A-Z glossary of agentic HR

How to use this glossary

This is a reference document, not a narrative article. Use it in three ways:

• During vendor evaluations. When a vendor uses a term, check it here. If their definition diverges from the industry-standard meaning, that is a signal worth investigating.
• In internal presentations. Copy definitions directly into decks and briefing documents. Consistent terminology accelerates alignment across HR, IT, and executive stakeholders.
• For onboarding. Share with team members who are new to the agentic HR space. Reading A through Z takes ten minutes and builds a working vocabulary.

A

Term	Definition and Context
Agent	An autonomous software entity that perceives its environment, reasons over context, and takes governed actions without step-by-step human instruction. Distinct from a chatbot (which responds to prompts) and a workflow bot (which follows a fixed script). An HR agent might detect a retention risk, assemble intervention options, and route a recommendation to an HRBP, all without being asked.
Agent Runtime	The execution environment where agents operate. Manages lifecycle, concurrency, memory, tool access, and governance enforcement. Think of it as the operating system for agents. Without a runtime, agents are just prompts with no persistence or coordination.
Agentic HR	The application of autonomous agents to HR processes, replacing portal-based, human-initiated workflows with proactive, context-aware, governed automation. Not a product name. A category shift in how HR technology delivers value.
Agentic Reasoning	The capacity of an agent to evaluate a situation, weigh multiple factors, consider constraints, and select an appropriate action. Goes beyond pattern matching. Involves multi-step inference, goal decomposition, and trade-off analysis.
Autonomous Action	An action initiated by an AI system without a direct human trigger. Governed autonomy means the system can act independently within defined boundaries. Ungoverned autonomy (no boundaries) is what responsible vendors avoid.

B

Term	Definition and Context
Boundary Problem	The limitation that arises when an agent or AI system can only access data within a single application. A Workday agent that cannot see data in your ATS, LMS, or collaboration tools operates within a boundary. Cross-system context solves this.
Business Process Framework (BPF)	A predefined structure of process steps, conditions, and routing rules in an HCM system. BPFs are deterministic and rigid. Agents reason dynamically. The tension between BPFs and agentic reasoning is a key architectural question in modernization.

C

Term	Definition and Context
Chain of Thought	A reasoning pattern where an AI system breaks a problem into sequential steps and works through them explicitly. In agentic HR, chain-of-thought reasoning makes agent decisions auditable because you can inspect each step in the logic.
Chatbot	A conversational interface that responds to user queries, typically with retrieval-based or generative answers. A chatbot answers questions. An agent takes action. The distinction matters because many vendors label chatbots as agents.
Context Engine	The inference layer that assembles cross-system data into a unified understanding of people, roles, skills, and organizational dynamics in real time. Not a database. Not a search index. A reasoning substrate that connects disparate signals into actionable context.
Context Window	The amount of information an LLM can process in a single interaction. Measured in tokens. A larger context window does not equal better reasoning. What matters is which information fills the window and how it is selected.
Copilot	A UI-embedded AI assistant that helps users complete tasks within an application. Operates in a single-app context. Requires the user to be in the application. Contrast with a flow-of-work agent that operates across applications and meets people where they are.
Cross-System Context	The ability to assemble and reason over data from multiple systems simultaneously. A retention analysis that combines compensation (HCM), engagement (survey tool), performance (talent suite), and external signals (market data) requires cross-system context. Single-vendor copilots typically lack this.

D

Term	Definition and Context
Decision Orchestration	The coordination of multiple data sources, analytical models, business rules, and approval workflows to arrive at a recommended or automated decision. Agents perform decision orchestration. Dashboards display data and leave orchestration to humans.
Deep Reasoning	Multi-step analytical processing where an agent evaluates trade-offs, considers constraints, and models outcomes before recommending action. Distinct from shallow pattern matching or simple retrieval.

E

Term	Definition and Context
Embedding	A numerical vector representation of text, skills, roles, or other entities that captures semantic meaning. Embeddings allow systems to compute similarity (e.g., how close is “project management” to “program coordination”). The foundation of modern skills matching and talent intelligence.
Explainability	The ability of an AI system to articulate why it made a specific recommendation or took a specific action. In HR, explainability is not optional. Employees, managers, and regulators all require it. “The model said so” is not an explanation.

F

Term	Definition and Context
Flow of Work	The tools and environments where employees spend their day: Teams, Slack, email, calendars. Not the HR portal. Delivering HR intelligence in the flow of work means the employee never leaves their primary workspace.
Foundation Model	A large-scale AI model (e.g., GPT-4, Claude) pre-trained on broad data and adapted for specific tasks. Foundation models provide general reasoning. Domain-specific context and governance layers make them useful for HR.

G

Term	Definition and Context
Governed Autonomy	The operating model where agents can act independently within defined guardrails. Business rules, approval thresholds, and escalation paths constrain what agents can do. The agent is autonomous, but not unsupervised.
Guardrails	Constraints placed on AI systems to prevent harmful, biased, or unauthorized actions. In agentic HR, guardrails include approval workflows, compensation band limits, data access controls, and escalation triggers. Guardrails are what make autonomy safe.

H

Term	Definition and Context
Hallucination	When an AI system generates information that is plausible-sounding but factually incorrect. In HR, hallucination risk is highest when agents lack structured data and rely solely on generative models. Grounding agent responses in verified data from systems of record is the primary mitigation.
Human-in-the-Loop (HITL)	A design pattern where a human reviews and approves agent recommendations before they are executed. Critical for high-stakes HR decisions (compensation changes, termination flags, promotion recommendations). The loop can be tight (approve every action) or loose (approve exceptions only).

I

Term	Definition and Context
Intelligence Layer	The technology stack that sits between systems of record and the employee experience. Includes the context engine, knowledge graph, agent runtime, and governance framework. The layer that transforms stored data into proactive intelligence.
Intent Detection	The ability of an agent to understand what a user is trying to accomplish from natural language input or behavioral signals. “I want to grow” might mean career development, compensation increase, or team expansion depending on context.

K

Term	Definition and Context
Knowledge Graph	A structured representation of entities (people, skills, roles, projects) and the relationships between them. Unlike a relational database, a knowledge graph makes relationships first-class citizens. “Who has Python skills AND has led cross-functional projects AND is within two levels of a VP role?” is a graph query, not a SQL query.

L

Term	Definition and Context
LLM (Large Language Model)	A neural network trained on large text datasets that can generate, summarize, and reason over natural language. LLMs power the conversational and reasoning capabilities of agents. They are necessary but not sufficient for agentic HR. Without structured data and governance, an LLM is just a text generator.

M

Term	Definition and Context
Memory (Agent)	The ability of an agent to retain context across interactions and over time. Session memory lasts one conversation. Persistent memory accumulates across weeks and months. An agent that remembers your career goals from three months ago is fundamentally more useful than one that starts fresh every time.
Multi-Agent System	An architecture where multiple specialized agents collaborate to solve complex problems. A workforce planning scenario might involve a demand-forecasting agent, a skills-gap agent, and a sourcing-strategy agent working together. Coordination between agents is the hard problem.

O

Term	Definition and Context
Ontology (Skills)	A formal, structured representation of skills and their relationships (parent/child, adjacent, prerequisite). A skills ontology defines what “data science” means and how it relates to “machine learning,” “statistics,” and “Python.” Taxonomic ontologies are static hierarchies. Dynamic ontologies evolve with market data.
Orchestration	The coordination of multiple agents, tools, data sources, and workflows to complete a complex task. The orchestration layer decides which agent handles which part of a request, manages handoffs, and ensures the overall outcome is coherent.

P

Term	Definition and Context
Persistent Context	Information that an agent retains across sessions and uses to improve over time. An agent with persistent context knows your team composition, open roles, recent decisions, and stated priorities without being told each time. Without persistence, every interaction starts from zero.
Portal Architecture	A design model where users must navigate to a dedicated application to access functionality. The dominant paradigm in HR tech for 20+ years. The opposite of flow-of-work delivery. The reason most HR tools see 20-40% adoption for non-mandatory tasks.
Proactive Intelligence	The ability of a system to surface insights and recommendations without being asked. A proactive agent notifies a manager about a succession gap before it becomes a crisis. A reactive system waits for the manager to run a report.

R

Term	Definition and Context
RAG (Retrieval-Augmented Generation)	A pattern where an LLM retrieves relevant documents or data before generating a response. Reduces hallucination by grounding responses in actual data. In HR, RAG might pull policy documents, employee records, or market benchmarks before answering a question.
Reasoning Engine	The component of an agent that evaluates options, weighs trade-offs, and determines the best course of action. Not the same as a rules engine (which follows if/then logic). A reasoning engine can handle ambiguity, incomplete information, and novel situations.

S

Term	Definition and Context
Session-Based AI	An AI interaction that begins and ends within a single conversation, with no memory carried forward. Most chatbots and copilots operate this way. Adequate for transactional queries (“How many PTO days do I have?”). Inadequate for sustained processes (career development, succession planning).
Skills Inference	The process of deducing skills a person likely has based on their experience, role, projects, and other signals, even if those skills are not explicitly listed in a profile. Inference dramatically expands the usable skills data in an organization beyond self-reported profiles.
System of Action	Software that reads from systems of record, reasons across the data, and takes governed action. Requires cross-system context, autonomous reasoning, and action-taking capabilities. The layer that transforms stored data into workforce decisions.
System of Record	Software that stores, manages, and enforces transactions on employee data. Your HCM (Workday, SuccessFactors, Oracle) is the canonical example. Essential for compliance and data integrity. Not designed for cross-system reasoning or proactive action.

T

Term	Definition and Context
Talent Intelligence	The practice of using data and AI to understand workforce capabilities, gaps, trends, and opportunities. Encompasses skills mapping, labor market analytics, succession modeling, and retention prediction. Talent intelligence is the input. Agents are the action layer.
Tool Use (Agent)	The ability of an agent to invoke external tools, APIs, or systems to gather information or execute actions. An agent that can query your ATS, update a requisition, and send a Slack message is using tools. Tool use is what separates agents from chatbots that only generate text.

W

Term	Definition and Context
Workforce Context Engine (WCE)	Gloat’s specific implementation of a context engine. Assembles real-time understanding of people, roles, skills, and organizational dynamics by integrating data from HCM, ATS, LMS, performance, and collaboration systems. The foundation that agents reason over.
Workflow Automation	The execution of predefined process steps triggered by events or conditions. Traditional workflow automation follows fixed paths. Agentic workflows adapt dynamically based on context. The distinction: a workflow does what it was told. An agent figures out what needs to be done.