Agentic AI glossary.
Plain-English definitions of the terms that come up when you evaluate enterprise Agentic AI: agents, orchestration, governance, and the IBM watsonx stack. Each entry is a quick, citable answer with how we apply it in production.
All terms
Agent Catalog (watsonx Orchestrate)
The place in IBM watsonx Orchestrate to find, publish, and reuse agents and tools: prebuilt from IBM, from partners, or built by your own team.
Read definition →ArchitectureAgent control plane
The operating layer to deploy, monitor, secure, and govern an entire fleet of AI agents from one place, managing the agentic estate at scale.
Read definition →ArchitectureAgent2Agent (A2A) protocol
An open protocol that lets AI agents from different frameworks or vendors discover one another and collaborate through a common interface.
Read definition →Agentic AIAgentic AI
AI systems that take action by applying logic, making decisions, and running multi-step workflows end to end across enterprise systems, with human oversight.
Read definition →ArchitectureAgentOps
The practice of running AI agents in production: tracing, evaluation, cost and latency monitoring, and incident response for agent behavior.
Read definition →Agentic AIAI agent
A software component that pursues a goal by perceiving context, reasoning about what to do, and acting through tools and APIs, then iterating on the result.
Read definition →GovernanceAI governance
The policies, controls, and processes that keep AI systems safe, compliant, transparent, and accountable across their lifecycle.
Read definition →AI-assisted developmentIBM Bob
IBM's AI coding agent for enterprise teams: understand, build, test, and modernize code in the IDE or terminal, on IBM SaaS or self-hosted.
Read definition →AI-assisted developmentBobcoins (IBM Bob)
The usage units for IBM Bob. Every plan includes an allowance, and Enterprise pools Bobcoins across a team with limits per team or user.
Read definition →ArchitectureFoundation model
A large AI model trained on broad data that can be adapted to many tasks through prompting or fine-tuning. Granite is IBM's enterprise family.
Read definition →IBM watsonxIBM Granite
IBM's family of open foundation models for enterprise work, released under Apache 2.0 and available on watsonx or your own infrastructure.
Read definition →GovernanceHuman-in-the-loop (HITL)
A design pattern where a person reviews, approves, or can override an AI system's decisions at defined, high-stakes points.
Read definition →ArchitectureModel Context Protocol (MCP)
An open standard that connects AI agents and applications to tools, data, and systems through one common interface instead of one-off integrations.
Read definition →ArchitectureMulti-agent orchestration
Coordinating several specialized agents, with handoffs and shared context, to complete a process that no single agent owns end to end.
Read definition →ArchitectureRetrieval-augmented generation (RAG)
A technique that grounds an AI model's responses in trusted external data retrieved at query time, improving accuracy and reducing hallucination.
Read definition →IBM watsonxIBM watsonx BI
IBM's AI agent for business insights: ask questions about your data in plain language, with answers grounded in a governed semantic layer.
Read definition →IBM watsonxIBM watsonx Orchestrate
IBM's platform for building, deploying, and orchestrating AI agents that automate work across enterprise applications, with governance.
Read definition →IBM watsonxIBM watsonx.ai
IBM's enterprise studio to build, train, tune, and deploy AI models, including the Granite foundation models, under enterprise controls.
Read definition →IBM watsonxIBM watsonx.data
IBM's open, hybrid data lakehouse to query data across cloud, on-prem, and hybrid on open formats, feeding trusted data to AI, agents, and analytics.
Read definition →GovernanceIBM watsonx.governance
IBM's toolkit to govern AI across its lifecycle: model risk, lineage, drift and bias monitoring, and policy controls with audit trails.
Read definition →