Skip to content
AI BI Academy
Architect · Foundations · L200

Claude Certified Architect – Foundations

The design-level exam. This is where you leave 'does the code work' behind and start defending choices about orchestration, isolation, reliability, and cost. Scenario-based: four scenarios drawn from a pool of six.

Exam at a glance$125
Questions
60 (4 × 15)
Duration
120 minutes
Passing score
720 / 1000
Validity
12 months
Format
Scenario — 4 of 6
Delivery
Online-proctored or Pearson VUE
Counts toward CPN tier
Yes
Who it's for
Solution architects

Your diagrams end up on VP whiteboards. You choose orchestration patterns and defend them.

Tech leads & staff engineers

You decide fork-vs-subagent, hook-vs-prompt, and when the Batch API pays for itself.

Delivery leads

You own reliability and cost across a whole Claude solution, not a single endpoint.

The syllabus

Domain breakdown & weights

Click any domain to expand its objectives

The heart of the exam — how work is decomposed, isolated, and coordinated.

  • Context isolation: fork session vs spawn subagent
  • The subagent trap — pass findings explicitly
  • Delegation via the Task tool
  • Single-agent vs multi-agent tradeoffs

The scenario bank

Four scenarios, drawn from a pool of six.

Each scenario carries 15 questions. Prepare all six and you can afford to lose your weakest one. Knowing them up front tells you exactly which cross-domain skills to drill.

01
Customer Support Resolution Agent
Orchestration · MCP · Reliability
02
Code Generation with Claude Code
Claude Code · Reliability
03
Multi-Agent Research System
Orchestration · MCP · Reliability
04
Developer Productivity with Claude
MCP · Claude Code · Orchestration
05
Claude Code for Continuous Integration
Claude Code · Prompt Engineering
06
Structured Data Extraction
Prompt Engineering · Reliability
Exam-day cheat sheet

The distinctions it quietly leans on

Memorise which side of each split a question is testing — it eliminates wrong answers fast.

Enforcement
Programmatic (hooks, gates) — for money/safety consequences
Prompt-based — for style and format only
Errors
Access failure (timeout) → retry
Empty result → no action
Config scope
~/.claude/CLAUDE.md — personal, not shared
project/.claude/CLAUDE.md — team, version-controlled
Session
Resume — when prior context is still valid
Fresh start + injected summary — when results are stale
Accuracy
Aggregate (97%) — can hide failure modes
Segmented by doc-type × field — the honest metric
Prep track

Free courses, in the right order.

Every course below is free on the Anthropic Academy. Work them top to bottom — tick each off to track your readiness.

Your progress0 / 8

Tap a course to mark it complete and watch your readiness climb.

What actually shows up

Traps & high-value focus areas

The subagent trap is tested verbatim

Subagents never inherit the coordinator's context. 'Analyze the findings' fails — you must pass the findings text explicitly. Expect this exact distinction.

Improve tool descriptions first

Any option that reaches for a routing classifier or tool consolidation before improving tool descriptions is almost always wrong.

Never bet money on a prompt

If a failure has financial, security, or safety consequences, the answer is a programmatic hook or gate — not a prompt instruction.

Aggregate accuracy is a red flag

Any option that reports only aggregate accuracy for a decision about reducing human review should be eliminated.

Ready to earn Architect?

Follow the mapped prep track, drill the mock exam, and book when your readiness says go.

Explore other credentials