Constraint Solving
Definition
Finding an action plan that satisfies limits such as budget, time, permissions, dependencies, policies, or resource availability.
Agent Planning & Reasoning terms and explanations from the Agentic AI Glossary.
Definition
Finding an action plan that satisfies limits such as budget, time, permissions, dependencies, policies, or resource availability.
Definition
A Critic Agent reviews another agent's plan, answer, or tool trajectory and points out errors, risks, or missing evidence.
Definition
A branching structure that represents choices and outcomes, helping agents or systems select actions under different conditions.
Definition
A map of tasks or components that shows which items must happen before others can safely run.
Definition
Planning that changes during execution as the agent receives new observations, tool results, failures, or user constraints.
Definition
A recorded timeline of execution activity that helps engineers debug behavior and understand execution history.
Definition
An Executor Agent carries out the steps created by a planner, usually by calling tools, APIs, or workflow actions.
Definition
Breaking a broad objective into smaller subgoals that can be planned, assigned, executed, and evaluated.
Definition
A reasoning approach that explores connected ideas or partial solutions as a graph instead of a single linear chain.
Definition
Planning in layers, where a high-level goal is divided into phases, tasks, and concrete execution steps.
Definition
Planning across many steps, tools, or time periods where memory, checkpoints, and recovery become important.
Definition
A meaningful checkpoint that shows progress toward a larger goal and helps decide whether to continue, revise, or stop.
Definition
A pattern where one component creates a plan and another component follows the plan while reporting progress and failures.
Definition
Carrying out planned steps through model calls, tools, APIs, retrieval, human approvals, or workflow actions.
Definition
Changing a plan after discovering missing information, failed assumptions, blocked tools, or better available options.
Definition
Checking that a plan is feasible, safe, complete, and aligned with the user's goal before execution.
Definition
A Planner Agent breaks a high-level goal into ordered steps, dependencies, and checkpoints before execution begins.
Definition
Using search over possible states or actions to find a path from the current situation to a desired goal.
Definition
Planning only the next few steps, useful when the environment changes quickly or uncertainty is high.
Definition
Turning a task into clear ordered steps that can be executed, checked, retried, or delegated.
Definition
A smaller unit of work inside a larger task, often assigned to a tool, worker agent, or workflow step.
Definition
Splitting a complex request into manageable subtasks with clear order, dependencies, and completion criteria.
Definition
A reasoning approach that explores multiple possible solution paths and chooses the most promising branch.
Definition
A Verifier Agent checks whether an output, tool result, or completed task satisfies the requirements and quality criteria.
Definition
A retry strategy that waits longer after repeated failures, reducing pressure on rate-limited or unstable services.
Definition
Restoring a workflow from a saved state so an agent can resume after failure without starting over.
Definition
The ability to process multiple sessions, tasks, or tool calls at the same time.
Definition
A queue that stores failed messages or tasks for later inspection, repair, replay, or manual handling.
Definition
A reduced-capability operating state used when a model, tool, data source, or service is unavailable.
Definition
The degree to which the same input and configuration produce the same result, important for testing and auditability.
Definition
Logic that detects exceptions, communicates failures clearly, retries when safe, and prevents uncontrolled agent behavior.
Definition
Code-level handling of unexpected runtime problems such as bad inputs, unavailable APIs, or invalid tool outputs.
Definition
A specific way an agent or system can fail, such as hallucination, wrong tool use, timeout, or unsafe escalation.
Definition
An alternate path used when the preferred model, tool, retrieval source, or workflow cannot complete safely.
Definition
The ability to continue operating safely when components fail, slow down, or return partial results.
Definition
Reducing functionality in a controlled way instead of failing completely when resources or services are limited.
Definition
A design property where repeating the same action does not create duplicate or harmful side effects.
Definition
Behavior where the same request may produce different outputs because of sampling, model updates, tool state, or timing.
Definition
Managing provider or API limits through throttling, queuing, backoff, caching, or routing to alternatives.
Definition
Recovery Strategy is a planned approach for recovery. It guides design, deployment, measurement, governance, and continuous improvement.
Definition
The ability of an AI system to deliver correct, safe, and available behavior across normal and edge-case conditions.
Definition
The ability to rerun an evaluation or agent trace and obtain comparable behavior for debugging and compliance.
Definition
The system's ability to absorb failures, adapt, and keep delivering acceptable service.
Definition
Repeating a failed request or tool call under controlled limits, often with backoff and idempotency protection.
Definition
The ability to handle noisy inputs, edge cases, bad retrieval, ambiguous requests, or tool errors without breaking.
Definition
Investigating the underlying reason for a failure, such as bad data, prompt design, tool errors, or infrastructure limits.
Definition
A failure state that avoids harm, data leakage, unsafe actions, or misleading output even when the task cannot finish.
Definition
The ability to maintain quality, reliability, and cost control as usage grows.
Definition
Automatic detection, retry, repair, or rerouting after an execution failure.
Definition
Securely managing the lifecycle and context of individual user or task sessions.
Definition
Restoring task variables, memory, progress, and context after interruption or system failure.
Definition
A maximum wait time for a model, tool, or workflow step before the system stops, retries, or falls back.
Definition
Agent Graph is a workflow pattern for agent graph. It coordinates agent steps, tools, state, approvals, handoffs, or recovery paths.
Definition
Agentic Workflow is a workflow pattern for agentic workflow. It coordinates agent steps, tools, state, approvals, handoffs, or recovery paths.
Definition
Checkpointing is a workflow pattern for checkpointing. It coordinates agent steps, tools, state, approvals, handoffs, or recovery paths.
Definition
Critic-Reviewer is a workflow pattern for critic-reviewer. It coordinates agent steps, tools, state, approvals, handoffs, or recovery paths.
Definition
DAG is a workflow pattern for dag. It coordinates agent steps, tools, state, approvals, handoffs, or recovery paths.
Definition
Directed Acyclic Graph is a workflow pattern for directed acyclic graph. It coordinates agent steps, tools, state, approvals, handoffs, or recovery paths.
Definition
Durable Execution is a workflow pattern for durable execution. It coordinates agent steps, tools, state, approvals, handoffs, or recovery paths.
Definition
Escalation Flow is a workflow pattern for escalation flow. It coordinates agent steps, tools, state, approvals, handoffs, or recovery paths.
Definition
Evaluator-Optimizer is a workflow pattern for evaluator-optimizer. It coordinates agent steps, tools, state, approvals, handoffs, or recovery paths.
Definition
A design where systems react to events, triggers, messages, or state changes rather than only direct user requests.
Definition
Event-Driven Workflow is a workflow pattern for event-driven workflow. It coordinates agent steps, tools, state, approvals, handoffs, or recovery paths.
Definition
Human Approval Flow is a workflow pattern for human approval flow. It coordinates agent steps, tools, state, approvals, handoffs, or recovery paths.
Definition
Multi-Step Workflow is a workflow pattern for multi-step workflow. It coordinates agent steps, tools, state, approvals, handoffs, or recovery paths.
Definition
Coordinating models, tools, agents, state, policies, and workflows into one reliable execution path.
Definition
Orchestrator-Worker is a workflow pattern for orchestrator-worker. It coordinates agent steps, tools, state, approvals, handoffs, or recovery paths.
Definition
Parallelization is a workflow pattern for parallelization. It coordinates agent steps, tools, state, approvals, handoffs, or recovery paths.
Definition
Planner-Executor is a workflow pattern for planner-executor. It coordinates agent steps, tools, state, approvals, handoffs, or recovery paths.
Definition
RAG Workflow is a workflow pattern for rag workflow. It coordinates agent steps, tools, state, approvals, handoffs, or recovery paths.
Definition
Routing is a workflow pattern for routing. It coordinates agent steps, tools, state, approvals, handoffs, or recovery paths.
Definition
State Machine is a workflow pattern for state machine. It coordinates agent steps, tools, state, approvals, handoffs, or recovery paths.
Definition
Supervisor-Worker is a workflow pattern for supervisor-worker. It coordinates agent steps, tools, state, approvals, handoffs, or recovery paths.
Definition
Tool-Use Flow is a workflow pattern for tool-use flow. It coordinates agent steps, tools, state, approvals, handoffs, or recovery paths.
Definition
Workflow is a workflow pattern for workflow. It coordinates agent steps, tools, state, approvals, handoffs, or recovery paths.
Definition
Automating a multi-step business or technical process through tools, rules, and agent decisions.
Definition
Workflow Orchestration is a workflow pattern for workflow orchestration. It coordinates agent steps, tools, state, approvals, handoffs, or recovery paths.
Definition
A best-first search method that combines path cost so far with a heuristic estimate of remaining cost.
Definition
A property of a search algorithm or heuristic that guarantees an optimal solution when its assumptions hold.
Definition
A search strategy that expands the most promising node first according to an evaluation function.
Definition
A search strategy that works forward from the start and backward from the goal until the two searches meet.
Definition
The average number of child nodes generated from each node in a search space.
Definition
A search strategy that explores all nodes at the current depth before moving to the next level.
Definition
A search approach that explores possible states without using problem-specific guidance.
Definition
A collection of nodes that a search algorithm has already expanded.
Definition
A maximum search depth used to prevent depth-first search from exploring endlessly.
Definition
A search strategy that follows one path as deeply as possible before backtracking.
Definition
The length of the shortest sequence of operators from an initial state to a goal state.
Definition
A connection in a problem-space graph that represents an operator or transition between states.
Definition
A first-in, first-out data structure commonly used to implement breadth-first search.
Definition
The desired state that a search algorithm is trying to reach.
Definition
A search strategy that expands the node estimated to be closest to the goal.
Definition
A function that estimates how close a state is to a goal or how costly the remaining path may be.
Definition
A search approach that uses domain knowledge to choose more promising paths first.
Definition
A search strategy that uses heuristic information about the problem to improve efficiency.
Definition
The starting state from which a search problem begins.
Definition
A search method that repeatedly runs depth-first search with increasing depth limits until it finds a solution.
Definition
A last-in, first-out data structure commonly used to implement depth-first search.
Definition
A local search method that keeps several candidate states and repeatedly selects the best successors.
Definition
A state that is better than nearby states but may not be the best solution overall.
Definition
A search method that improves a candidate solution by moving to neighboring solutions.
Definition
A point in a search graph that represents a state or partial solution.
Definition
A scoring function that measures how good a candidate solution is.
Definition
A collection of generated nodes that a search algorithm has not expanded yet.
Definition
An action or transition rule that changes one problem state into another.
Definition
The accumulated cost of moving through a sequence of states or actions.
Definition
A specific search problem defined by an initial state and a goal state.
Definition
The set of states and operators in which a search problem is solved.
Definition
A graph representation of a search problem where nodes are states and edges are operators.
Definition
A search method that expands nodes according to heuristic values and manages open and closed lists.
Definition
A method for exploring states and actions to find a solution to a problem.
Definition
A local search method that sometimes accepts worse moves early so it can escape poor local solutions.
Definition
A problem where one agent must find a sequence of moves from a start state to a goal state.
Definition
The maximum amount of memory an algorithm needs while solving a problem.
Definition
A new state that results from applying an operator to a current state.
Definition
The amount of computation or number of generated nodes an algorithm may require.
Definition
An optimization problem that seeks the lowest-cost tour visiting each city once and returning to the start.
Definition
A search strategy that expands the path with the lowest accumulated cost first.
Explore more chapters or test your knowledge with quizzes.