Skip to content

Study notes: Associate – Foundations

My working notes from preparing for this exam, organized by blueprint domain. They condense what matters; they do not replace the exam guide, and nothing here comes from the live exam, which is covered by a non-disclosure agreement.

How to use these notes: read the study guide first for the facts and blueprint, then work through this page domain by domain, checking that every line reads as obvious to you. Anything that does not is a study target.

Domain 2 first: output evaluation (21%)

The largest domain, so it earns the first slot. The mindset the exam rewards is professional skepticism applied cheaply:

  • Treat every specific-looking detail in an output (citation numbers, statistics, names, dates) as unverified until checked against an authoritative source. Confident tone and self-reported confidence are not accuracy signals.
  • Know the common failure shapes: fabricated specifics, plausible-but-wrong summaries, silent omissions, and bias inherited from framing.
  • The verification workflow worth internalizing: identify claims that would matter if wrong, check those against the source of record, and route anything regulated, legal, or high-stakes to human review.
  • Adapting output for an audience (tone, format, length, artifact versus inline) is scored material, not an afterthought.
flowchart LR
    O[Claude output] --> C{Claims that matter if wrong?}
    C -->|yes| V[Verify against the source of record]
    C -->|no| A[Adapt for audience and format]
    V -->|holds up| A
    V -->|does not| R[Rework the prompt or escalate]

Prompting and task execution (14%)

  • Structure beats cleverness: state the role, the task, the context, the constraints, and the output format. Decompose big requests into steps rather than one sprawling prompt.
  • Iteration is the method: inspect what is wrong with an output, change one thing, rerun. Different task types (analysis, research, drafting, brainstorming) justify different strategies, and the exam expects you to match them.

Product and model selection (12%)

The tradeoff table to know cold:

Model Character Fits
Haiku Fastest, lowest cost High-volume, straightforward tasks
Sonnet Balanced Most day-to-day work
Opus Most capable, highest cost and latency Complex reasoning that justifies the spend
  • Feature selection is tested the same way: Projects for recurring context, research mode for sourced answers, Artifacts for documents and structured deliverables, plain chat for quick tasks.
  • Context is finite. Know when to start a fresh conversation, when to summarize and carry forward, and when to persist knowledge in a Project instead of re-pasting it.

Workflow integration and solution design (16%)

  • The valued skill is translating a business problem into a Claude workflow: analyze the requirement, pick where Claude adds value, integrate it into the existing process rather than around it, and be able to state its value and limits to a stakeholder plainly.
  • Redesign beats bolt-on: the exam guide's own language is process reimagination, not just task automation.

Configuration and knowledge management (12%)

  • Projects: instructions define behavior, knowledge sources define what it knows. Keep instructions short, specific, and maintained; stale knowledge is a quality bug you own.
  • Connectors (Google Drive, Gmail) extend reach; managing what is connected is part of governance, not just convenience.

Governance, risk, and responsible use (15%)

  • The pattern the sample questions reward: make the task safe, then do it. Anonymize or redact regulated identifiers before upload rather than abandoning the task or trusting an instruction to the model to compensate.
  • Know your organization's data classes and which may never leave approved systems. "Instruct Claude not to retain it" is not a control.
  • Appropriate-use judgment is tested: some tasks are wrong for AI in your context regardless of capability.

Troubleshooting and optimization (10%)

  • Diagnose before rewriting: is the failure missing context, ambiguous instruction, wrong feature, or wrong model? Fix the identified cause, not everything at once.
  • Optimization means cheaper and faster at equal quality: reusable prompts, Projects for repeated context, and the smallest model that holds quality.

Recall list

Facts worth having instantly available on exam day: 60 questions, 120 minutes, 720 of 1,000 to pass, closed book, results on screen immediately, badge via Credly, credential valid 12 months with a free renewal assessment, retake waits of 14, 30, then 90 days, and the Associate exam does not count toward partner tier eligibility.

These notes are the maintainer's own summary and carry no official standing. Study guide · Repository index