Validates that you can build, integrate, and ship production-grade applications, agents, and workflows: Claude API, Agent SDK, Claude Code, MCP servers, evals, security, and cost optimization.
Who it's for: AI/ML engineers, technical leads, and software engineers with hands-on Claude or LLM experience (Python/TypeScript, REST APIs).
Practice bank in production — be first in line
We're building the full Developer question bank against the official blueprint below. Leave your email and get notified at launch (plus early free access).
Domains and weights from the official Anthropic exam guide (v1.0). Your score report shows percent-correct per domain.
Agent architecture and tradeoffs, Claude Agent SDK, custom agent loops, subagents, and agentic frameworks.
Claude API mechanics (messages, tools, streaming, caching, batch), application design across interfaces, CLAUDE.md and configuration management.
Rules, Skills, Commands, Agents, session management, headless mode, and settings.json configuration.
Error type identification, recovery strategies, and trace analysis to isolate failure modes.
LLM fundamentals, model tiers and tradeoffs, token budgeting, prompt caching, and cost modeling.
Context window management, drift prevention, prompt engineering methods, and output validation patterns.
Prompt injection mitigation, guardrail layering, hooks for safety controls, and secrets management.
Tool implementation and function calling, tool description writing, and MCP server development and deployment.
Same style and cognitive level as the real exam. Click an answer to see the rationale.
Your application sends a 6,000-token system prompt with static policy content on every request, followed by a short user message. Costs are growing linearly with traffic. Which change most directly reduces cost without losing context?
An agent loop should stop when Claude has finished its task. Which termination condition is the correct primary mechanism?
A coding agent has tools to read files, write files, and execute shell commands. Untrusted repository content may contain prompt-injection payloads. Which defense most directly limits the blast radius of a successful injection?
Meanwhile, the free Claude lessons cover a large part of this blueprint.