (D5) — Projects configuration (instructions + knowledge files), connectors such as Google Drive and Gmail for grounding Claude in live organizational data.
Governance (D6) — data sensitivity classification (what may be pasted into Claude), privacy obligations, and operating within your organization's AI policy; transparency about AI involvement (Diligence, in 4D terms).
Troubleshooting (D7) — diagnosing underperforming prompts: is the failure missing context, ambiguous instructions, wrong model, exhausted context window, or a task the model fundamentally can't do?

How questions read

Scenario-based, like the architect exams but at the workflow level:

"A marketing manager pastes a 90-page PDF and asks for a summary; Claude's answers degrade after several follow-ups. What should they do?"

Distractors are usually plausible but wrong-layer — e.g., "switch to Opus" when the real issue is context exhaustion, or "write an API script" when the associate-correct answer is escalate or reconfigure the Project.

Preparation pointers

claude-cowork and claude-ai-features — the product surface the exam lives on.
ai-fluency-4d — the framework behind D1/D2/D4/D6.
ai-capabilities-limitations — the mental model behind D2 and D7.
model-family — cold-know the Haiku/Sonnet/Opus tradeoffs for D3.
Hands-on: build one Project with knowledge files + a connector, and deliberately break a long conversation to practice restart/summarize/persist.