Skip to content

Industrial agentic terminology

General

  • Agentic AI: AI systems designed to act as autonomous “coworkers” that can take goals, make decisions, and coordinate across tools or sub-agents to complete multi-step business processes. Unlike passive AI (just answering queries), Agentic AI is about doing work on behalf of humans, following rules, using context, and handing off subtasks as needed.
  • Generative AI: AI models that can create new content, such as text, images, or code, by learning from large datasets and generating outputs that resemble human-created material. Generative AI is typically used for creativity, language, and content automation (e.g., drafting documents, summarizing, creating responses).
  • Predictive AI: AI models trained to forecast outcomes based on historical data. Predictive AI focuses on patterns and probabilities, for example, predicting demand, identifying churn risk, or estimating delivery times, rather than creating new content or executing autonomous workflows.
  • Digital worker (vs. “agent”): Preferred term because it signals AI that helps people rather than replaces jobs; leadership guidance is to avoid “agent” in external framing.
  • Orchestrator (agent): An agent that coordinates the overall plan and hands work off to sub-agents/tools; any agent can be an orchestrator if it has “an army of sub-agents.”
  • Sub-agent: A delegated agent used when the task involves reasoning/decision-making (e.g., email extraction). Current constraint: sub-agents cannot have sub-agents.
  • Tool: Used when input→output is deterministic with no decision-making (e.g., “create order” API). May wrap multiple API calls.
  • Deterministic vs. decision-making: Deterministic → tool; requires reasoning/choices → sub-agent.

Platform / IFS-Specific Terms

  • Projection (IFS projection): IFS concept: an entity plus its REST APIs (CRUD and actions).
  • Connector (IFS connector): Integration that imports IFS projections so tool scaffolding is auto-generated.
  • API task: The specific endpoint call inside a tool, populated from a projection’s API spec.
  • Form (tool input schema): Input fields a tool expects, auto-derived from required API parameters.
  • GNI response task: Logic that shapes a tool’s output to only the fields requested by the agent at runtime.
  • Defaults (site/user defaults): Auto-fetched values from IFS (e.g., customer/site defaults) that populate records when not present in input.
  • Transactional entity (IFS): Multi-step API flow (e.g., cancel purchase order) that requires several API calls but is often wrapped as one tool.

Process, Testing & Debugging Terms

  • Triggers: Ways an agent starts: email, cron, Slack, Teams (planned), or CX ticket events. Triggers can have instructions/prompt logic.
  • Approvals / human-in-the-loop: Points where the agent seeks human approval/notification (Slack, email, Teams).
  • Plan (plan view): Execution plan showing which sub-agents/tools were chosen and in what order.
  • Events (run events/logs): Per-task status and I/O; successful tasks show green checkmarks, exposing inputs/outputs for debugging.
  • Evaluation / eval: Method to test agents: run sample prompts, inspect plan/events, rerun as simulation tests.