Submitting the final quiz unlocks your grade, Leaderboard points, and Certificate.
◆ Only one answer is 100%. Others may count 75% or 50% — your final grade reflects this. At the end you'll see the correct answers for all questions.
⚠️ Submit once and be patient — your grade can take up to 30 minutes.
After you submit, our pipeline processes your result automatically. This can take up to 30 minutes — that's normal. If you submit a second time thinking something went wrong, the pipeline will detect the duplicate and only your first submission counts . Your grade, Leaderboard points, and Certificate will not be processed for the repeat. Submit once, then wait.
After up to 30 minutes, find your grade at members.builderscamp.com and your Certificate at members.builderscamp.com/certificates .
Good luck. 🤞
1. What is the single biggest determinant of AI agent performance?
A) The quality of individual prompts — how specific, detailed, and well-structured each request is. B) The Claude model version in use — newer models consistently outperform older ones on engineering tasks. C) The quality of context management — what the agent knows, how fresh it is, and how it's organized. D) The number of MCP integrations and external tools connected to the agent.
2. What is the most important constraint on what belongs in CLAUDE.md?
A) It should contain only rules that are always true, project-wide, and persistent across sessions — never situational rules. B) It should be updated at the end of every session with new learnings from that conversation. C) It should be kept under a strict line count so it doesn't consume too much of the context window. D) It should serve as a comprehensive reference for all project conventions, including those that only apply in certain contexts.
3. What is 'memory rot' and what is the recommended solution?
A) Memory rot is when the agent begins hallucinating; the solution is providing more specific, step-by-step instructions. B) Memory rot is when context windows fill up during complex tasks; the solution is working on smaller, more isolated features. C) Memory rot is when the model's base training becomes outdated; the solution is switching to a newer model version regularly. D) Memory rot is when long sessions accumulate outdated decisions, abandoned approaches, and contradictory instructions; the solution is starting fresh sessions with a short rehydration briefing and clear source documents.
4. What best describes a 'Skill' when working with AI coding agents?
A) An executable onboarding document packaging workflow steps, known pitfalls, quality gates, and a definition of done for recurring work. B) A shorthand alias for a frequently typed prompt — reduces keystrokes but requires re-explanation when context changes. C) A persistent memory entry that stores the outcome of the most recent agent session for future reference. D) A model capability flag that unlocks advanced reasoning modes for complex engineering tasks.
5. Why use specialized agents over a single general-purpose agent?
A) Specialized agents keep each role focused — the builder isn't distracted by review concerns, and the reviewer isn't biased by having written the code. B) Specialized agents are easier to configure because each one requires fewer permissions than a general-purpose agent. C) Specialized agents reduce memory rot and context pollution by isolating concerns — each agent has a focused context, and the human makes the final decisions. D) Specialized agents can run autonomously end-to-end, eliminating the need for human review checkpoints.
6. What makes Hooks fundamentally different from rules written in CLAUDE.md?
A) CLAUDE.md is maintained by developers; Hooks are generated automatically by the agent based on observed patterns. B) CLAUDE.md applies to the whole project; Hooks apply only to specific files or directories. C) CLAUDE.md is the agent's primary instruction file; Hooks extend it by connecting to external tools and APIs. D) CLAUDE.md contains recommendations the agent should follow; Hooks are automated controls that enforce rules regardless of what the agent outputs.
7. How should a team decide which MCPs and plugins to enable — and when?
A) Enable all available MCPs at project start to give the agent maximum capability from day one. B) Treat MCPs as expensive resources — enable only what's needed, prefer read-only access, and activate write access deliberately. C) Add MCPs progressively as features require them, keeping all previously added integrations permanently enabled. D) Disable MCPs after each session to avoid context buildup, then re-enable them at the start of the next session.
8. In what order should the stages of AI-assisted engineering work be structured?
A) Research → Plan → Build → Ship → Done — each stage has distinct goals, validation criteria, and outputs. B) Explore → Prototype → Validate → Iterate — prioritizing learning cycles over linear execution. C) Design → Implement → Test → Deploy — following standard software development lifecycle stages. D) Prompt → Generate → Accept → Ship — move fast by eliminating unnecessary review stages.
9. How should 'Done' be defined when working with AI coding agents?
A) Done means the AI expresses high confidence in its output and the code runs without immediate errors. B) Done means a specialized reviewer agent has approved the implementation without major findings. C) Done is an objective state defined by pre-established acceptance criteria and passing tests — not the agent's confidence level. D) Done means the implementation closely matches the original prompt, with no major deviations from what was described.
10. What does the 'continuous improvement' principle require in this operating model?
A) Each session should gradually reduce the number of human approvals required as the agent proves its reliability. B) Each session should capture lessons, generate improvement proposals, and update skills and rules — all subject to human approval before taking effect. C) Improvement proposals from sessions should be automatically applied to skills and rules to eliminate the review bottleneck. D) After enough sessions, CLAUDE.md should be replaced by the agent's own learned context, removing the need for manually authored rules.
