Site map
Every page, in one place.
Auto-generated from the same seed data that powers the rest of the site. Add a new lesson, blog post, case study, or editable page and it shows up here on the next deploy without anyone having to edit this page.
11 sections · 17 groups · 115 links
Quick start
The fastest paths into the site.
Curriculum
All 30 lessons across 5 exam domains.
Overview
Agentic Architecture & Orchestration · 27%
- Agentic Architecture — domain landing
- Agent vs. WorkflowWhen to give Claude autonomy vs. when to pin a fixed sequence.
- Orchestrator-Worker PatternOne Claude routes subtasks to specialized subagents.
- Evaluator–Optimizer LoopUse Claude to grade Claude's own output and iterate.
- Parallel Fan-OutMap-reduce over Claude calls when the work shards cleanly.
- Agent State & MemoryPersisting decisions and intermediate results across turns.
- Human-in-the-LoopWhere to surface confirmation prompts without breaking flow.
- Evaluating Agent QualityGolden tasks, success rates, and step-level traces.
Tool Design & MCP Integration · 18%
- Tool Design & MCP — domain landing
- Tool Schema FundamentalsDesigning input schemas Claude reliably populates.
- Tool Error ContractsReturn errors that teach Claude how to recover.
- MCP OverviewWhat the Model Context Protocol is and when to use it.
- MCP Resources vs. ToolsWhen to expose data as a resource vs. a tool call.
- Testing Tools in IsolationRun tools end-to-end without invoking Claude.
Claude Code Configuration & Workflows · 20%
- Claude Code — domain landing
- Claude Code: What It IsAnthropic's official agentic CLI for software work.
- settings.json ConfigurationWhere the harness reads permissions, hooks, env, and model.
- HooksShell commands that fire on Claude Code events.
- Slash CommandsReusable named workflows the user invokes with /name.
- SkillsModular capability packs with their own prompts and resources.
- Claude Agent SDKBuilding custom agents on the same primitives as Claude Code.
Prompt Engineering & Structured Output · 20%
- Prompt Engineering — domain landing
- System PromptsWhere to put role, constraints, and output contract.
- Structured OutputJSON mode, schema-guided generation, and post-validation.
- Few-Shot ExamplesHow many examples, in what order, of what shape.
- Prompt Caching5-minute server-side cache that cuts tokens by 90% on hits.
- Extended ThinkingWhen to enable thinking budgets and what they cost.
- Iterating Prompts Like CodeVersion, test, and review prompts.
Context Management & Reliability · 15%
- Context & Reliability — domain landing
- Token BudgetsWorking within context windows without truncating critical state.
- Context CompactionSummarize old turns to keep recent detail.
- Retries & BackoffIdempotency and exponential backoff for API errors.
- ObservabilityLogging traces, costs, and quality signals.
- GuardrailsInput/output filters that catch policy and PII issues.
- Safe RolloutsFeature flags, shadow runs, and gradual ramps.
14-Day Plan
Adaptive daily schedule, mastery map, and per-day task lists.
Practice & testing
Every interactive test surface on the site.
Dashboard area
The signed-in / live-state side of the site.
Case studies
Long-form production walkthroughs, one per domain.
- All case studies (index)
- Building a customer-success copilot at ScaleOpsHow a B2B SaaS team went from one brittle chatbot to a production-grade copilot in six weeks.
- How DataLayer built its integrations platform on MCPReplacing a tangle of bespoke connectors with reusable MCP servers.
- Migrating Rails 5 → 7 at LumenPayA 400K-line monolith and a five-person team — how Claude Code took the timeline from quarters to weeks.
- Cutting LLM costs 62% at OmniSupportHow a customer-support platform retuned prompts, caching, and structured output without sacrificing quality.
- Shipping a research agent at Research CoFrom a slick demo to a system that handles thousand-document briefs without face-planting.
- PrintWize: from one prompt to a real agent in three sprintsHow a 30-person print-on-demand startup outgrew a single Claude call.
- NorthStar Analytics: cutting $40K/month in LLM costA four-week dive into prompt caching, model tiering, and compaction.
- LedgerLine: a tax-prep bot that never files without a humanHow a SMB tax startup shipped autonomous drafting with strict HITL.
- HelioGrid: a 40-skill internal Claude Code libraryEncoding tribal knowledge so juniors ramp in three weeks instead of nine.
- OrbitalDevs: from bespoke connectors to MCPA CI/CD platform replacing 12 hand-written wrappers with reusable MCP servers.
Journal
25 posts, newest first.
- Journal index
- MCP in 2026: where the ecosystem actually is2026-05-17 · Snapshot of the MCP ecosystem this month — what's stable, what's emerging, what's still a research project.
- What's actually worth memorizing for test day2026-05-16 · The short list of facts that show up verbatim on the exam, and the longer list of patterns that show up in scenarios.
- Why we built the BKT mastery map2026-05-15 · A binary 'lesson completed' flag is a lie. Mastery is probabilistic. Here's why that matters for spaced practice.
- Sequencing your final 72 hours before the exam2026-05-14 · A schedule for the three days right before you sit the real test. Light on volume, heavy on retrieval.
- Cloud progress sync just went live2026-05-13 · Sign in with Google and your lessons, BKT mastery, mock attempts, and 14-Day Plan now follow you across devices.
- Updating your study plan for the May 2026 model lineup2026-05-12 · With Opus 4.7, Sonnet 4.6, and Haiku 4.5 all current, here's what to revisit in our curriculum this week.
- Choosing between Opus 4.7, Sonnet 4.6, and Haiku 4.52026-05-10 · A decision flowchart for the current model lineup. Memorize this — at least one exam question is a near-verbatim variant.
- Haiku 4.5 for production: where it actually wins2026-05-09 · Haiku 4.5 punches above its price tier. Three patterns where swapping it in saves real money without quality loss.
- Sonnet 4.6: the workhorse most candidates should default to2026-05-08 · Sonnet 4.6 is the model the exam assumes when scenarios don't specify. Know its sweet spot and its trade-offs cold.
- Claude Opus 4.7 with 1M context: what changes for the exam2026-05-07 · Anthropic released Opus 4.7 with a 1M-token context window. The exam objectives don't change — but how you answer some scenarios will.
- Shipping your first agent to production2026-05-06 · A pragmatic checklist: evals, observability, budgets, kill switch.
- Three recipes for bulletproof structured output2026-04-30 · JSON mode alone isn't enough. Combine it with examples, validation, and a single retry.
- Skills vs. slash commands: when to use which2026-04-23 · Slash commands are for things you invoke; skills are for things you teach.
- Leave headroom: the 70% rule2026-04-16 · Plan for ~70% of the window so you have room for outputs and unexpected tool result blocks.
- Using Claude Code hooks for team-wide policy2026-04-09 · Hooks are the right place to enforce 'no force-push' and 'no skipping tests' across every developer's machine.
- Why you should build agents without tools first2026-04-02 · Before adding tools, see if a pure-LLM loop solves your task. Most teams over-engineer.
- Mock exam strategy: how to use your 60 minutes2026-03-25 · Two passes. First pass: 90 seconds per question, mark anything uncertain. Second pass: spend the rest on the marked ones.
- Prompt cache economics: why static-first ordering matters2026-03-19 · The 5-minute cache cuts input cost by ~90% on hits. Almost everyone loses cache hits by putting volatile content first.
- MCP in 15 minutes2026-03-11 · The Model Context Protocol in plain language: tools, resources, and prompts — when to use which.
- What to study first for the Claude Certified Architect exam2026-03-04 · Start with agentic patterns and tool design — they're the two biggest scoring blocks and the most transferable to real work.
- The 80/20 of the certification2026-02-26 · Twelve concepts get you to 720. Memorize these and the rest is gravy.
- Setting up your prep environment2026-02-12 · Get an Anthropic API key, a sandbox repo, and Claude Code wired up before you read anything else. You'll absorb 2-3x more.
- Reading Anthropic's docs efficiently2026-02-05 · Three reading orders that compress weeks of reading into a long weekend.
- The exam blueprint: what 'foundations' means in practice2026-01-29 · Five domains, 27/18/20/20/15 weighting. 'Foundations' doesn't mean trivia — it means production patterns most teams get wrong on day one.
- Why we built this site2026-01-15 · The official Claude Certified Architect exam launched. We needed a single-page path through 5 sprawling domains. So we built one.
Reference & audio
Outbound resources and the read-aloud surface.
About & policies
Site information, terms, and privacy.
Account
Sign-in (Google OAuth) when the deployment is configured for it.
Site
Meta routes.