Claude Certified Developer — Foundations glossary

23 terms the exam expects you to know, defined in plain language.

A

API documentation as ground truth(ground truth docs, official docs): The official API documentation serves as the authoritative source for resolving disagreements about API behavior.
Actionable Error Messages: Error messages that contain sufficient context to guide recovery actions.
Attention problems: Issues related to model focus that are often confused with context window limitations.

Coordinator(coordinator): An agent that orchestrates and manages the execution of subagents.

Decomposition problems: Issues related to breaking down tasks that are often confused with context window limitations.
Deterministic mechanisms(deterministic vs probabilistic, deterministic): Software engineering constructs such as hooks, preconditions, schemas, and tool constraints that guarantee behavior.

Failure Category Conflation(failure categories): Mistaking one type of error for another, such as timeout vs empty result or syntax vs semantic errors.
Failure category(failure categories): A distinct class of system error such as timeout vs empty result, syntax vs semantic, or drift vs overflow.
Failure category distinctions: Differentiating between related failure types such as timeout vs empty result, syntax vs semantic errors, drift vs overflow.

Integrating MCP and the SDKs(integrating mcp and the sdks): The process of combining Model Context Protocol with software development kits for AI applications.

Messages API(messages api): An API for sending and receiving messages between users and an AI model.

Parallel subagents(parallel subagents): Subagents that execute simultaneously from multiple Task calls in a single turn.
Performance, Cost, and Reliability(performance, cost, and reliability): Considerations for optimizing efficiency, cost-effectiveness, and dependability of AI systems.
Probabilistic mechanisms(probabilistic mechanisms, probabilistic): Techniques such as prompts and examples that encourage but do not guarantee a specific behavior.
Programmatic enforcement(programmatic enforcement): The use of deterministic mechanisms to enforce behavior in high-stakes scenarios like financial, legal, or safety.
Prompt-level fix(prompt-level fixes): An adjustment to prompts or examples that influences model behavior without programmatic guarantees.

Right-Sized Solution Selection(right-sized solutions): The practice of choosing the simplest adequate fix over overengineered alternatives.
Right-sized solution(right-sized solutions): The principle of choosing the simplest adequate fix (e.g., description rewrite) over overengineered infrastructure.
Right-sized solutions: Choosing the simplest effective fix to avoid overengineering.

Structured Output and Prompt Techniques(structured output and prompt techniques): Methods for obtaining formatted output from AI and designing prompts for specific responses.

Task call(task calls): An invocation that creates a subagent to perform a specific task.
Tool Use and Function Calling(tool use and function calling): A capability enabling an AI model to invoke external tools or functions to perform actions.
Turn(turn): A single round of interaction in which one or more Task calls can be made.