Glossary (Agentic)

An autonomous or semi-autonomous AI system that pursues goals, makes plans, uses tools, and acts in an environment. WAF++ requires every agent to have a unique identity, authentication, and auditability. Related: WAF-AGN-010.

Known limitations or failure modes of an agent (knowledge cutoff, tool gaps, accuracy issues). The Agent Debt Register documents and prioritizes these limitations. Related: WAF-AGN-100.

A network of heterogeneous agents with specialized roles that cooperate through defined interfaces. Related: WAF-AGN-060.

Coordination of multiple agents across architecture and process boundaries, including task assignment, event communication, and failure recovery. Related: WAF-AGN-060.

Consistency of an agent’s behavior with human values, organizational rules, and regulatory requirements. Verified through guardrails, tests, and human-in-the-loop checks. Related: WAF-AGN-050.

Explicit recording of the intermediate steps a model goes through when making a decision. Important for debugging, explainability, and compliance. Related: WAF-AGN-030.

Behavior that arises from the interaction of multiple agents and was not explicitly programmed. Can be creative or risky. Related: WAF-AGN-050, WAF-AGN-060.

Capability of a model to generate structured calls to external functions or APIs. Foundation of tool use in agents. Related: WAF-AGN-020.

Safety and policy mechanisms that constrain agent outputs and actions. Include content filters, action boundaries, rate limiting, and fail-safes. Related: WAF-AGN-050.

Human review or approval before the execution of critical agent actions. Ensures accountability and compliance. Related: WAF-AGN-080.

Large Language Model. A transformer-based language model with a large parameter count that enables reasoning, planning, and text generation.

Document describing a model’s capabilities, limitations, training data, evaluation results, and usage risks. Supports transparent model selection and compliance.

A system with multiple agents that cooperate or interact in defined roles. Requires orchestration and clear communication. Related: WAF-AGN-060.

A specialized agent that controls other agents, delegates tasks, and monitors overall progress. Related: WAF-AGN-060.

Temporal and complexity scope of the planning an agent can perform. Ranges from single-step actions to long-term, multi-step plans.

Attack in which inputs are manipulated so that a model performs unintended actions. Countermeasures include input/output filtering and strict tool permissions. Related: WAF-AGN-020, WAF-AGN-050.

Complete, time-ordered record of an agent’s decision-making. Essential for audit and debugging. Related: WAF-AGN-030, WAF-AGN-070.

Many homogeneous agents working in parallel and decentralized toward a shared goal. Related: WAF-AGN-060.

An agent’s ability to invoke external tools, APIs, or code runners. Must be governed through whitelists, schemas, and logging. Related: WAF-AGN-020.